Você está na página 1de 12

Consumption Theory in Terms of Revealed Preference

Paul A. Samuelson
Economica, New Series, Vol. 15, No. 60. (Nov., 1948), pp. 243-253.
Stable URL:
http://links.jstor.org/sici?sici=0013-0427%28194811%292%3A15%3A60%3C243%3ACTITOR%3E2.0.CO%3B2-0
Economica is currently published by The London School of Economics and Political Science.

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/about/terms.html. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained
prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in
the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/journals/lonschool.html.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

The JSTOR Archive is a trusted digital repository providing for long-term preservation and access to leading academic
journals and scholarly literature from around the world. The Archive is supported by libraries, scholarly societies, publishers,
and foundations. It is an initiative of JSTOR, a not-for-profit organization with a mission to help the scholarly community take
advantage of advances in technology. For more information regarding JSTOR, please contact support@jstor.org.

http://www.jstor.org
Fri Aug 17 13:39:23 2007

Consumption Theory

in Terms of Revealed Preference

DECADE ago I suggested that the economic theory of consumer's


behaviour can be largely built up on the notion of " revealed.preference ".
By comparing the costs of different combinations of goods at different
relative price situations, we can infer whether a given batch of goods
is preferred to another batch ; the individual guinea-pig, by his market
behaviour, reveals his preference pattern-if there is such a consistent
pattern.
Recently, Mr. Ian M. D. Little of Oxford University has made an
important contribution to this field,l In addition to showing the
changes in viewpoint that this theory may lead to, he has presented
an ingenious proof that if enough judiciously selected price-quantity
situations are available for two goods, we may define a locus which
is the precise equivalent of the conventional indifference curve.
I should like, briefly, to present an alternative demonstration of
this same result. While the proof is a direct one, it requires a little
more mathematical reasoning than does his.

If we confine ourselves to the case of two commodities, x and y,


we could conceptually observe for any individual a number of pricequantity situations. Since only relative prices are assbmed to matter,
each observation consists of the triplet of numbers, (p,/p,, x, y). By
manipulating prices and income, we could cause the individual to
come into equilibrium at any (x, y) point, at least within a given area.
We may also make the simplifying assumption tha-t one and only one
price ratio can be associated with each combination of x and y.
Theoretically, therefore, we could for any point (x, y) determine a
unique p,/p, ; or
(1)
pzlp,=f (2, Y)
wheref is an observable function, assumed to be continuous and with
continuous partial derivative^.^
1 I. M. D. Little : " A Reformulation of the Theory of Consumers' Behaviour", Oxford
Economic Papers, Neur Series, No. I , January, 1949 ; P. A. Samuelson : Foundatioizs of
Economic A n a l ~ ~ s (1947),
is
Ch. V and VI ; P. A. Samuelson : " A Note on the Pure Theory of
Consumer's Behaviour; and an Addendum ", Economics (1938), Vol. V (New Series), pp. 61-71,
353-354.
2 Mathematically, the above continuity assumptions are over-strict.
Also, ure shall make
the unnecessarily strong assumption that in the region under discussion the price-quantity
relations have the " simple concavity " property : J(afiay) - (af/ax)> O.

The central notion underlying the theory of revealed preference,


and indeed the whole modern economic theory of index numbers,
is very simple. Through any observed equilibrium point, A, draw the
budget-equation straight line with arithmetical slope given by the
observed price ratio. Then all combinations of goods on or within
the budget line could have been bought in preference to what was

actually bought. But they weren't. Hence, they are all " revealed "
to be inferior to A. No other line of reasoning is needed.
As yet we have no right to speak of " indifference ", and certainly
no right to speak of " indifference slopes
But nobody can object
to our summarising our observable information graphically by drawing
a little negative " slope element " a t each x and y point, with numerical
gradient equal to the price ratio in question.

".

194.81

C O K S U h Z P T I O N T H E O R Y I N TERMS O F R F V F A L E D PREFERENCE

245

This is shown in Figure I by the numerous little arrows. These


little slopes are all that we choose to draw in of the budget lines which
go through each point and the directional arrows are only drawn in
to guide the eye. I t is a well known observation of Gestalt psychology
that the eye tends to discern smooth contour lines from such a representation, although strictly speaking, only a finite number of little
line segments are depicted, and they do not .for the most part run into
each other.1 (In the present illustration the contour lines have been
taken to be the familiar rectangular hyperbola: or unitary-elasticity
curves and f (x, y) takes the simple form p,/p,:y/x.)
There is an exact mathematical counterpart of this phenomenon of
Gestalt psychology. Let us identify a little slope, dyldx, with each
price ratio, - p,/p,. Then, from (I), we have the simplest differential
equation

