Chapter 7

Revealed
Preference
 Revealed Preference Analysis
 Suppose we observe the demands
(consumption choices) that a
consumer makes for different
budgets. This reveals information
about the consumer’s preferences.
We can use this information to ...
Revealed Preference Analysis
– Test the behavioral hypothesis that
a consumer chooses the most
preferred bundle from those
available.
– Discover the consumer’s
preference relation.
 Assumptions on Preferences
 Preferences
– do not change while the choice data
are gathered.
– are strictly convex.
– are monotonic.
 Together, convexity and monotonicity
imply that the most preferred
affordable bundle is unique.
 Assumptions on Preferences
x2
If preferences are convex and
monotonic (i.e. well-behaved)
then the most preferred
x2* affordable bundle is unique.

x1* x1
 Direct Preference Revelation
 Suppose that the bundle x* is chosen
when the bundle y is affordable.
Then x* is revealed directly as
preferred to y (otherwise y would
have been chosen).
 Direct Preference Revelation
x2
The chosen bundle x* is
revealed directly as preferred
to the bundles y and z.
x*
z
y
x1
 Direct Preference Revelation
 That x is revealed directly as
preferred to y will be written as

x y.
D
 Indirect Preference Revelation
 Suppose x is revealed directly
preferred to y, and y is revealed
directly preferred to z. Then, by
transitivity, x is revealed indirectly as
preferred to z. Write this as

x z
I
  
so x y and y z x I
z.
D D
 Indirect Preference Revelation
x2
z is not affordable when x* is chosen.

x*

x1
 Indirect Preference Revelation
x2
x* is not affordable when y* is chosen.

x*
y*
z

x1
 Indirect Preference Revelation
x2
z is not affordable when x* is chosen.
x* is not affordable when y* is chosen.

x*
y*
z

x1
 Indirect Preference Revelation
x2
z is not affordable when x* is chosen.
x* is not affordable when y* is chosen.
So x* and z cannot be compared
x* directly.
y*
z

x1
 Indirect Preference Revelation
x2
z is not affordable when x* is chosen.
x* is not affordable when y* is chosen.
So x* and z cannot be compared
x* directly. 
y* But x*x* y*
D
z

x1
 Indirect Preference Revelation
x2
z is not affordable when x* is chosen.
x* is not affordable when y* is chosen.
So x* and z cannot be compared
x* directly. 
y* But x*x* y*
D
z and y*

z
D
x1
 Indirect Preference Revelation
x2
z is not affordable when x* is chosen.
x* is not affordable when y* is chosen.
So x* and z cannot be compared
x* directly. 
y* But x*x* y*
D
z and y*

z
D 
so x* z.
x1 I
 Two Axioms of Revealed
Preference
 To apply revealed preference
analysis, choices must satisfy two
criteria -- the Weak and the Strong
Axioms of Revealed Preference.
 The Weak Axiom of Revealed
Preference (WARP)
 If the bundle x is revealed directly as
preferred to the bundle y then it is
never the case that y is revealed
directly as preferred to x; i.e.
 
x y not (y x).
D D
 The Weak Axiom of Revealed
Preference (WARP)
 Choice data which violate the WARP
are inconsistent with economic
rationality.
 The WARP is a necessary condition for
applying economic rationality to
explain observed choices.
 The Weak Axiom of Revealed
Preference (WARP)
 What choice data violate the WARP?
 The Weak Axiom of Revealed
Preference (WARP)
x2

y
x
x1
 The Weak Axiom of Revealed
Preference (WARP)
x2
x is chosen when y is available

so x y.
D

y
x
x1
 The Weak Axiom of Revealed
Preference (WARP)
x2
x is chosen when y is available

so x y.
D
y is chosen when x is available

so y x.
D
y
x
x1
 The Weak Axiom of Revealed
Preference (WARP)
x2
x is chosen when y is available

so x y.
D
y is chosen when x is available

so y x.
D These statements are
y
x inconsistent with
each other.
x1
Checking if Data Violate the WARP
 A consumer makes the following
choices:
– At prices (p1,p2)=($2,$2) the choice
was (x1,x2) = (10,1).
– At (p1,p2)=($2,$1) the choice was
(x1,x2) = (5,5).
– At (p1,p2)=($1,$2) the choice was
(x1,x2) = (5,4).
 Is the WARP violated by these data?
Checking if Data Violate the WARP
Choices
(10, 1) (5, 5) (5, 4)
Prices
($2, $2) $22 $20 $18

