A global cost Malmquist productivity index using data envelopment analysis

Author(s): G Tohidi, S Razavyan and S Tohidnia

Source: The Journal of the Operational Research Society, Vol. 63, No. 1 (JANUARY 2012),
pp. 72-78
Published by: Palgrave Macmillan Journals on behalf of the Operational Research Society
Stable URL: http://www.jstor.org/stable/41353909
Accessed: 06-07-2016 17:24 UTC

REFERENCES
Linked references are available on JSTOR for this article:
http://www.jstor.org/stable/41353909?seq=1&cid=pdf-reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted
digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about
JSTOR, please contact support@jstor.org.

Operational Research Society, Palgrave Macmillan Journals are collaborating with JSTOR to
digitize, preserve and extend access to The Journal of the Operational Research Society

This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 17:24:53 UTC
All use subject to http://about.jstor.org/terms
Journal of the Operational Research Society (2012) 63, 72-78 2012 Operational Research Society Ltd. All rights reserved. 0160-5682/12

wvm.palgrave-journals.com/jors/

A global cost Malmquist productivity index

using data envelopment analysis
G Tohidi1*, S Razavyan2 and S Tohidnia1
islamic Azad University , Central Tehran Branch, Tehran , Iran; and 2 Islamic Azad University,
South-Tehran Branch, Tehran, Iran

This paper proposes a global cost Malmquist productivity index, new cost Malmquist productivity
index, that is circular and that gives a single measure of productivity change. The index is inspired by
the global Malmquist productivity index as extended to productivity measurement. Decomposition of
the proposed cost Malmquist productivity index is presented. Numerical results are presented for an
example taken from the literature to illustrate the proposed algorithm.
Journal of the Operational Research Society (2012) 63, 72-78. doi:10.1057/jors.201 1.23
Published online 13 April 2011

Keywords: linear programming; data envelopment analysis; allocative efficiency; circularity; Malmquist index

1. Introduction index with technology formed from data of all producers in

all periods. The proposed index generates a single measure
In recent years, the measurement of productivity change
of productivity change, it is immune to LP infeasibility,
has enjoyed a great deal of interest among researchers
and it satisfies circularity. This has been shown before by
studying firm performance and behaviour. In this frame-
Berg et al (1992), where a common (base) technology can
work, the Malmquist index was first introduced in pro-
be used to make the circular Malmquist index. Portela and
ductivity literature by Caves et al (1982). Fre et al (1989)
Thanassoulis (2008) introduced a meta-Malmquist index.
decomposed productivity change into technical efficiency
This index can be decomposed into circular components
change (TEC) and technical change and used non-
of efficiency change and technical change. Simplicity that
parametric mathematical programming models for its
enables the easy manipulation of data and the feasibility of
computation.
problems for computing the index under variable returns to
The Malmquist productivity index introduced by Caves
scale (VRS) are the main advantages of this index.
et al (1982) is not circular, that is, we may not derive the
The Malmquist index has seen many applications and
productivity change between time periods t and t + 2 when
extensions (Fre et al , 1998). But it does not capture allo-
we know the productivity change between time periods t
cative efficiency, which reflects the distance between the
and t + 1 , and between time periods t + 1 and t + 2. On the
actual and minimum cost at which a production unit may
other hand, its adjacent period components can provide
secure its outputs once any technical inefficiency of the unit
different measures of productivity change.
has been eliminated (Fre and Grosskopf, 1992). Mania-
Sufficient conditions on the adjacent period technologies
dakis and Thanassoulis (2004) proposed an approach to
were stated by Fre and Grosskopf (1996) for the index to
decompose productivity change so that the contribution of
satisfy circularity, and to average the same measures of
allocative efficiency change (AEC) is identified. In parti-
productivity change. Infeasibility can occur when linear
cular, a cost Malmquist index, which is defined in terms of
programming (LP) techniques are used to compute and
cost rather than input distance functions, is developed and
decompose the index. Xue and Harker (2002) provide
computed using non-parametric LP models. The index is
necessary and sufficient conditions on the data to avoid LP
applicable when producers can be assumed to be cost
infeasibility.
minimizes and input-output quantity and input price data
Pastor and Lovell (2005) demonstrate that the source of
are available. But this index is not circular, LP infeasibility
all three above mentioned problems is the specification of
can occur, and its adjacent period components can provide
adjacent period technologies in the construction of the
different measures of productivity change.
index. Then Pastor and Lovell (2005) proposed a global
In this paper we propose a new cost Malmquist pro-
* Correspondence: G Tohidi, Department of Mathematics, Islamic Azad ductivity index and show that it is possible to specify a base
University, Central Tehran Branch, Tehran, Iran. cost boundary in a way that solves all three problems,
E-mail: ghatohidi@yahoo.com without having to impose restrictive conditions on either