It is known mathematically that this defines a unique curve through


any given point, and a (one-parameter) family of curves throughout
the surrounding ( x , y) plane. These solution curves (or " integral
solutions " as they are often called) are such that when any one of
them is substituted into the above differential equation, it will be
found to satisfy that equation. Later we shall verify that these solution
curves are the conventional " indifference curves " of modern economic
theory. Also, and this is the novel part of the present paper, I shall
show that these solution curves are in fact the limiting loci of revealed
preference-or in Mr. Little's terminology they are the " behaviour
curves " defined for specified initial points. This is our excuse for
arbitrarily associating the differential equation system (2) with our
observable pattern of prices and quantities summarised in (I).
Mathematicians are able to establish rigorously the existence of
solutions to the differential equations without having to rely upon
the mind's eye as a primitive " differential-analyser " or " integrator ".2
Also, mathematicians have devised rigorous methods for numerical
solution of such equations to any desired (and recognisable) degree of
accuracy.
I t so happens that one of the simplest methods for proving the
existence of, and numerically approximating, a solution is that called
the " Cauchy-Lipschitz " method after the men who first made it
Every student of elementary physics has dusted iron filings on a piece of paper suspended
on a permanent magnet. The little filings become magnetised and orient themselves in a simple
pattern. To the mind's eye these appear as "lines of force" of the magnetic field.
a The usual proof found in such intermediate texts as F. R. Moulton, Differential Equations,
But the earlier
Ch. X I I - X I I I , is that of Picard's "method of successive approximations
rigor& proofs are by the Cauchy-Lipschitz method, which is very closely related to the economic
theory of index numbers and revealed preference. See also, R. G. D. Allen, Ma~hematicalAnalysis
for Economists, 1938, Ch. XVI.

".

246

ECONOMICA

[NOVEMBER

rigorous, even though it really goes back to at least the time of Euler.
In this method we approximate to our true solution .curve by a
connected series of straight line-segments, each line having the slope
dictated by the differential equation for the beginning point of the
straight line-segment in question. This means that our differential
equation is not perfectly satisfied at all other points ; but if we make
our line-segments numerous and short enough, the resulting error from
the true solution can be made as small as we please.
Figure 2 illustrates the Cauchy-Lipschitz approximations to the
true solution passing through the point A (10,30) and going from
x= 10, to the vertical line x = 15. The top smooth curve is the true
unitary-elasticity curve that we hope to approximate. The three
lower broken-line curves are successive approximations, improving
in accuracy as we move to higher curves.
Our crudest Cauchy-Lipschitz approximation is to use one linesegment for the whole interval. We pass a straight line through A
with a slope equal to the little arrow at A, or equal to - 3. This is
nothing but the familiar budget line through the initial point A ; it
intersects the vertical line x = 15, at the value y = 15 or at the point
marked Z'.'
(Actually, from the economic theory of index numbers and consumer's choice, we know that this first crude approximation Z' : (x, y)
= (15, 15) clearly revealed itself to be " worse " than (x, y) = (10,30)
-since the former was actually chosen over the latter even though
both cost the same amount. This suggests that the Cauchy-Lipschitz
process will always approach the true solution curve, or " indifference
curve ",from below. This is in fact a general truth, as we are about
to see.) Can we not get a better approximation to the correct solution
than this crude straight line, AZ' ? Yes, if we use two line-segments
instead of one. As b e f ~ r elet us first proceed on a straight line through
A with slope equal to A's little arrow. But let us travel on this line
only two-fifths as far as before : to x = 12, rather than x= 15. This
gives us a new point B' (12, 24), whose directional arrow is seen to
have the slope of - 2. Now, through B' we travel on a new straight
line with this new slope; and our second, better, approximation to
the true value at x = 15, is given by the new intersection, Z", with
the vertical line, at the level y = 18. (The " true " value is obviously
at Z on the smooth curve where y must equal 20 if we are to be on
the hyperbola with the property xy= 10 x 30= 15 x 20; and our
second approximation has only $ the error of our first.)
The general procedure of the Cauchy-Lipschitz process is now clear.
Suppose we divide the interval between x = 10 and x = 15 into 5 equal
segments ; suppose we follow each straight line with slope equal to
its initial arrow until we reach the end of the interval, and then begin a
new straight line. Then as our numerical table shows, we get the still
A Numerical Appendix gives the exact arithmetic underlying this and the following
figure.

better approximation, y= 199. In Figure 2, the broken line from A


to Z"' shows our third approximation..
In the limit as we take enough sub-intervals so that the size of each
line-segment becomes indefinitely small, we approach the true value
of y= 20, and the same is true for the true value at any other x point.
How do we know this ? Because the pure mathematician assures us
that this can be rigorously proved.

In economic terms, the individual is definitely going downhill along


any one Cauchy-lips chit^, curve. For just as A was revealed to be
better than Z', so also was it revealed to be better than B'. Note too
that Z" is on the budget line of B' and is hence revealed to be inferior

248

ECONOMICA

[NOVEMBER

to B', which already has been revealed to be worse than A. I t follows


that Z" is worse than A.
By the same reasoning 2"' on the third approximation curve is
shown to be inferior to A, although it now takes four intermediate
points to make this certain. I t follows as a general rule : any CauchyLipschitz path always leads to a final point worse than the initial.
And strictly speaking, it is only as an infinite limit that we can hope
to reveal the neutral case of " indifference" along the true solution
curve to the differential equation.

We have really proved only one thing so far : all points below the
true mathematical solution passing through an initial point, A, are
definitely " revealed to be worse " than A.
We have not rigorously proved that points falling on the solution
contour curve are really " equal " to A. Indeed in terms of the strict
algebra of " revealed preference " we have as yet no definition of
what is meant by " equality " or " indifference ".
Still it would be a great step forward if we could definitely prove
the following: all points above the true mathematical solution are
definitely " revealed to be better " than A.
The next following section gives a direct proof of this fact by defining
a new process which is similar to the Cauchy-Lipschitz process and
which definitely approximates to the true integral solution f r o m above.
But it may be as well to digress in this section and show that by indirect
reasoning like that of Mr. Little, we may establish the proposition
that all points above the solution-contour are clearly better than A.
I shall only sketch the reasoning. Suppose we take any point just
vertically above the point Z and regard it as our new initial point.
The mathematician assures us that a new " higher " solution-contour
goes through such a point. Let us construct a Cauchy-Lipschitz process
leftward, or backwards. Then by using small enough line-segments we
may approach indefinitely close to that point vertically above A which lzes
on the new contour line above A's contour. A will then have to lie below
the leftward-moving Cauchy-Lipschitz curve, and is thus revealed to be
worse than any new initial point lying above the old contour line. Q.E.D.
We may follow Mr. Little's terminology and give the name " behaviour
line" to the unique curve which lies between the points definitely
shown to be better than A, and those definitely shown to be worse
than A. This happens to coincide with the mathematical solution to
the differential equation, and we may care to give this contour line,
by courtesy, the title of an indifference curve.l
1 If our preference field does not have simple concavity-and
why should it ?-we may observe
cases where A is preferred to B a t some times, and B to A a t others. If this is a pattern of
consistency and not of chaos, we could choose to regard A and B as "indifferent " under those
circumstances. If the preference field has simple concavity, " indifference " will never explicitly
reveal itself to us except as the rcs~llts of an infinite limiting process.

Let me return now to the problem of defining a new approximating


process, like the conventional Cauchy-Lipschitz process, but which :
(I) approaches the mathematical solution from above rather than
below, and which (2) definitely reveals the economic preference of
the individual at every point.

Our new process will consist of broken straight lines ; and in the
limit these will become numerous enough to approach a smooth curve.
But the slopes of the straight line-segments will not be given by their
initial points, as in the Cauchy-Lipschitz process. Instead, the slope
will be determined by t h e j n a l point of the sub-interval's line-segment.

After the reader ponders over this for some time and considers its
geometrical significance, he may feel that he is being swindled. How
can we determine the slope at the line's final point, without first
determining the final point ? But, how can we know the final point
of the line unless its slope has already been determined ? Clearly, we
are a t something of a circular impasse. To determine the slope, we
seem already to require the slope.
The way out of this dilemma is perfectly straightforward to anyone
who has grasped the mathematical solution of a simultaneous equation.
The logical circle is a virtuous rather than a vicious one. By solving the
implied simultaneous equation, we cut through the problem of circular
interdependence. And in this case we do not need an electronic computer to solve the implied equation. Our human guinea-pig, simply by
following his own bent, inadvertently helps to solve our problem for us.
I n Figure 3, we again begin with the initial point A. Again we
wish to find the true solution for y at x = 15. Our first and crudest
approximation will consist of one straight line. But its slope will
be determined at the end of the interval and is initially unknown.
Let us, therefore, through A swing a straight line through all possible
angles. One and only one of these slopes will give us a line that is
exactly tangent to one of the little arrows at the end of our interval.
Let Z' be the point where our straight line is just tangent to an arrow
lying in the vertical line. It corresponds to a y value of 22&, which is
above the true value of y = 20.
Economically speaking, when we rotate a straight " budget line"
around an initial point A, and let the individual pick the best combi-pation of goods in each situation, we trace out a so-called " offer curve ".
This curve is not drawn in on the figure, but the point Z' is the intersection of the offer curve with the vertical line. I t should be obvious
from our earlier reasoning that 2' and any other point on the offer
curve is revealed to be better than A, since any such equal-cost point
is chosen over A.
So much for our crude first approximation. Let us try dividing the
interval between x = 10 and x = 15 up into two sub-intervals so that
two connected straight lines may be used. If we wish the first line
to end at x = 12, we rotate our line through A until its final slope is
just equal to the indicated little arrow (or price ratio) along the vertical
line x= 12. For the simple hyperbole in question, where - p,/p,=
dy- - y/x, our straight line will be found to end at the point B", whose

dx-

( x , y) coordinates are (12, 25$) and whose arrow has a slope of just
less than (- 2).
We now begin at B" as a new initial point and repeat the process
by finding a new straight line over the interval from x= 1 2 to x = 15.
Pivoting a line through all possible angles, we find tangency only at
the point Z", where y = 21?, which is a still better approximation to
the true value, y= 20.

The interested reader may easily verify that using more sub-intervals
and intermediate points will bring us indefinitely close to the true
solution-contour.1 I t is clear therefore that our new process brings
us to the true solution in the limit, but unlike the Cauchy-Lipschitz
process, it now approaches the solution from above. And we can use
the word " above " in more than a geometrical sense. Along the
new process lines, the individual is revealing himself to be getting
better off. For just as A is inferior to Z', it is by the same reasoning
inferior to B", which is likewise inferior to Z" ; from which it follows
that A is inferior to Z".
I t should be clear, therefore, that no matter how many intermediate
points there are in the new process, the consumer none the less reveals
himself to be travelling uphill. I t follows that every point above the
mathematical contour line can reveal itself to be better than A.

This essentially completes the present demonstration. The mathematical contour lines defined by our differential equation have been
proved to be the frontier between points revealed to be inferior to
A, and points revealed to be superior. The points lying literally on a
(concave) frontier locus can never themselves be revealed to be better
or worse than A. If we wish, then, we may speak of them as being
indifferent to A.
The whole theory of consumer's behaviour can thus be based upon
operationally meaningful foundations in terms of revealed p r e f e r e n ~ e . ~

He m a y verify that using the points x == 10,11, 1 2 , 13, 14, 15 brings us to within ot
zo, as shown in the second table of the Numerical Appendix.
The above remarks apply without qualification to two dimensional problems where the
problem of "integrability" cannot appear. In the multidimensional case there still remain
some problems, awaiting a solution for more than a decade now.

252

[NOVEMBER

ECONOMICA

NUMERICAL
APPENDIX
In the Cauchy-Lipschitz process, the straight line going from (xo, yo)
to (x,, y,) is defined by the explicit equation
(0)

Y0

y = y o - f ( ~ 0 ,~ o ) ( x - x o ) = ~ o - - ( x - x o )
X0

where dy/dx= -f (x, y) is the differential equation requiring solution


-in this case being = -y/x. The three approximations given in
Figure 2 are derived numerically in the following table.

First Approximation
initial point
30 - 3 (15 - lo)=

10

15

Second Approximation
initial point
30-3(12-10)24- 2 (15 - IZ)=

10

I2

15

30

- 30/1o=

-3

- 30/10=
- 24/12=

-3

15

30
24
I8

-2

Third Approximation

initial point

30- 3 (11 - IO)=

In the new process which approaches the true solution, y = 3oo/x,


from above, the straight lines hdve their slopes determined by the
final point of each interval, or by the implicit equation

(4 Y l = ~ o - f ( ~ l ~ ~ l ) ( ~ l - ~ o )

In the case where f (x, y)= ylx, we have

Our numerical approximations are given in the following table :

10

First Appvoximation

initial point

10

Second Approximation

initial ~ o i n t

10

Third Approximation

initial point

30

It may be mentioned that the third Cauchy-Lipschitz approximation


satisfies the equation 2 7 0 1 ~which is less than the true solution, ~ O O / X;
and the third approximation of the new upper process satisfies the
equation 330/x, which happens to be equally in excess of the true
solution.

[In Figure 3, the point between A and Z" should be labelled B"].

Você também pode gostar