($2, $1) $21 $15 $14

($1, $2) $12 $15 $13

 Checking if Data Violate the WARP
Choices
(10, 1) (5, 5) (5, 4)
Prices
($2, $2) $22 $20 $18

($2, $1) $21 $15 $14

($1, $2) $12 $15 $13

Red numbers are costs of chosen bundles.

 Checking if Data Violate the WARP
Choices
(10, 1) (5, 5) (5, 4)
Prices
($2, $2) $22 $20 $18

($2, $1) $21 $15 $14

($1, $2) $12 $15 $13

Circles surround affordable bundles that

were not chosen.
 Checking if Data Violate the WARP
Choices
(10, 1) (5, 5) (5, 4)
Prices
($2, $2) $22 $20 $18

($2, $1) $21 $15 $14

($1, $2) $12 $15 $13

Circles surround affordable bundles that

were not chosen.
 Checking if Data Violate the WARP
Choices
(10, 1) (5, 5) (5, 4)
Prices
($2, $2) $22 $20 $18

($2, $1) $21 $15 $14

($1, $2) $12 $15 $13

Circles surround affordable bundles that

were not chosen.
 Checking if Data Violate the WARP

Ch o i c e s
(10,1) (5,5) (5,4) (10, 1) (5, 5) (5, 4)
Prices
($2,$2) $22 $20 $18 (10,1) D D

($2,$1) $21 $15 $14 (5, 5) D

($1,$2) $12 $15 $13 (5 ,4) D

 Checking if Data Violate the WARP

Ch o i c e s
(10,1) (5,5) (5,4) (10, 1) (5, 5) (5, 4)
Prices
($2,$2) $22 $20 $18 (10,1) D D

($2,$1) $21 $15 $14 (5, 5) D

($1,$2) $12 $15 $13 (5 ,4) D

Checking if Data Violate the WARP

(10,1) is directly (10, 1) (5, 5) (5, 4)

revealed preferred
to (5,4), but (5,4) is (10,1) D D
directly revealed
preferred to (10,1), (5, 5) D
so the WARP is
violated by the data. (5 ,4) D
Checking if Data Violate the WARP
x2

(5,4) (10,1)
D


(10,1) (5,4)
D

x1
 The Strong Axiom of Revealed
Preference (SARP)
 If the bundle x is revealed (directly or
indirectly) as preferred to the bundle
y and x  y, then it is never the case
that the y is revealed (directly or
indirectly) as preferred to x; i.e.
 
x y or x y
D I
 
not ( y x or y x ).
D I
 The Strong Axiom of Revealed
Preference
 What choice data would satisfy the
WARP but violate the SARP?
 The Strong Axiom of Revealed
Preference
 Consider the following data:

A: (p1,p2,p3) = (1,3,10) & (x1,x2,x3) = (3,1,4)

B: (p1,p2,p3) = (4,3,6) & (x1,x2,x3) = (2,5,3)

C: (p1,p2,p3) = (1,1,5) & (x1,x2,x3) = (4,4,3)

 The Strong Axiom of Revealed
Preference
A: ($1,$3,$10)
Choice
(3,1,4). A B C
Prices

B: ($4,$3,$6) A $46 $47 $46

(2,5,3).
B $39 $41 $46

C: ($1,$1,$5) C $24 $22 $23

(4,4,3).
 The Strong Axiom of Revealed
Preference
Choice
A B C
Prices
A $46 $47 $46

B $39 $41 $46

C $24 $22 $23

 The Strong Axiom of Revealed
Preference
Choice In situation A,
A B C
Prices bundle A is
A $46 $47 $46 directly revealed
preferred to
B $39 $41 $46 bundle C;

A C.
C $24 $22 $23 D
 The Strong Axiom of Revealed
Preference
Choice In situation B,
A B C
Prices bundle B is
A $46 $47 $46 directly revealed
preferred to
B $39 $41 $46 bundle A;

B A.
C $24 $22 $23 D
 The Strong Axiom of Revealed
Preference
Choice In situation C,
A B C
Prices bundle C is
A $46 $47 $46 directly revealed
preferred to
B $39 $41 $46 bundle B;

C B.
C $24 $22 $23 D
 The Strong Axiom of Revealed
Preference
Choices
A B C A B C
Prices
A $46 $47 $46 A D

B $39 $41 $46 B D

C $24 $22 $23 C D

 The Strong Axiom of Revealed
Preference
Choices
A B C A B C
Prices
A $46 $47 $46 A D

B $39 $41 $46 B D

C $24 $22 $23 C D

The data do not violate the WARP.

 The Strong Axiom of Revealed
Preference
We have that A B C
  
A C, B A and C B A D
D D D

so, by transitivity, B D
  
A B, B C and C A. C D
I I I

The data do not violate the WARP but ...

 The Strong Axiom of Revealed
Preference
We have that A B C
  
A C, B A and C B A I D
D D D

so, by transitivity, B D I
  
A B, B C and C A. C I D
I I I

The data do not violate the WARP but ...

 The Strong Axiom of Revealed
Preference

B A is inconsistent A B C
D

with A B. A I D
I

B D I

C I D

The data do not violate the WARP but ...

 The Strong Axiom of Revealed
Preference

A C is inconsistent A B C
D

with C A. A I D
I

B D I

C I D

The data do not violate the WARP but ...

 The Strong Axiom of Revealed
Preference

C B is inconsistent A B C
D

with B C. A I D
I

B D I

C I D

The data do not violate the WARP but ...

 The Strong Axiom of Revealed
Preference
A B C
The data do not violate
the WARP but there are A I D
3 violations of the SARP.
B D I

C I D
 The Strong Axiom of Revealed
Preference
 That the observed choice data satisfy
the SARP is a condition necessary
and sufficient for there to be a well-
behaved preference relation that
“rationalizes” the data.
 So our 3 data cannot be rationalized
by a well-behaved preference
relation.
 Recovering Indifference Curves
 Suppose we have the choice data
satisfy the SARP.
 Then we can discover approximately
where are the consumer’s
indifference curves.
 How?
 Recovering Indifference Curves
 Suppose we observe:
A: (p1,p2) = ($1,$1) & (x1,x2) = (15,15)
B: (p1,p2) = ($2,$1) & (x1,x2) = (10,20)
C: (p1,p2) = ($1,$2) & (x1,x2) = (20,10)
D: (p1,p2) = ($2,$5) & (x1,x2) = (30,12)
E: (p1,p2) = ($5,$2) & (x1,x2) = (12,30).
 Where lies the indifference curve
containing the bundle A = (15,15)?
 Recovering Indifference Curves
 The table showing direct preference
revelations is:
Recovering Indifference Curves
A B C D E
A D D
B
C
D D D D
E D D D
Direct revelations only; the WARP
is not violated by the data.
 Recovering Indifference Curves
 Indirect preference revelations add
no extra information, so the table
showing both direct and indirect
preference revelations is the same as
the table showing only the direct
preference revelations:
 Recovering Indifference Curves
A B C D E
A D D
B
C
D D D D
E D D D
Both direct and indirect revelations; neither
WARP nor SARP are violated by the data.
 Recovering Indifference Curves
 Since the choices satisfy the SARP,
there is a well-behaved preference
relation that “rationalizes” the
choices.
 Recovering Indifference Curves
x2 A: (p1,p2)=(1,1); (x1,x2)=(15,15)
B: (p1,p2)=(2,1); (x1,x2)=(10,20)
E C: (p1,p2)=(1,2); (x1,x2)=(20,10)
D: (p1,p2)=(2,5); (x1,x2)=(30,12)
B
E: (p1,p2)=(5,2); (x1,x2)=(12,30).
D
A C

x1
 Recovering Indifference Curves
x2 A: (p1,p2)=(1,1); (x1,x2)=(15,15)
B: (p1,p2)=(2,1); (x1,x2)=(10,20)
E C: (p1,p2)=(1,2); (x1,x2)=(20,10)
D: (p1,p2)=(2,5); (x1,x2)=(30,12)
B
E: (p1,p2)=(5,2); (x1,x2)=(12,30).
D
A C

x1
Begin with bundles revealed
to be less preferred than bundle A.
 Recovering Indifference Curves
x2 A: (p1,p2)=(1,1); (x1,x2)=(15,15).

x1
 Recovering Indifference Curves
x2 A: (p1,p2)=(1,1); (x1,x2)=(15,15).

x1
 Recovering Indifference Curves
x2 A: (p1,p2)=(1,1); (x1,x2)=(15,15).

