Está en la página 1de 12

1

THE DEMAND FOR MEAT IN CANADA: AN APPLICATION OF THE TRANSFORMATION OF THE VARIABLES.

Zuhair A. Hassan and Stanley R. Johnson*


The transformation of variables analysis has been applied i this study to estimate the demand for n five meat products. Results show that the demand functions take a variety of forms. These forms are not the frequently employed linear or log-linear functions. The elasticities derived from the appropriate functional forms are at times very different from those that would have resulted from an arbitary specification. The estimated expressions can also be viewed as linear functions with variational parameters. Lanalyse de transformation des variables a Ctk appliquk dam ce rapport afin destimer la demande de cinq produits de viande. Selon les rbsultats, les fonctions qui dkrivent la demande prennent diverses formes. Ces formes ne sont pas celles de fonctions lintaires ou log-linkires courantes. Les elasticitks dkrivtes de ces formes fonctionnelles approprik sont quelquefois t r t s diffkrentes de celles qui rtsultent dune specification arbitraire. Les expressions estimtes peuvent aussi &re considkrks comme des fonctions linkaires qui contiennent des paramttres variationnels.

Introduction The analysis of consumer demand at disaggregated commodity levels is usually conducted without the benefit of the richer (Ipriori information available for complete demand systems. Arguments in the demand functions may include own price, prices of one or two commodities thought to be substitutes and income. Past experience and institutional information on consumption habits guides the selection of the substitute variables for inclusion in the equations. The estimated demand function must be regarded as rough approximations of the complete system which determines behavior with respect to the commodities in question. Possibly because of the lack of (Ipriori information the applied work on consumer demand at disaggregated commodity levels has employed a variety of functional forms. Although the forms may permit a closer approximation of the more general demand system, they impose strong conditions on the demand elasticities: typically the n structural characteristics of major interest. Thus, while flexibility i the choice of functional form may be useful from the viewpoint of improved statistical resolution in the sample data, the choice should be made on a systematic basis. Recent developments in the transformation of variables have made it possible to select functional forms in a very general context. This is accomplished by estimating two additional parameters in a comparatively simple algebraic specification of the model to be studied. The present paper applies this functional form analysis to the demand for meat in Canada. Demand equations for beef, pork, veal, chicken and turkey are studied. Results obtained are informative from an applied viewpoint and indicative of the importance of the choice of functional form.
*

Agriculture Canada and University of Missouri. Columbia, respectively. A computer program provided by Charley Huang and Robert Raunikar, University of Georgia, was used in the analysis.

Canadian Journal of Agricultural Economics 27(3). 1979

2 Functional Form The transformation applied in the analysis of functional form was suggested by Box and Cox [2]. As the computer software for handling this and similar nonlinear problems has become available, the analysis of functional form has been applied in a number of contexts [1,3,4,8,9]. The transformation employed in the study of functional form is A P P ... + Pk (Xki - l)/p + f j (yi - i)/A = Po + (Xh - I)/ (11

for i = 1.2, ..., n. In this model yi is the dependent or endogenous variable, xji, j = 1,2, ..., k, the explanatory or non stochastic independent variables, ~i a structural disturbance, /3j = 0,1,2, ..., k, the structural parameters and finally, A and p the transformation parameters. For n sample observations the model in (1) can be written in matrix form as

y* = x*$

+ f.

(2)

where the dimensions of the matrices and vectors in equation (2) are implied by the correspondences with equation (1). The functional form for equation (1) is determined by values for parameters A and pFor example letting A = p = 1, equation (1) becomes
yi =

00 +

xfi

+ ... + PpXki + C i

(3)

with yi a linear function of the explanatory variables and the unobservable structural disturbance. Similarly, for A = 1 and p = -1 equation (1) is of the inverse functional form. o equation (1) approaches the double log At the other extreme, as 1 o and p functional form. This is shown by observing that any positive and finite number say z, can be written as z = e b and that elm can be expanded as
+ +

Now consider the dependent variable from equation (1). From the result shown in equation (4) it is clear that we can write

For A o the expression ( 5 ) has the limiting value, (yA - 1)/ = lnyi 15, p. 4 7 An 6. identical result holds for the explanatory variables in equation (1) when the transformation form parameter p goes to zero. As these are limiting expressions for A and p, the transformation is continuous in the neighborhood of zero. The log inverse and semi log functional forms are also special cases of equation (1). From the above results it should be apparent that if A o and p = -1 then equation (1) approaches the log inverse functional form. A similar reasoning will show that for A = 1 and p o equation (1) becomes semi-log functional form. Thus equation (1) through the variation of parameters A and p in the interval (-1, 1) can provide a family of
+

functional forms. It is of interest for the present analysis that the family of functional forms includes those which have been customarily applied in the analysis of consumer demand.
Estimation If the structural disturbance from equation (1) is assumed normally and independently distributed with mean zero and finite variance 02, then the likelihood function for the sample can be expressed as

L (J,02.1, p) = (2n oq-n/2 exp [-(y' -

- x*p) (y* - x*p/203J, -- -

(6)

where J is the Jacobian for the transformation to the observed random variables yi. Estimates for the parameters, p, 9, and p, can be obtained by maximizing the likelihood function (6)as expressed in log form. For given d u e s of A and p, the likelihood function can be maximized for the parameter vector - and 9. resulting /3 The compressed likelihood function is of the form n L (Lp) = U n o 2 (A.p) + (1- 1) 1h y i (7) 2 i= 1 where 2 (1.p) is the variance estimate for She structural disturbances conditioned on the parameters A and p. Hence (7) is maximized by choosing A and JA to minimize 82 (A,p). Since A and p must be in the interval (-1,l) the maximum of (7)can be obtained by employing a numerical search over a grid of the parameter values. The pair of values for 1 and p which yield the ordinary least squares estimates of the parameters1 giving the smallest residual variance estimator then approximate the maximum likelihood estimates for A and p and corresponding, 1and 02 191. As the estimation has taken place by maximum likelihood, the likelihood ratio test can be used to obtain an approximate confidence region for 1and p. That is. from the standard likelihood ratio test it follows that

with x2 the Chi square statistic with two degrees of freedom and critical value IT. This 2'Q result is of use in evaluating the ability of the sample data to distinguish between the members of the family of functional forms.
The Model

The five equations linking per capita disappearance of beef, pork, veal, chicken and turkey to prices and consumer income are of the form Retail demand for beef PCFB; RPBF, RPPK, RPVL, PCDY, SUDi. Retail demand for pork: PCPK; RPBF, RPPK, PCDY, SUDi. Retail demand for veal: PCVL; RPBF, RPPK, PCDY, SUDi. Retail demand for chicken: PCCK RPCK, RPHB,PCDY, SUDi. Retail demand for turkey: PCTK; RPCK,RPTK, PCDY, SUDi. (9)
-

(10) (1 1)

(12) (13)

where PCBF is the per capita disappearance of beef in pounds; PCPK the per capita disappearance of pork in pounds; PCVL the per capita disappearance of veal in pounds, PCCK the per capita disappearance of chicken in pounds, PCTK the per capita disappearance of turkey in pounds, RPBF the consumer price index of beef (1971 = loo), RPPK the consumer price index of pork (1971 = loo), RPVL the consumer price index of veal (1971 = 100). RPCK the consumer price index of chicken (1971 = loo), RPHB the consumer price index of hamburger (1971 = loo), RPTK the consumer price index of turkey (1971 = loo), PCDY the per capita disposable income in current dollars and the SUDi are seasonal dummies for second, third and fourth quarter respectively; i = 2, 3, 4. For the above set of equations, per capita disappearance of each commodity is expressed as a function of the own price, price(s) of other commodities and per capita disposable income. The five demand equations are thus typical of those used to study consumption behavior in applied contexts at disaggregated commodity levels.
The Data

To estimate the various price and income parameters, use was made of time series data for Canada. Specifically, quarterly observations on per capita meat and poultry disappearance, consumer price indexes (1971 = 100) and per capita personal disposable income for the period first quarter 1965 to fourth quarter 1976 were employed. Sources of prices, income and disappearance data are [6,7] and files of the Livestock Division, Statistics Canada, respectively.
The Results

The original observations (except for the constant term and quarterly dummy variables) were first transformed according to equation (1) based on assumed values for the related parameters A and p. The ordinary least square procedure was then used to obtain estimates of demand parameters. Estimates of the five demand equations for selected values of A and p are presented in Tables 1-5. In the tables, the standard error for each of the structural coefficients is shown in the parenthesis beneath the reported estimate. The Durbin-Watson statistic (D-W), the coefficient of multiple determination (R2), the transformation parameters (A and p) and the value of the log of the likelihood function are presented in the last columns of the tables. In general, the estimated relationships conform with the general results from the theory and previous applied work. All estimated price coefficients have the expected signs and all the estimated income coefficients are positive. In addition, with the exception of a few coefficients, all the estimates for the structural coefficients are more than twice their corresponding standard errors. The results also reveal: The likelihood functions for individual meats are maximized at substantially different values of 1 and p: (1.0, 0.4) for beef (Table l), (0.20, -0.40) for pork 00 (Table 2), (0.50, 0.20) for veal (Table 3), ( . ,0.40) for chicken (Table 4) and (1.0, -1.0) for turkey (Table 5). This implies that functional form best approximating the true structure over the sample period is different from one commodity to another.

TABLE 1 Explanatory Variables

suo2
SU03

DEMAND EQUATIONS F R BEEF. BOX-COX PROCEDURE, (FIRST OUARTER 1965 O

- FOURTH OUPRTER 1976)


RPVL
PCOY

Funct 1on
SU04
RPPK

Constant
RPBF

0-Wb

R2
A
1 1 18.53'

Linear
2.50*

0.00R6
0.0464 0,9211 (o.00~)

19.93 (0.6786)~

-0.4077 (0.2813)

-0.5200 (0.2869)

-0,1477 (0.3151)

-0.0889 (0.0246)

(n.0222)
z.50* 0.9116

fl.00R3 (0.0014)

OoubleLogarithmic -0.0134 (0.0126)

1.88 (0.1247)

-0.0223
(0.0127) 2.52* 2.39* 0.8955 0.0993 0.9237 0.9225 1

-0.0147 (0.0141)

-0.4274 (0.1210)

0.0662 (0.0497)

o.ngw (0.1058)

0.3704 (0.0555)

o
0

17.92~

Semi Logarithmic -0.2621 (0.2806) 2.35 (2.34) -n.i318 (11.10) 15.50 (257.97) 8277.60 (1098.93) 342.74 (47.31) 9.07 (1.23) -0.0057 -0.4590 (0.2825) -0.0170 (0.0138) -53.39 (14.26)
17.80 (6.09)

-0.3595 (0.3137) -10.98 ( 2.69) 1.85 (1.10)

-6.73
-0.0247 (0.0155)
-0.5868 (0.3602) -1373.71 (331.41) 472.27 (141.64)

18.93d

(2.75)

Log-Inverse

-303.78 (37.84)

o
1

-1

13.88
-1 1
0.40

(0.0139)
-0.0875 (0.3237) -0.3306 (0.3205)

Inverse

-7367.29 (878.99)

2.35*

12.62 19.~7~

General
-0.3258 (0.2777)

14.93 (1.34) -0.4916 (0.2812) 0.5313 (0.3629) 0.5633 (0.0829) -0.2685 (0.3108) -1.59 (0.4079) 0.2029 (0.1625)

2.5~

a b c

*
i n d i c a t e s inconclusive t e s t a t the f i v e percent l e v e l .

Standard e r r o r s are i n parentheses. 0-W i s the Durbin-Watson s t a t i s t i c ; The maximum l l k e l i h o o d estimate. The 95 percent confidence region.

O EQUATIONS FOR PORK, BOX-COX PROCEDURE, (FIRST QUARTER 1965


FOURTH QUARTER 1976)

RPPK

m.

Explanatory Variables SU04


RPBF

Constant PCDY 0.0091 (0.0012) 0.8201 1 15.78 0.6544* D-Wb

suo2

SUD3

~
1

16.07 (0.7170)a 0.2731 (0.0743) 0.4827 (0.0593) 0.~539. 0.8949 0 0 29.2gd 6.89 (0.8345) n.8367* 0.8932 ~8.32~ 28.4gd 2B.05d 30.15' 0.8912 0 1 0.20 -0.40 -1 -1 0.8920 0.8987 1 0.8755* 0.8443* 0.8804* 263.31 (49.71) 3821.26 (690.76) 10.59 (1.45) 3.54 (1.04) 33.77 (10.29) 436.51 (142.99) 3.03 (0.8891) -10.03 (0.5818) -1426.21 (85.42) -101.09 (6.14) -12.39 (0.7337) -0.~799 (0.0521 )

-0.1028 (0.2968)

-0.9172 (0.2993)

-1 ,6179 (0.3085)

n.0373 (0.0095)
-n.n939 (0.0075)

2.49 (0.1319) -1.58 (0.2386)

-0.0067 (0.0163)

-0.n599 (0.0163)

-0.1140 (0.0169)

11.87 (1.85)

-0.0860 (0.2300)

-0.8582 (0.2297)

-193.41 ( 38.51)

-0.0088 (0.0168)

-0.0527 (0.0165)

-0.1036 (0.01 74)

-2819.57 (535.17)

-0.1156 (0.2334)

-0.7558 (0.2303)

-1.4405 (0.2420)

-5.94 (1.30)

-0.oi20 (0.02n)

-0.0970 (0.0270)

-0.1868 (0.0282)

o r s are i n parentheses. Durbin-Watson s t a t i s t i c ; *indicates s i g n i f i c a n t 0-U a t the f i v e percent l e v e l . I l k e l f h o o d estimate. ent confidence region.

TABLE 3 Explanatory Variables Constant


SUOZ
PCOY
A
y
~~

FOURTH QUARTER 1976) SU03


1

DEMAND EQUATIONS FOR VEAL, BOX-COX PROCEDURE, (FIRST QUARTER 1965

RPBF
R2

Function SU04 RPVL 0-Wb 0.76'


1

Linear

-0.1710

o.ooio
0.8662 1 (0.0001) 1.01 (0.1375)
0

85.97

2.61 (0.1492)'
-0.1709 (0.0687)

(0.0680)
-0.1116 (0.0469)
-3.11 (0.3231)

-0.2921 (0.0697)
0.0218 (0.0050)

-0.0402 (0.0042) 0.87* 0.8911 0

OoubleLogarl thmic 2.94 (0.4664)

-0.0866 (0.0476)
1.09
(0.1688)

-0.2348 (0.0488) -0.3216 (0.0599) -4.39 (0.3963) -267.60


0.8286
(40.24)

1.10 (0.4026) 2.16 (0.4939) 1.04' 0.7797* 1.00' 0.9447' 0.8772 0.9039 0.9044 44.68 (55.21) 829.90 (172.81) 1 0

91.89

Semi Logarithmic 3.52 (0.5723) -0.0794


(0.0606)

-0.1585 (0.0585) -0.1106 (0.0589) -0.2428 (0.0633) -0.1779 (0.0576) -607.71 (139.73)

94.05d

Log-Inverse

0
1 0.50

-1

81.07

Inverse -647.42 (154.75)


'

-0.1494 (0.0671) -395.31 (44.57) 0.2621 (0.0361) 884.51 (191.40)

