• No results found

Economic Commission for Latin America, United Nations.

2 This body of literature always uses the term "real" with the meaning "deflated". cf. Ch. I, Section B, §11.

The basic principles are quite similar to those underlying the

comparison of GNP over several years for a country. Because of inflation

and changes in the price structure, in order to compare real changes in

output or in expenditures, one must devise a way to determine real prices

or, more explicitly, one must value each item of total output with the

same yardstick. When inter-country comparisons are concerned, the prob­

lems are even more acute because the price structure can be very different.

The Gilbert and Kravis approach consists of breaking down total

GNP for two different countries into equivalent groups of products. Then

within each group a certain number of items are selected. These items

must be identical in the two countries, hence all sorts of difficulties

related to quality differences arise. A distinction is drawn between

economic (cost) differences that are measurable and non-economic differ­

ences (related to know-how or skill) that are not measurable.

The next step consists of gathering price or quantity data for

these identical items. Here another question arises; should the factor

costs or the market prices be chosen? It all depends upon the major

concern: whether it is to obtain the relative real cost of production

or the relative real worth for the purchaser. Factor costs will corres­

pond to a relative productivity comparison, while market prices will

correspond to a relative expenditure comparison. The OEEC studies of

Gilbert and Kravis use the first valuation. Once the relative prices are

gathered they are used in order to revalue the expenditures in each group

of products in equivalent prices and then to aggregate them in comparable

sums.

Obviously the two countries P and R produce or consume differ­

ent relative quantities and Qr and face different relative prices P^

and P . Hence the so-called index number problem is present.

We will obtain two values for the internal purchasing power

used as weights, i.e. EP EP IPPQr EPrQr (Paasche) (Laspeyres)

and two values for the corresponding quantity comparisons according to

which country’s relative prices are used as weights, i.e.

EQPPP E(fpP

ZQPPr EQrPr

(Paasche) (Laspeyres)

These two indices are often geometrically averaged to obtain the Fisher

index.

Gilbert and Kravis affirm that the differences between the

Paasche and the Laspeyres are not a consequence of a difference in the

pattern of taste but are due to the fact that the countries compared

experience different levels of development. They assume that as countries

reach similar levels of real income, their price structure will become

similar. This is indeed a very bold but necessary statement.1

/

(b) Multilateral Comparisons

A serious limitation of the earlier PPP studies has been partly

overcome in the last study by Kravis, Kenessey, Heston and Summers.

Indeed, the earlier attempts had to be restricted to binary comparisons;

a country was chosen as a base to which the other countries were compared;

then if the other countries were compared between themselves using the

results yielded by their comparisons to the base country no consistency

1 This is more or less equivalent to stating that the two countries are

at different levels of the same indifference map. cf. assumption 2,

was present, (i.e. Fisher’s circular test1 was not met).

In A System of International Comparisons of Gross Produet and

Purchasing Power, Kravis et.al. affirm that they perform multilateral

comparisons which are base-country invariant and which meet Fisher’s

circular test. Their method named "country product dummy method" permits

them to deal with missing price data at the most detailed level of the

comparison and at the same time to incorporate in the study the informa­

tion carried by all the price data of all the countries. The method was

developed by Summers (1973) who proves that the coefficients of the

country dummies in a regression of the logarithms of all the available

prices (P) against dummies for all the countries (X^,) except one and

against dummies for all the items (Y^) in a specific category are the

PPP relative to the country without dummy.2 Furthermore, he argues that

PPP meets the requirements of country-base invariance, additive consist­

ency, the circular test and the factor reversal test. From an economet­

ric point of view, a rather abundant use of dummy variables is usually

not advisable. In fact, a very similar problem had already been studied

by Kloek and Theil (1965). They developed a method based on limited

information theory and were able to construct a single comparable index

of total expenditure for the six members of the European Coal and Steel

Community. From a theoretical point of view, Summers’ method is

questionable as it is replacing the missing price data for a certain

1 q(Q2»Q1) ^CQ-^Qq) = q(Q2,Q0) and

p(P2 »Pi) p(Pr P0) = P(P2 ,P0) where q(Q2 »Q1) is the quantity index

between country (or period) 2 and 1 and p(P2 ,P^) is the corresponding price index etc.