A is directly revealed preferred

to any bundle in

x1
 Recovering Indifference Curves
x2 A: (p1,p2)=(1,1); (x1,x2)=(15,15)
B: (p1,p2)=(2,1); (x1,x2)=(10,20).
E

D
A C

x1
 Recovering Indifference Curves
x2 A: (p1,p2)=(1,1); (x1,x2)=(15,15)
B: (p1,p2)=(2,1); (x1,x2)=(10,20).

x1
 Recovering Indifference Curves
x2
A is directly revealed preferred
to B and …

x1
 Recovering Indifference Curves
x2
B is directly revealed preferred
to all bundles in
B

x1
 Recovering Indifference Curves
x2
so, by transitivity, A is indirectly
revealed preferred to all bundles
in
B

x1
 Recovering Indifference Curves
x2
so A is now revealed preferred
to all bundles in the union.
B
A

x1
 Recovering Indifference Curves
x2 A: (p1,p2)=(1,1); (x1,x2)=(15,15)

E C: (p1,p2)=(1,2); (x1,x2)=(20,10).

D
A C

x1
 Recovering Indifference Curves
x2 A: (p1,p2)=(1,1); (x1,x2)=(15,15)

C: (p1,p2)=(1,2); (x1,x2)=(20,10).

A C

x1
 Recovering Indifference Curves
x2 A is directly revealed
preferred to C and ...

A C

x1
 Recovering Indifference Curves
x2 C is directly revealed preferred
to all bundles in

x1
 Recovering Indifference Curves
x2 so, by transitivity, A is
indirectly revealed preferred
to all bundles in

x1
 Recovering Indifference Curves
x2 so A is now revealed preferred
to all bundles in the union.

B
A
C

x1
 Recovering Indifference Curves
x2 so A is now revealed preferred
to all bundles in the union.
Therefore the indifference
curve containing A must lie
B
everywhere else above
A this shaded set.
C

x1
 Recovering Indifference Curves
 Now, what about the bundles
revealed as more preferred than A?
 Recovering Indifference Curves
x2 A: (p1,p2)=(1,1); (x1,x2)=(15,15)
B: (p1,p2)=(2,1); (x1,x2)=(10,20)
E C: (p1,p2)=(1,2); (x1,x2)=(20,10)
D: (p1,p2)=(2,5); (x1,x2)=(30,12)
B
A E: (p1,p2)=(5,2); (x1,x2)=(12,30).
D
A C

x1
 Recovering Indifference Curves
x2 A: (p1,p2)=(1,1); (x1,x2)=(15,15)

D: (p1,p2)=(2,5); (x1,x2)=(30,12).
A
D

x1
 Recovering Indifference Curves
x2 D is directly revealed preferred
to A.

A
D

x1
 Recovering Indifference Curves
x2 D is directly revealed preferred
to A.
Well-behaved preferences are
convex
A
D

x1
 Recovering Indifference Curves
x2 D is directly revealed preferred
to A.
Well-behaved preferences are
convex so all bundles on the
line between A and D are
A
preferred to A also.
D

x1
 Recovering Indifference Curves
x2 D is directly revealed preferred
to A.
Well-behaved preferences are
convex so all bundles on the
line between A and D are
A
preferred to A also.
D
As well, ...