This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 17:24:53 UTC
All use subject to http://about.jstor.org/terms
 G Tohidi et al - A global cost Malmquist productivity index using data envelopment analysis 73
the technologies or the data. That is, the proposed index is
circular, LP infeasibility cannot occur, and its adjacent
period components provides a single measure of produc-
tivity change. We call the index presented in this paper the
global cost Malmquist index.
The rest of the paper is organized as follows. In section 2,
we express technical background. In Section 3, the global
cost Malmquist index and its components are presented.
Section 4 presents the global cost Malmquist index when
the production technology exhibits VRS. In section 5, we
compute the proposed index and its components. Section 5
provides a numerical example. Section 6 concludes.
2. Technical background
Consider panel data of j = 1 , . . . , J producers and
t= 1, . . . , T time periods. In time period t , producers are
using inputs + , to produce outputs 1.. Define
now the production technology of period t , which is
= {(, ): X* can produce /}. The subscript V indicates
that the production technology satisfies constant returns
to scale (CRS).
Assume that Tlc is non-empty, closed, convex and
bounded. Alternatively, define the technology of produc-
tion in terms of the input distance function (Shephard,
1953, 1970) as:
= sup{0 : (x'/e,y') e , 0>O}, (1)
where the subscript i denotes input orientation. D'(y' x*)
characterizes the technology of production completely in
the sense that D'(y' x*)^' is sufficient for (x' y')e7*c and
'(, X ') = 1 if and only if (' ) lie on the frontier. -( ,
X*) is reciprocal to Farrell's input oriented measure of
technical efficiency (Farrell, 1957), which is
'(/,*') = min { : ( ', ') e T., > 0}. (2)
When input prices, w' e R"t , are available one may define
technology in terms of the cost function, which is
Ct(y'wt) = min{wV : (*',/) e rc',w'> 0}, (3)
where C*{y' wl) defines the minimum cost of producing a
given output vector y' given the input prices / and the
technology of period t. The set of input vectors x' which
correspond to the scalar C*(y' w ') lie on an isocost line that
defines a cost boundary as:
Iso *(y' w) = {x* : wtxt = '( , >')}- (4)
The input oriented measure of overall (or cost) efficiency
Farrell (1957) for (' ') under input prices is defined as
follows:
'( , x', w') = ' (5)
This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 17:24:53 UTC
All use subject to http://about.jstor.org/terms
If overall efficiency is less than one it will be either
because production is based on excessive input usage or
because it takes place at the wrong input mix in light of
input prices, or both. The first factor is captured by the
input technical efficiency measure in (2) and the second by
input oriented measure of allocative efficiency (AE), which
is defined as follows:
wtxt
= OE ;x ><(/,*<). (6)
The cost Malmquist (CM) productivity index of periods
t, t + 1 and their geometric mean are defined (Maniadakis
and Thanassoulis, 2004) as follows, respectively.
/*'+'/'(/+1)1
[ w'x'/C'iy', w') J' ( '
rM,+ 1 = k'+1x'+7C+V+1,w'+1)l l j
[ w'+'X'/C+H/.H-'+M J' l j
''+1/'(+|, w') w'+lx'+l/C'+1(y+1,w'+1)l
= w'x'/C'iy','1) X w'+l x' / C'+1(y' , w'+I) J '
(9)
where w'x' = Yln=i w'nx'n> n denotes nth input and '(,>')
is as defined in (3) with reference to the CRS technology.
Define now a global production technology (Pastor and
Lovell, 2005), which is:
7? = conv{7'c1U...U7'cr}.
The global Malmquist productivity index is defined on
(Pastor and Lovell, 2005) as:
(.0)
where the input distance function Df(x , y) = sup {6:(x/0, y)
elf,e>0}. Clearly, given the definition of the distance
function, when Mc> 1 we have a deterioration in pro-
ductivity between t and t+ 1. Productivity remains uncha-
nged if Mc= 1 and improve if Mc< 1-
3. A global cost Malmquist productivity index
In this section, we are going to compare a decision making
unit (DMU) with itself at different time points to measure
productivity changes by reference to a common cost fron-
tier when input-output quantity and input price data are
available. Towards this end, we define a common cost
frontier. It makes possible to define a new cost Malmquist
productivity index that is circular, there is no need to
resort to the geometric mean in the calculation and LP
74 Journal of the Operational Research Society Vol. 63, No. 1

-G G

infeasibility cannot occur when the production technology IsoC (y W )

exhibit VRS. For defining the common cost frontier, we ' IsoC'(/ w')

need a common input price vector that remains unchanged

in all time periods 1 , . . . , T. '' ' R x'
Any input price vector can be used for defining the
common cost frontier. We define the common input prices
^ r+1 f+1 ^
wG e R' , as wG = > YlJ=' h = y>0, where the
weights XjJ = 1 , . . . , T are decision-makers' preference ^ w ^ w x ' ' ^ : w x
over w' 7=1, . . . , . In fact, for A=l, j = 0(j^t,
7=1, ...,7) and ,+ 1 = 1, = 0(7#/+1,7=1,
we have wG = wt and wG = ' respectively. In the other
words, instead of wG we can use the price of any period
of time. Also the cost function CG(y'wG ) is defined as
follows:
'Vko^(3^,') Q z
CG(/, wG) = min{wGx : (x,) G TG}. () je,
Figure 1 The global cost productivity index under CRS
The set of activities (x, y)elf that correspond to the technology.
scalar CG(y, wG) lie on an isocost line that defines a new
cost boundary. This new cost boundary is defined as:

Iso G{y, wG) = {(x,y) : wGx = CG{y , wG)}. (12) 3.1. Decomposition of the proposed cost Malmquist
productivity index
Since the vector wG is constant, there is only one
The CMG index in (13) can be decomposed in a similar
benchmark cost boundary for all time periods 1 , . . . , T
manner as the CM index in (9) (Maniadakis and
that is defined in (12).
Thanassoulis, 2004).
Now we define the proposed Malmquist index on TG as:

CMG kG*'+1/CG(/+1,>vG)l mv 3.1.1. First stage decomposition of the CM index. The

[ wGx'/CG(y,wG) J- mv ^
The cost ratio wGx'/CG(y',wG) measures the extent to
which the observed cost in period t can be reduced while
still securing the output vector y* under the global
technology and input price vector wG. This ratio measures
the distance between the observed cost wGx' and the cost
boundary (,'). This distance will have a minimum
value of one. This (cost) distance is the reciprocal to
Farrell's measures of overall efficiency (Farrell, 1957). The
cost ratio wGxt + 1/(*+ l,wG) in (13) is defined in an
analogous manner. Clearly, a CMG index value less than
one implies productivity progress, a value greater than one
implies regress and a value of one indicates constant
productivity.
The global cost index is illustrated in Figure 1 for the
case where two inputs (xb x2), are used to produce one out-
put, y. RCQZ is the boundary of period t production
technology while IKLJ is that of period t + 1 . IKLQZ is
the boundary of global production technology. Production
takes place at point G:(x' ') in period t and shifts to point
B:(x+ ' yt+l) in period t+ 1. The output is standardized
to one unit. In terms of the distances in Figure 1, the CMG
index in (13) is (OB/OF)/(OG/OP). This demonstrate
graphically that the CMG index uses as benchmark the
global cost boundary (ie Iso G(y, wG)).
CMG index can be decomposed into OECG and CTCG, as
follows:
g _wt+lxt+/Ct+(yt+'wt+l) wGxt+ 1 / CG (yt+ 1 , wG )
wtxt/Ct(yt, w') wt+lxt+l /Ct+l(yt+l,wt+l)
x w x / ( >w) _ qecg x CTCG 14)
where the component outside the brackets (OECG) is the
overall efficiency change component in the decomposition
of CM index (Maniadakis and Thanassoulis, 2004) and
the term inside the brackets (CTCG) provides a new
measure of cost technical change. The first ratio of CTCG
measures the distance between the new cost boundary and
the cost boundary in period t+ 1 along ray (xt+l9 yt+x).
Similarly, the second ratio measures the distance between
the new cost boundary and the cost boundary in period
t along ray (x' '). In terms of the distances in Figure 1,
OECg = OEC = (OB/OE)/(OG/OM) and CTCG is:
CXCG = OB/OF X OG/OM = OE/OF
OB/OE X OG/OP = OM/OP
The value of CTCG may be either greater or less than or
equal to one. For example, in Figure 1, CTCG< 1 indicates

This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 17:24:53 UTC
All use subject to http://about.jstor.org/terms
 G Tohidi et al- A global cost Malmquist productivity index using data envelopment analysis 75
the cost boundary in period t along ray (x' y') is farther
away from the new cost boundary than is the cost
boundary in period /+ 1 along ray (xt+ *, y* + 1).
3.1.2. Second stage decomposition of the CM index. The
terms obtained in the first stage decomposition of the
CMG index can themselves be decomposed as follows.
The decomposition of OECG: The OECG component in
(14) can be decomposed into TECG and AECG terms as
follows:
o *'+1)
D'(y >',x')
wl+ 'xt+' j 1 (+l ) w'+l ))<+' (yt+ 1 ) +1 ))
x w'x'/iC'iy', wl)D'i(y', x'))
= TECgxAECg, (15)
where TECG and AECG are technical efficiency change
and allocative efficiency change components of CM index
(Maniadakis and Thanassoulis, 2004). In terms of the
distances in Figure 1, the two terms are:
TECg= TEC = OB/QD AECg= AEC = QD/QE
OG/OA' OA/OM
The decomposition of CTCG: The CTCG term in (14) can
be further decomposed as follows:
CTCG = 'D?(y+l,*+l) D'{,x'y
'.D'+l (y+1 , x'+l ) Df(yl,x')_
wGx'+1/ (CG(/+I , wG)Df(y,+l , jc'+1 ))
w'+'x'+i/ (C'+1 (j>'+1 , wt+x )D'+X (y+1 , xt+l ))
wlxl/(C4yl,w')D'{y',x')) '
X w0x'/(CG(yt,wG)Df(yl,xt))
- TCG x PE. (16)
Each one of the cost ratios of the two indices in the
brackets in (14) are separated within (16) into a technical
part and a residual price effect (PE) part. The first term in
the right-hand side of (16) is the technical change (TCG)
component of the global Malmquist (Mc) index, intro-
duced by Pastor and Lovell (2005). In terms of the
distances in Figure 1, this term is:
_ OB/OD OG/OA
OB/OD X OG/ON
The terms in the second brackets (PE) in (16) capture
the residual impact of relative input price changes on the
shift of the cost boundary. In terms of the distances in
This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 17:24:53 UTC
All use subject to http://about.jstor.org/terms
Figure 1, this term is:
G = OD/OF OA/OM
OD/OE ON/OP
The numerical values of the components resulting from
the decomposition of the CMG index are interpreted in the
same manner as the index itself. A value below one indi-
cates progress, greater than one regress and one indicates
that performance stayed constant.
4. Another decomposition of the global cost Malmquist
index
When the production technology exhibits VRS, Portela
and Thanassoulis (2008) decompose the meta-efficiency
VRS score of unit j observed at time t as follows:
nm(VRS) pm(CRS)
nm(CRS) _ - fT(VRS) A ^
ujt - ujt A nT(VRS) nm(VRS)
Ujt Ujt
= eJt{VRS) x TGV,-, x ME,-,.
That is, meta-efficiency decomposes into within-period-
efficiency in relation to a VRS frontier 0]}VRS)' technolo-
gical gap between the VRS frontier in t and the VRS
meta-frontiers ff^VRS)ie]}VRS), labelled TGV, and meta-
scale-efficiency Q"l{ CRS)0^ VRS) , labelled MSE. Note, the
MSE captures the distance between the CRS and VRS
meta-frontiers at the input-output mix of unit j as observed
in period t. Thus, the circular meta-Mulmquist index,
defined as Ml/^i^ = e^xRS)ie^CRS) can be decomposed
as follows (Portela and Thanassoulis, 2008):
ufCRS) eltl(VRS) TGV/f-n MSEy+1
v+ 1 ej(VRS) jev,-, MSE,-, '
That is, the index can be decomposed into pure technical
efficiency change 6jt' '( VRS)/ Qf vrs^ frontier shift between
VRS frontiers TGV^+i/TGV^ and meta-scale-efficiency
change MSE7,+ 1/MSE7,.
According to the above discussion, we decompose the
cost ratio (wGxt)/(CG(CRS'yti wG)) as follows:
wGx' wGx'
CG(CRS)(y,iWG ) = C'(rRS)(y',w')
C'(VRSHy',wl)
X CG(VRS)(ytiWG)
CG{ VRS)(y' wG)
X C^^ly^w0)'
where, the cost ratio wGxt/Ct(^VRS)(yt, wl) measures the
distance between the observed cost wGxt and the cost
boundary '(5)(,/) under VRS technology of period t,
technological gap between the cost boundary in t under
VRS technology of period t and the global cost boundary
76 Journal of the Operational Research Society Vol. 63, No. 1