-0.1812 (0.0653)

-0.3330 (0.0701)

157.55 (61.15)

-1
0.20

88.05 95.11'

2.90 -0.1211 -0.1383 -0.2659 0.6574 -1.46 (0.3246) (0.0504) (0.0497) (0.05161 (0.1656) (0. 13491 a Standard e r r o r s are i n oarentheses. b 0-W i s the Ourbin-Hatson s t a t i s t i c ; 'indicates s i g n i f i c a n t 0-W a t the f i v e percent l e v e l . c The maximum l i k e l i h o o d estimate. d The 95 percent confidence region.

General

TABLt 4
.-__

FOURTH QUARTER 1976) Constant


SUOZ
PCOY h R2

DEHANO EQUATIONS FOR CHICKEN, BOX-COX PROCEDURE, (FIRST QUARTER 1965

PPHB
0-Wb
.66*

Funct ion

Explanatory Variables SU03 SU04 RPCK

u
1

L
44.47

Linear

0.8288
(0.1 719) 0.1083 (0.0228) 0.3387 (0.0533) .64* 0.9252 (0.1780) 3.81 (0.5277) 295.03 (52.35) 1893.19 (416.64) 0.0434 (0.0047 ) 0.1251 (0.0247) 1.05 (0.19) 0.1021 (0.0224) -0.1060 (0.0122) 0.0569 (0.007s) -333.09 (73.23) 222.93 (60.51) -52.65 (9.20) 32.40 (7.60) -4.33 (0.6368) -0.6511 (0.081 5) 0.5673 (0.0675) 0.9082

5.35 (0.291 2 0.0058 ( 0 .O006) 0.8990 -0.2338 (0.1657) -8.07 -272.54 (41.03) -1774.05 -0.0731 (0.0236) -0.4787 (0.1881) -0.0699 (0.0214) -0lb261 (0.021 1 ) -0.1342 (0.1855) -0.0204 (0.0233) -0.4514 (0.1702) -0.1560 (0.1680) -0.0702 -0.0238 (0.021 8) (0.021 5 )

-0.4552 (0.1634)

-0.1970 (0.161 7)

-0.0437 (0.0053)

0.0260 (0.0032 )

DoubleLogarithmic

47.44d

Semi Logarithmic

2.28 (0.4171)

.74*
.60*

0.8909 0.8923

1 0

0 -1

42.64 43.60

Log-Inverse

Inverse

.61* 1.59*

0.8671 0.9110

1 0

-1 0.40

37.90 48.10'

(326.52)
1.31 (0.0764)

General

a Standard e r r o r s are i n parentheses. b D-W i s the Durbin-Watson s t a t i s t i c ; *indicates inconclusive 0-W a t the f i v e percent l e v e l .

c d

The maximum l i k e l i h o o d estimate. The 95 percent confidence reglon.

TABLE 5

Constant
RPTK

DEPANO EQUATIONS FOR TURKEY. BOX-COX PROCEDURE, (FIRST QUARTER 1965

- FOURTH 1976)
PCDY
0-Wb

Funct 1on -0.00~5 0.9859


1

SU02

SU03

Explanatory V a r l abl es SUD4 RPCK

RZ
L 1 82.~12~

Linear
(0.0036)

5.01 (0.1251la (0.0003) 0.1451 (0.1279)


0 0

-3.48 (0.0731) -1.26 (0.0420) -1.05 (0.0416) -0.8493 (0.0448) 0.1471 (0.2766) -0.3582 (0.2883) 1.56* 0.9634

-3.14 (0.0727)

-2.73 (0.0798)

0.0050 (0.0033)

o.ononi

i.v*

DoubleLogari thmlc 1.63 (0.2918)


4.53 .

71.53

Semilogarlthalc

1.60*
1.61* 1.64* 0.9861 0.9641

0.9859

1
0

83.00d

(0.5092) -0.8519
(0.0434)

-3.47 (0.0733) 0.1693 (0.2233) 128.55 (75.24) 173.35 (131.89) -1.26 (0.0416) -3.46 (0.0729) -3.14 (0.0722) -2.75 (0.0762) 70.74 (57.74) -125.96 (60.84) -1.04 (0.0412) 14.61 (32.93) -44.11 (34.70)

-3.14 (0.0726)

-2.75 (0.0782)

0.6598 (0.4827)

-1.03 (0.5032)

Log-Inverse -1 14.51 (105.38)

-97.55 (60.11)

-1 1 -1

70.29 83.22'

Inverse

a b c d

Standard e r r o r s are i n parentheses. 0-W i s the Durbin-Watson s t a t l s t i c ; *indicates inconclusive 0-W a t the f i v e percent l e v e l . The aaxlmua l l k e l i h o o d estimate. The 95 percent confidence region.

10

The 95% confidence region of A and 1 respectively for individual meats include: (1.0, l.O), (0.0, 0.0) and (1.0, 0.0) for beef; (0.0, O.O), (1.0, O.O), (0.0, -1.0) and (1.0, -1.0) for pork; (1.0, 0.0) for veal; (0.0, 0.0) for chicken; (1.0, 1.0) and (1.0, 0.0) for turkey. These results indicate the functional form for the beef demand equation can be linear or double-log or semi-log. For pork the double-log, semilog and log inverse functional forms cannot be rejected. For veal and chicken the semi-log and the double-log functional forms cannot be rejected. Finally for turkey the linear form is within the confidence region. The results also show that the appropriate functional forms in some cases depart from the linear and double-log functions, expressions widely used in applied work. For the purpose of interpreting these results in a more useful context it is of interest to examine the price and income elasticities (Tables 6-7). The elasticities were obtained from the parameter estimates in Tables (1-5) by evaluating the usual price and income elasticities formula at the sample means. Applying the standard elasticity formula to equation (1) gives

where K and y a r e the means of the two variables and E . is the elasticity for Xj. This Y ! general form provides a convenient method for evaluating the elasticities at sample means and other assumed levels of the dependent and independent variables.

In examining the results in Tables 6 and 7, note as implied by the sign of the coefficient estimates, that al of the estimated direct price elasticities have the expected l negative sign (Table 6) and all of the estimated income elasticities are positive (Table 7). The results also indicate that the estimated elasticities (price and income) differ among various functional forms. For example, in the case of chicken the linear function showed the greatest response to own price (-0.7640), while the inverse function was the least responsive (-0.3717). The direct price elasticity implied by the maximum likelihood estimate of the appropriate general functional form of chicken was -0.7321. If the linear or the inverse functions were used instead to calculate the direct price elasticity for chicken, it would have been over-estimated by 4.35% and under-estimated by 49.23%, respectively. Similar comparisons for the other commodities can also be made. They indicate that errors on the order of 100 percent are possible in elasticities estimated from structural forms within the family but not optimal for the sample data.
Alternative Interpretation The non-linear functional form has an alternative interpretation, which is based on the idea of variational parameters. That is, the non-linear functional form can be considered as linear but with parameters that vary. This characterization of the estimated relationship although identical is useful as a basis for interpreting the results of the empirical work. As indicated earlier, the demand function specifications are but approximation of the true structure determining the consumption level. The linear functional form with variational parameters shows how the linear approximation must

TABLE 6 Beef Percentage Elasticity -3.13 98.11 97.49 101.88 79.62 87.46 -0.7321 -0.3717 50.77 -0.4048 -0.4206 57.45 -0.3551 -0.6056 82.72 -0.4143 -0.6511 88.93 -0.3582 88.48 102.34 87.72 -0.7640 -3.11 -3.25 -2.54 -2.79 -3.19 104.35 -0.3204 79.15
o f General

Veal Elasticity Elasticity Percentage o f General Elasticity -0.9103 -0.8799 -0.8943 -0.7534 -0.7672 -0.8358 91.79 90.14 107.00 105.27 108.91 91.65 94.30 106.68 110.92 125.57 Pork Percentage o f General Chicken Percentage o f General Turkey Percentage o f General

OWN PRICE ELASTICITIES FOR MEAT AND POULTRY

~~

Function

Elasticity

Linear

-0.4154

lbubleLogarithmic

-0.4274

SemiLogarithmic

-0.4835

Log-Inverse

-0.5027

Inverse

-0.5691

General

-0.4532

TABLE 7

Pork
Veal Percentage Elasticity
o f General

INCOME ELASTICITIES FOR MEAT AND POULTRY

Function 80.12 104.27 112.38 124.07 131.86 0.4367 0.3547 81.22 0.3386 77.53 1.06 0.8445 0.8552 0.4978 114.0 0.8120 0.4827 110.53 0.5113 117.08 0.5963 69.72 118.10 94.94 123.94 98.74

Elasticity

Beef Percentage o f General Elasticity Percentage o f General

Elasticity

Chicken Percentage o f General


0.6300

Elasticity 101.20 0.5673 0.5335 0.3794 0.3401 0.6225 91.13 85.70 60.94 54.63 0.0021 0.1451 0.0675 0.1653 0.0889

Turkey Percentage o f General 2.36 163.21 75.92 185.93

Linear

0.2846

OoubleLogarithmic

0.3704

1.01

Semi Logari thmic

0.3992

Log-Inverse

0.4407

Inverse

0.4684

General

0.3552

c .

12

adjust to explain the sample data. It can be easily shown that equation (1) can be written as

* P-1 A-1 . where /3j = /lj (Up)(Xji /yi ) ,J = 1, 2 ..., 1 and Po and Ej have been appropriately redefined. What this implies is that for A # p # 0,each coefficients value depends upon the magnitude of the dependent and independent variables j, the transformation parameters A and P and a constant pj. Hence, any member of the family of functions described by equation (1) has a very simple interpretation, provided that one is prepared to view the parameters in a variational context. Conclusion

The transformation of variables analysis has been applied to estimate the demand for five meat products. Results show that the demand functions take a variety of forms. In addition these forms are not the frequently employed linear and log-linear expressions. The elasticities calculated from the appropriate forms are at times 100 percent different from those that would have resulted from an arbitrary specification. It also has been shown that the estimated expressions can be viewed as linear functions with variational parameters. This makes it possible to study the effects of the transformation parameters A and p on the results of the empirical application in a straight-forward manner.

REFERENCES
1 Benus, J., Knnta and H. Shapiro. The Dynamics of Household Budget Allocation to Food Expenditure. The Review of Economics and Statistics, 57 (May. 1976) 129-138.

2 Box, G.E.P. and D.R. Cox. An Analysis of Transformation. Journal of the Royal Statistical Society, Series B, 26 (1964) 211-243. 3 Chang, H.S. Functional Forms and the Demand for Meat in the United States. TheReview of Economics and Statistics, 59 (February, 1977), 355-359. 4 Kau, J.B. and C.F. Lee. The Functional Form in Estimating the Density Gradient: An Empirical Investigation. Journal of the American Statistical Association, 77 (June, 1976)
326-327. 5 Kmenta, J. Elements ofEconomerrics. New York, New York: TheMacmillan Company, 1971. 6 Statistics Canada. Prices and Price Indexes. Catalogue No. 62-001, monthly. 7 Statistics Canada. National Income and Expenditure Accounts. Catalogue No. 13-201,

Annual. 8 Zarembka, P. Functional Form in the Demand for Money. Journal of the American Statistical Association, 63 (June, 1968) 502-5 11. 9 Zarembka. P. Transformation of Variables in Econometrics, in Frontiers ofEconometrics, ed. P. Zarembka, New York, New York: Academic Press, 1974, 81-104.

También podría gustarte