x1
 Recovering Indifference Curves
x2 all bundles containing the
same amount of commodity 2
and more of commodity 1 than
D are preferred to D and
therefore are preferred to A
A also.
D

x1
 Recovering Indifference Curves
x2

bundles revealed to be
strictly preferred to A

A
D

x1
 Recovering Indifference Curves
x2 A: (p1,p2)=(1,1); (x1,x2)=(15,15)
B: (p1,p2)=(2,1); (x1,x2)=(10,20)
E C: (p1,p2)=(1,2); (x1,x2)=(20,10)
D: (p1,p2)=(2,5); (x1,x2)=(30,12)
B
A E: (p1,p2)=(5,2); (x1,x2)=(12,30).
D
A C

x1
 Recovering Indifference Curves
x2 A: (p1,p2)=(1,1); (x1,x2)=(15,15)

E: (p1,p2)=(5,2); (x1,x2)=(12,30).

x1
 Recovering Indifference Curves
x2
E is directly revealed preferred
to A.
E

x1
 Recovering Indifference Curves
x2
E is directly revealed preferred
to A.
E Well-behaved preferences are
convex

x1
 Recovering Indifference Curves
x2
E is directly revealed preferred
to A.
E Well-behaved preferences are
convex so all bundles on the
line between A and E are
preferred to A also.
A

x1
 Recovering Indifference Curves
x2
E is directly revealed preferred
to A.
E Well-behaved preferences are
convex so all bundles on the
line between A and E are
preferred to A also.
A
As well, ...

x1
 Recovering Indifference Curves
x2 all bundles containing the
same amount of commodity 1
and more of commodity 2 than
E E are preferred to E and
therefore are preferred to A
also.
A

x1
 Recovering Indifference Curves
x2
More bundles revealed
E to be strictly preferred
to A

x1
 Recovering Indifference Curves
x2
Bundles revealed
E earlier as preferred
to A
B

A C
D

x1
 Recovering Indifference Curves
x2

All bundles revealed

E to be preferred to A
B
A
C
D

x1
 Recovering Indifference Curves
 Now we have upper and lower
bounds on where the indifference
curve containing bundle A may lie.
 Recovering Indifference Curves
x2

All bundles revealed

to be preferred to A

x1
All bundles revealed to be less preferred to A
 Recovering Indifference Curves
x2

All bundles revealed

to be preferred to A

x1
All bundles revealed to be less preferred to A
 Recovering Indifference Curves
x2
The region in which the
indifference curve containing
bundle A must lie.

x1
 Index Numbers
 Over time, many prices change. Are
consumers better or worse off
“overall” as a consequence?
 Index numbers give approximate
answers to such questions.
 Index Numbers
 Two basic types of indices
– price indices, and
– quantity indices
 Each index compares expenditures
in a base period and in a current
period by taking the ratio of
expenditures.
 Quantity Index Numbers
 A quantity index is a price-weighted
average of quantities demanded; i.e.
t t
p1x1  p 2x 2
Iq 
b b
p1x1  p 2x 2
 (p1,p2) can be base period prices (p1b,p2b)
or current period prices (p1t,p2t).
 Quantity Index Numbers
 If (p1,p2) = (p1b,p2b) then we have the
Laspeyres quantity index;
b t b t
p1 x1  p 2 x 2
Lq 
b b b b
p1 x1  p 2 x 2
 Quantity Index Numbers
 If (p1,p2) = (p1t,p2t) then we have the
Paasche quantity index;
t t t t
p1x1  p 2x 2
Pq 
t b t b
p1x1  p 2x 2
 Quantity Index Numbers
 How can quantity indices be used to
make statements about changes in
welfare?
 Quantity Index Numbers
b t b t
p1 x1  p 2 x 2
 If Lq  1 then
b b b b
p1 x1  p 2 x 2
b t b t b b b b
p1 x1  p 2 x 2  p1 x1  p 2 x 2

so consumers overall were better off

in the base period than they are now
in the current period.
 Quantity Index Numbers
t t t t
p1x1  p 2x 2
 If Pq  1 then
p1t xb
1  p t b
2x 2
t t t t t b t b
p1x1  p 2x 2  p1x1  p 2x 2