under VRS global technology along ray (x' /), labelled The term (',') can be computed using the following
TGVg, and global-cost-scale-efficiency CG^VRS>i{y' wG)/ model:
N
CG(CRS)(yt wG labelled gCSE,. Note that GCSE, mea-
sure the distance between global cost boundary under CRS CG(y', wG) = min Y wknx >
= 1
global technology and global cost boundary under VRS R

global technology along ray (x' /). st tyfi* >'>

Now the global cost Malmquist index, defined as 7=1
R
CMG(CAS) = (,( JG )} can be decomposed as
follows: Y^jXjn^Xn, n=l,...,N,
7=1

*j> 0, =1,...,,
CMG(CRS) = WGX,+ l/C,+ ^VR^{y,+ 'wt+l) x5=0, =1,...,, (18)
wGx' /C'(VRSi(y', w')
where R=T x J is the number of observed DMUs
Ct+l(VRS)(yt+lWt+iyCG(VRS)(yl+lWG}
in TG that are denoted by (xj, yj), 0=1, . . . ,R). The
x C'<ras) (/, W>)/CG<VRS) (y<, wG)
terms Ct+'y'+' wt+x) and (y'+l, wG) can be
CG( VRS> (/+' 5 wG) jCG{CRS) {y,+l , wG) computed using models (17) and (18), respectively, after
x CG(VRS)(y>, wG)/CG<CRS)(y', wG) changing round the time periods t and t + 1.
We can compute the term D'(y' x) using the following
OEC ::TGVg'+i ::GCSE'+' model (Fre et al , 1989):
TGVC GCSE, '
''(* ,*)'~ = min
j
where, OEC is the overall efficiency change, cost boundary
shift between the cost boundary in t under VRS techno- st- E ^ >" m = 1 ' ' M'
7=1
logy of period t and the cost boundary in t + 1 under VRS
J
technology of period t+ 1, TGVg,+ /TGV^ and global-
cost-scale-efficiency change GCSE,+ 1/GCSE,. Y,*jXJn<exL n=l,...,N,
7=1