His equation is the following:

ln P 3, X + + 3 aj 1 al n-1 n is number of countries, j = 1 , .. A is number of items, a = 1, .., A. X -1 + Y-. Y . a,n-1 1 lj + y ay aA Aj.+ v

country by some average of the experience of all the other countries

included in the comparison, thereby imposing on the country some hybrid

foreign price structure. Hence we are back to the same basic theoretical

criticism of overlooking the difference in taste.

(c) Extension of the PPP method with non-monetary indicators

We must note that the PPP approach was a monetary approach.

However, an application of the PPP method using non-monetary indicators

was developed by Beckerman (1966). This study is very similar to Heston’s

(1973) and differs only in the following respect.1 The dependent

variables used in the regressions are not exchange rate GNP, but are the

’real consumption’ data estimated by various PPP studies. They are re­

gressed against several non-monetary indicators. Then the coefficients

yielded by the multiple regression are used to predict 'real consumption’

for other countries on the basis of their scores on the selected non­

monetary indicators.

This method combines the monetary and the non-monetary approaches.

However, the main problem faced by the author concerns the small number

of existing purchasing power parity studies available at the time.

2

These equations are based on the data for nine developed western countries

plus four other observations for Japan, USSR, China and India. As these

are mainly developed countries, the author does not have a very represent­

ative sample of countries in order to carry out predictions for a larger

group of developed and less developed countries. Several equations also

Heston's method was discussed earlier.

2 Beckerman also uses for each country two observations corresponding to two different years but does not include a dummy in order to

isolate the variations due to the time trend in the pooled regressions as he assumes that the observations are independent because they are five years apart.

had to be derived in order to predict the level in various countries.

If PPP studies were available for a larger and more representative

sample of countries, this method would offer a very valuable short-cut

to the estimation of PPP consumption in various countries.

(d) Criticism of the PPP methods

The PPP methods can be criticised on two levels: first on

pragmatic grounds and second on theoretical grounds.

First, the PPP studies are very lengthy and expensive to

undertake. They also involve a lot of corrections, adjustments,

definitions that are judgmental and arbitrary. Hence there is room for

a lot of personal bias, and the nationality of the investigator might be

reflected in the choice of the basic basket of goods chosen.

However, these are not very serious shortcomings. On the other

hand, the PPP methods face more ominous criticism on theoretical grounds.

Indeed, it was shown that the PPP methods always yield Paasche-type and

Laspeyres-type indices and the justification for presenting two such

indices is as follows: the exact index is assumed to be bracketed by

them. Furthermore, some average of the two indices offers a sort of

approximation to this exact index. Unfortunately it was shown in the

section on the theory of index numbers that the Paasche and the

Laspeyres quantity indices provides lower and upper bounds to the exact

index only under very restrictive assumptions which were then spelled

out. The question is now to appraise how realistic these assumptions

are. Hence we will first discuss the likelihood of a common indifference

map for all the countries and then raise the question of homotheticity

of the utility function.

Common indifference map

An exact index exists only if the various countries have the

same tastes, i.e. if they share the same indifference map or the same

to their budget constraints and their relative prices.

This is a very restrictive assumption. Indeed, if we accept

the existence of national tastes embedded in a national utility function,

we know that the national tastes in the USA are very different from the

national tastes in Mali; these two countries could not possibly share

the same indifference map. Of course tastes might be similar in similar

countries; for instance the EEC countries might have very resembling

utility functions. So if we are comparing very similar countries, the

assumption of a common indifference map for practical purposes should not

distort the results too much. However, if we are comparing a whole range

of countries from the very poor to the highly developed ones, the

assumption becomes quite unrealistic.

Homotheticity of the utility function

Such a condition implies unit income elasticity for all goods

as the budget constraint shifts upward, i.e. as income increases. More

explicitly, it means that, when faced by the same relative prices, if

the average citizen of a rich country consumes twice as much of any one

commodity as the citizen of a poor country consumes, he will also

choose to consume twice as much for instance of any other commodity;

twice as much bread as well as twice as many cars etc.