so consumers overall are better off

in the current period than in the base
period.
 Price Index Numbers
 A price index is a quantity-weighted
average of prices; i.e.
t t
p1x1  p 2x 2
Ip 
b b
p1 x1  p 2 x 2
 (x1,x2) can be the base period bundle
(x1b,x2b) or else the current period
bundle (x1t,x2t).
 Price Index Numbers
 If (x1,x2) = (x1b,x2b) then we have the
Laspeyres price index;
t b t b
p1x1  p 2x 2
Lp 
b b b b
p1 x1  p 2 x 2
 Price Index Numbers
 If (x1,x2) = (x1t,x2t) then we have the
Paasche price index;
t t t t
p1x1  p 2x 2
Pp 
b t b t
p1 x1  p 2 x 2
 Price Index Numbers
 How can price indices be used to
make statements about changes in
welfare?
 Define the expenditure ratio
t t t t
p1x1  p 2x 2
M
b b b b
p1 x1  p 2 x 2
 Price Index Numbers
 If
p1t xb
1  p t b
2x 2
t t t t
p1x1  p 2x 2
Lp   M
b b b b b b b b
p1 x1  p 2 x 2 p1 x1  p 2 x 2
then

t b t b t t t t
p1x1  poverall
so consumers  better
2x 2 are  pin2the
p1x1off x2
current period.
 Price Index Numbers
 But, if
t t t t t t t t
p1x1  p 2x 2 p1x1  p 2x 2
Pp   M
b t b t b b b b
p1 x1  p 2 x 2 p1 x1  p 2 x 2
then

b t b t b b b b
p1 x1  p 2overall
so consumers
x 2 were x1  poff
p1better 2 xin
2 the
base period.
 Full Indexation?
 Changes in price indices are sometimes
used to adjust wage rates or transfer
payments. This is called “indexation”.
 “Full indexation” occurs when the
wages or payments are increased at the
same rate as the price index being used
to measure the aggregate inflation rate.
 Full Indexation?
 Since prices do not all increase at
the same rate, relative prices change
along with the “general price level”.
 A common proposal is to index fully
Social Security payments, with the
intention of preserving for the elderly
the “purchasing power” of these
payments.
 Full Indexation?
 The usual price index proposed for
indexation is the Paasche quantity
index (the Consumers’ Price Index).
 What will be the consequence?
 Full Indexation?
t t t t
p1x1  p 2x 2
Pq 
t b t b
p1x1  p 2x 2

Notice that this index uses current

period prices to weight both base and
current period consumptions.
 Full Indexation?
x2
Base period budget constraint
Base period choice

x2b

x1b x1
 Full Indexation?
x2
Base period budget constraint
Base period choice

x2b
Current period budget
constraint before indexation

x1b x1
 Full Indexation?
x2
Base period budget constraint
Base period choice
Current period budget
x2b constraint after full indexation

x1b x1
 Full Indexation?
x2
Base period budget constraint
Base period choice
Current period budget
x2b constraint after indexation

Current period
choice
after indexation
x1b x1
 Full Indexation?
x2
Base period budget constraint
Base period choice
Current period budget
x2b constraint after indexation

x t Current period
2
choice
after indexation
x1b x1t x1
 Full Indexation?
x2
(x1t,x2t) is revealed preferred to
(x1b,x2b) so full indexation makes
the recipient strictly better off if
relative prices change between
x2b the base and current periods.
x2t

x1b x1t x1
 Full Indexation?
 So how large is this “bias” in the US
CPI?
 A table of recent estimates of the
bias is given in the Journal of
Economic Perspectives, Volume 10,
No. 4, p. 160 (1996). Some of this list
of point and interval estimates are as
follows:
 Full Indexation?
Author Point Est. Int. Est.
Adv. Commission to 1.0% 0.7 - 2.0%
Study the CPI (1995)
Congressional 0.2 - 0.8%
Budget Office (1995)
Alan Greenspan 0.5 - 1.5%
(1995)
Shapiro & Wilcox 1.0% 0.6 - 1.5%
(1996)
 Full Indexation?
 So suppose a social security recipient
gained by 1% per year for 20 years.
 Q: How large would the bias have
become at the end of the period?
 Full Indexation?
 So suppose a social security recipient
gained by 1% per year for 20 years.
 Q: How large would the bias have
become at the end of the period?
20 20
 A: ( 1  0  01)  1  01  1  22 so after
20 years social security payments
would be about 22% “too large”.