Xj> 0, 7=1,...,/. (19)

5. Computation of the index and its components
Similarly, the term Df(y' xf) can be computed by the
Data envelopment analysis (DEA) can be used to compute following model:
the CMG index. Assume that in time period t , the kth unit
employs amounts * of input n(n = 1, . . . , TV), available at
[DG{y' )] _1 = min 9
prices Wknin =1, . . . , AO, to provide amount of output
R
m(m =1, . . . , M). For unit the cost denoted wGxt in (13)
s.t. Y V > y'km . m=l,...,M,
is wGxf = J2n= i wknxkn Similarly the costs denoted J= i

wGxt+l , wV and u>+1x'+1 are J2n=' wkn^kn> R

Hn=' ' and EiLi ''> respectively. Also, the Y^Jxjn^04n, n=

7=1
term '(, w') can be computed using a model such as
the following (Maniadakis and Thanassoulis, 2004): 0, j= (20)

N
The terms D'+ ' (y ~ '+ ') and Df(y'+ ' x'+ ') can be
computed using models (19) and (20), respectively, after
C'(y', w') = min ^2 w'knxn . changing round the periods t and /+ 1. To compute when
=1

J the production technology exhibits VRS all that is needed

S.t. > . m = 1 , . . . , M, is to add to DEA models the constraint J2j^j = 1-
7=1
J

Y! <, n = i,...,N, 6. Numerical example

7= 1
Maniadakis and Thanassoulis (2004) designed a simple
Xj> 0, j = 1 , numerical example to illustrate the use of the CM index.
x>0, n=l,...,N. (17) Now let us use it to compare the two indexes CM and

This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 17:24:53 UTC
All use subject to http://about.jstor.org/terms
 G Tohidi et al- h global cost Malmquist productivity index using data envelopment analysis 77

Table 1 Numerical example data for period zero Table 2 CMG and CM index and component values for unit E

Unit Input 1 Input 2 Output Input 1 Input 2 CM TEC AEC CM TEC AEC PE?
level level level price price
0.418 0.667 0.857 0.7 1.046 0.795 0.667 0.857 0.7 1.986
A 3 5 1 3 3
The bold values emphasize the difference between CM and CM values.
3 3 1 3 3
5 2 1 3 3
D 8 2 1 3 3
E 7.5 3.0 1 3 3 Here, the value under one for the TECG component of
CMG (TEC component of CM) reflects the improvement
in technical efficiency of unit E between period zero and
one. The value of TCG (and TC) below one reflects
technical progress and AECG( = 0.857) is suggesting unit E
CMG. Table 1 shows DMUs A-E. Each of five DMUs uses changed input mix to make better the time period one
two inputs to produce a single output where the output input prices. The PEG component of CMG (PEG = 1 .986)
value is unitized to one for each DMU, assuming CRS. and PE component of CM (PE= 1.0446) suggest that the
Also, it is assumed that all DMUs face the same input effect of input price changes between time periods zero and
prices. Assume that in period one all units except E reduce one was detrimental to productivity. The values under
their input levels by 30%, unit E moves to unit and the one for both CMG and CM indexes are indicating a pro-
price of input one is reduced by 33.33%, (Maniadakis and ductivity growth between zero and one. Nevertheless, the
Thanassoulis, 2004). CM index indicates higher productivity growth than the
The components of the CM index are computed in CMG index because the CMG index (0.795) captures a
Maniadakis and Thanassoulis (2004) in respect of unit E. large regress in the PEG component (1.986) and the value
We use the values of TEC and AEC components of CM of the PE component of CM index is smaller than the PEG
instead of computing the TECG and AECG components of component of the CMG.
CMG as follows:

TECg = TEC = 0.667, AECg = AEC = 0.857. 7. Conclusions

We assume wG = ' wl + ' w2 and use the data in Table 1
to compute the TCG and PEG components of CMG index.
By solving the presented LP problems in section 5 and
using wGxl = 11.5501, wGx = 27.7537, wV = 31.5 and
/+ + 1 = 10.5002, we have
" ?1.*1) I ~ 1 v 1.4999
" LAV.*1) >?(/, *) J ~ 1 v 2-1427
and
The global cost Malmquist productivity index developed in
the current study is applicable when producers are cost
minimizers and input-output quantity and input price data
are available. The index developed here is an extension of
the global Malmquist productivity index introduced by
Pastor and Lovell (2005). Its decomposition makes it
possible to identify the root sources of productivity change.
This index and each of its components is circular, it
provides the single measures of productivity change and its
components and the LP techniques that are used to
compute the index are feasible.

ppG ^'/(^('.^'.1))
vv'x'AC'O^v^AV,*1))
Acknowledgements - We are grateful to anonymous referees for their
W0JC0/(C(/,W0)Z)?(/,X0)) constructive comments that improved this paper significantly.
x wGx/(CG(j, wG)Z)f(/, Xo))
11.5501/(11.5501 x 1)
~ 10.5002/(10.5002 x 1) References

31.5/(18.0000x 1.4999)
Berg SA, Forsund FR and Jansen ES (1992). Malmquist indices of
27.7537/(11.5501 x2.1427) ' ' productivity growth during the deregulation of Norwegian
banking, 1980-89. Scand J Econ 94: S211-S228.
The CMG index of unit E can be computed as: Caves DW, Christensen LR and Diewert WE (1982). The economic
theory of index numbers and the measurement of input, output
CMG = TEC x AEC x TCG x PEg = 0.795. and productivity. Econometrica 50: 1393-1414.
Fre R and Grosskopf S (1992). Malmquist productivity indexes
and Fisher ideal indexes. Econ J 102: 158-160.
The results about unit E have been summarized in
Fre R and Grosskopf S (1996). Intertemporal Production Frontiers:
Table 2. With Dynamic DEA. Kluwer Academic Publishers: Boston.

This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 17:24:53 UTC
All use subject to http://about.jstor.org/terms
78 Journal of the Operational Research Society Vol. 63, No. 1

Fre R, Grosskopf S, Lindgren and Roos P (1989). Productivity
developments in Swedish hospitals: A Malmquist output index app-
roach. Discussion Paper no. 89-3, Southern Illinois University,
Illinois, USA.
Fre R, Grosskopf S and Russell RR (1998). Index Numbers:
Essays in Honour of Sten Malmquist . Kluwer Academic Publish-
ers: Boston.
Farrell MG (1957). The measurement of productive efficiency. J R
Stat Soc Ser A-G 120: 253-281.
Maniadakis N and Thanassoulis E (2004). A cost Malmquist
productivity index. Eur J Opnl Res 154: 396-409.
Pastor JT and Lovell CAK (2005). A global Malmquist producti-
vity index. Econ Lett 88: 266-271.
Portela MCAS and Thanassoulis E (2008). A circular Malmquist -
type index for measuring productivity. Aston University, UK
Working Paper RP08-02.
Shephard RW (1953). Cost and Productions. Princeton University
Press, Princeton: USA.
Shephard RW (1970). Theory of Cost and Production Functions.
Princeton University Press, Princeton: USA.
Xue M and Harker PT (2002). Note: ranking DMUs with infeasible
super-efficiency DEA models. Manage Sci 48: 705-710.
Received February 2009;
accepted December 2010 after two revisions

This content downloaded from 197.255.68.201 on Wed, 06 Jul 2016 17:24:53 UTC
All use subject to http://about.jstor.org/terms