Such a consumption pattern definitely goes against the

empirical evidence given by the Engel’s curves. We know that, as people

become richer, they consume relatively more of certain goods, so defined

as luxuries, and relatively less of other goods, so defined as necessities.

Since the Engel's path which shows all the points of tangency between

the price lines and the various indifference curves as the budget con­

straint shifts to higher levels is not a straight line, this provides a

Fortunately, Samuelson has also developed the theory in the

case of non-homotheticity1 and we can thus relate it to the results of

the PPP studies.

Non-homotheticity

As a consequence of the introduction of non-homotheticity, we

believe that the assumption of a common indifference map for all

countries becomes much more acceptable for the following reasons. It is

now consistent with the actual pattern of consumption for the various

countries ranked along their level of development. Thus Gilbert's and

Kravis's statement that the various countries do not really exhibit

difference in the pattern of taste, but represent different levels of

development, becomes more sensible when the existence of luxuries and of

necessities is acknowledged.

It would now be interesting to turn to some empirical studies

and to examine them in the light of these findings. Unfortunately, a

straightforward application of Samuelson's theory cannot be carried out

unambiguously. Indeed, Samuelson's demonstration is based on two commod­

ities, one being a necessity and the other a luxury. When we consider

actual inter-country comparisons, we have to classify the commodities

into necessities and luxuries and weight them with their prices in order

to obtain total expenditure on necessities and total expenditure on

luxuries. First, an index problem immediately appears. Which price

structure is relevant? Should we use each country's own prices? Secondly,

commodities are usually classified into necessities and luxuries with

the help of national surveys, i.e. with reference to the same group of

people. In an article commemorating the centenary of Engel's Law}

Houthakker (1957) presented a comparative study of household expenditure

patterns for 25 different countries; however, all his elasticities

were calculated from regressions based on national series.1 On the

other hand, Samuelson's demonstration rests on a classification of the

commodities into necessities and luxuries based on cross-country con­

sumption patterns. Since this does not refer to the same group of people,

the meaning of such classification becomes somewhat different. What

causes the tilt of the cross-country Engel’s path? The fact that some

goods are necessities and others are luxuries for the various countries,

(i.e. we assume one common indifference map for the various countries),

or the fact that tastes are really different in different countries,

(i.e. we actually have a specific indifference map for each country but

we are linking together along the Engel's path the respective indiffer­

ence curves where each country stands at this specific time). In Table

2.2, we will present various necessities-luxuries classifications based

on a number of national surveys and cross-country surveys. We will also

refer to pure income elasticities when the price elasticities have been

taken into account and to approximate income elasticities when the price

effect has not been isolated.

The last major difficulty in applying Samuelson’s theory to

empirical data is precisely linked to the above distinction between pure

and approximate income elasticities. Indeed, as prices and income change

simultaneously, it is very difficult to isolate the pure income effect.

As purchasing power parity studies have to collect price data, they are

in a better position to value the pure income elasticities by also

including prices as independent variables in their regressions. Unfort­

unately, with the exception of food and education, the two PPP studies

included in the above table do not show much agreement.

Table 2.2

Various Approximations of Income Elasticities

I. Houthakker - National Surveys of 25 Countries

Approximate Income Elasticities

Low Average High

Income Inelastic:

Food .43 .57 .69

Housing

(incl. fuel and light) .48 .81 1.11

(not incl. furniture)

Income Elastic:

Clothing 1.05 1.31 1.77

Misc.

II. a) Gilbert and Associates b) Kravis, Kenessey et.al.

Cross country - 9 European countries and the USA

Cross country - 1 0 countries

Pure income Pure income

Elasticities Elasticities

Income Income

Inelastic: Inelastic:

Food .54 Food .63

alcohol, bev. .77 supply and opera- ^

tion of housing

tobacco .88 purchased transport .58

Clothing .84 Education .78

Housing .81

Income

purchase of trans- ^ Elastic:

port equipment meat 1.21 Education .75 beverage 1.18 Income tobacco 1.15 Elastic: Clothing 1.40 non-alcoholic bev. 1.13

Housing (rent and ..

footwear 1.01 fuel) i * J

furniture 2.10 house furnishings 1.12

fuel and light 1.19 purchase of trans- ^

Table 2.2 (Cont’d)

Income Inelastic: