Professional Documents
Culture Documents
An Extended Fakre) Lovell Technical E$Ciency Measure: W. Briec
An Extended Fakre) Lovell Technical E$Ciency Measure: W. Briec
Abstract
Recently, Chambers et al. (Journal of Economic Theory 70 (1996) 407}419; Working Paper, Southern Illinois
University, Carbondale, 1996) have developed the concept of directional distance to measure technical e$ciency.
However, there is no guarantee that such a measure must intersect the e$cient subset (also called strong e$cient subset)
de"ned by Koopmans (Activity Analysis of Production and Allocation, 1951, Vol. 36, pp. 27}56), that is primarily the
Pareto e$cient subset. So, we develop a framework for measuring e$ciency in the full input/output space. Following
FaK re and Lovell (Journal of Economic Theory 19 (1978) 150}162), we introduce a graph-type extension of the Russell
measure. Moreover, we show that our new measure can be computed by using linear programming. 2000 Elsevier
Science B.V. All rights reserved.
Keywords: E$ciency indices; Graph measure; Technical e$ciency; Production technology; Distance functions
1. Introduction
In recent years, a number of studies on the theoretical and empirical measurement of technical e$ciency has been generated by researchers, and
two di!erent notions of technical e$ciency have
emerged in the economic literature. The "rst notion
due to Koopmans [1], de"nes a producer as technically e$cient if a decrease in any input requires
a increase of at least one other input. This de"nition
is closely related to the Pareto e$ciency concept,
0925-5273/00/$ - see front matter 2000 Elsevier Science B.V. All rights reserved.
PII: S 0 9 2 5 - 5 2 7 3 ( 9 9 ) 0 0 0 1 9 - 5
192
(1)
K
K
" (u, v)3RL>N; u7 h xH, v6 h yH, h3C .
>
H
H
H
H
(2)
The set C characterises the returns to scale chosen
by the producer. It is possible to characterise constant returns to scale (see [5]), various returns to
scale [14], non-increasing returns to scale [15,16],
non-decreasing returns to scale [17]:
(a) Constant returns to scale [5]:
C"
: C "RK .
!01
>
(b) Various returns to scale [14]:
(3)
(4)
(5)
K
C"
: C " h3RK ; h "1 .
401
>
G
G
(c) Non-increasing returns to scale [17]:
K
C"
: C " h3RK ; h "1 .
>
G
,'01
G
(d) Non-decreasing returns to scale [15,16]:
K
C"
: C
" h3RK ; h *1 .
(6)
,"01
>
G
G
Note that all the above speci"cations of the production set imply the convexity assumption. It is
however, possible to drop the convexity assumptions by using the FDH methodology by Tulkens
and Vanden-Eeckaut [18]. In such case the set
C can be characterized as follows:
193
K
C"
: C " h3+0, 1,K; h "1 .
$"&
G
G
(7)
194
Fig. 1.
1
j ; jx3(y), j 3[0, 1]
"min
G
G
Card(I(x))
GZ'V
H
1
"max
k ; ky3P(x), k *1
H
H
Card(J(y))
I
HZ(7
195
1
1
D?@(x, y)"max
d#
e;
20
G
H
r(x, a)
s(y,
b)
5
BC
GZ'6 '?
HZ(75(@
((I!dA)x, (I#eB)y)3
196
(8)
of the maximisation program computing the Russell proportional distance. Since T2 holds,
+(u, v)3RL>N; u6x, v7y, is closed and bounded,
>
thus (xH, yH) belongs to the production set T and
D?@(x, y) is de"ned. Moreover, there exists i3I(x)
20
5I(a) or j3J(y)5J(b) such that xH(x or yH'y .
G
G
H
H
Thus, (x, y) does not belong to the e$cient subset.
Consequently, *)()L+(x, y)3; D?@(x, y)"0,,
20
and since HL we have *)(H)L+(x, y)3H;
D?@(x, y)"0,. Conversely, if (x, y)3H, and
20
D?@(x, y)"0, since y'0 it is obvious that
20
(x, y)3*)( ). Thus, (x, y)3*)(H) and (a) is proved.
The proofs of (b) and (c) are similar.
(2) Assume that j"I!d. Assume that for all
i3+1,2, n, and j3+1,2, p, we have a "1 and
G
b "0, thus
H
D?@(x, y)
20
1
1
d#
e;
"max
G
H
s(y, b)
r(x, a)
GZ'?5'V
HZ(@5(W
BC
((I!dA)x, (I#dB)y)3
1
"max 1!
j ; (jx, y)3 ,
G
Card(I(x)) 3
H
G 'V
therefore, we have the relationship
D?@(x, y)
20
1
"1!min
j ; (jx, y)3 .
G
Card(I(x)) 3
H
G 'V
Consequently D?@(x, y)"1!E (x, y).
20
0
(3) The proof is similar to (2) making the transformation h"I#d.
(4) First, we prove the input monotonicity. Let
(x, y)3, with x6x and assume that X(x, y)"
+(d, e)3RL>N; ((I!dA)x, (I#eB)y)3,, thus we
>
have X(x, y)LX(x, y) and the input monotonicity
is proved. Now, assume that y7y'0, since
y'0 we have necessarily X(x, y)M(x, y). Thus,
the output monotonicity holds. We deduce immediately (4).
(5) Let L be a (n#p);(n#p) positive diagonal
matrices de"ned over RL>N such that
"
1
1
"max
d#
e;
G
H
r(x, a)
s(y,
b)
5
5
BC
GZ'? '6
HZ(@ (7
1
1
d#
e;
"max
G
H
s(y,
b)
r(x, a)
BC
GZ'?5'6
HZ(@5(7
((I!dA)x, (I#eA)y)3 .
Since L is an isomorphism over RL>N, thus
D?@ ( x, y)
*20
1
1
"max
d#
e;
G
H
r(x, a)
s(y,
b)
BC
GZ'?5'6
HZ(@5(7
((I!dA)x, (I#eA)y)3
197
But we have
+(d, e)3RL>N; ((I!dA)x, (I#eB)y)3, d "e "j,
>
G
H
L+(d, e)3RL>N; ((I!dA)x, (I#eB)y)3,.
>
Thus, we deduce immediately that D?@(x, y))
2$
D?@(x, y).
20
The fourth result proves that the generalised
Russell}FaK re}Lovell measure characterises the
observed decision-making units that are strongly
e$cient. In the same idea, it is possible to relate the
input and output e$cient subsets and the Russell
proportional distance. Property (4) shows that the
Russell proportional distance function satis"es the
weak monotonicity condition. A similar proof is for
instance given in Ref. [21]. (5) proves that the
proportional distance is invariant with respect to
a change in the units of measurement.
The property (6) states a comparison between the
Farrell and Russell proportional distances. A similar result is obtained by FaK re et al. [11] alternatively in the input or output space. From properties
(2) and (3), it is obvious that (6) is in accordance
with the result obtained by FaK re et al.
Now, we focus on the particular relationship
between the proportional distance and the Russell
input and output measurements of technical e$ciency. Assume that b"0. The input set dimension
is two. Fig. 2 illustrates the di!erent cases.
Assume that a "a "1, the Russell propor
tional distance function coincides with the Russell
measurement of technical e$ciency and we have
the relationship D?@"1!EG . The factors of
0
0
1
1
"max
d#
e;
G
H
r(x, a)
s(y,
b)
5
5
BC
GZ'? 'V
HZ(@ (W
Fig. 2.
198
1
1
d#
e ,
max
G s(yI, b)
H
r(xI, a)
GZ'?
HZ(@
BC
K
(I!dA)xI7 h xG,
G
G
K
(I#eB)yI6 h yG,
G
G
d, e60, h3C,
(P)
4. Conclusion
The foregoing discussion is entirely in the spirit
of the recent literature on measures of technical
e$ciency, discussing radial and non-radial
measures. The measure we have introduced above
is a generalisation, in the full-input space, of the
Russell measure of technical e$ciency, introduce
by FaK re}Lovell [2]. From the practical standpoint
this measure has the advantage to select a strong
e$cient vector onto the frontier of technical
Acknowledgements
I am grateful to C. Blackorby for helpful suggestions and I would like to thank two anoymous
referees for comments that greatly improved the
exposition of the paper.
References
199