based on simultaneous proportional modifications of inputs and outputs; it generalizes
Debreu’s and Farrell’s measure and is equally straightforward to interpret.
To estimate the proportional distance function, we use a non-parametric approach (see
Banker and Maindiratta, 1988; Varian, 1984). The technology can be written as:
T = ](xt,yt),xt ≥∑θjxjj,yt ≤∑θjyjt,∑θj = 1,θj ≥ 0, j = ι,∙
-,J!>. (3)
The linear program that calculates the values of the directional distance function is given by7:
Dt(xt, yt) = maxδt
s.t. xt -δtxt ≥ ∑θj xtj , (4)
j
yt +δtyt ≤ ∑θj ytj ,
j
∑θj =1, j=1KJ.
j
Suppose that an individual bank is represented by a production vector(xt, yt) with
corresponding technologyTt, and then the production vector is changed to(xt+1, yt+1) with
corresponding technology Tt+1 . In order to assign a cardinal measure to the productivity
change we can use the directional distance function in one of two ways; corresponding to
using either the initial technology at t or the final technology at t+1 as reference. In this case,
the Luenberger productivity indicator proposed by Chambers (1996) can be employed to
evaluate productivity change. The productivity indicator is constructed as the arithmetic mean
of the productivity change measured by the technology at Tt+1 and the productivity change
measured by the technology at Tt .
The Luenberger productivity indicator is defined as8:
L(zt, zt+ι ) = 2 [Dt+ι(zt ; g)- D+ι (zt+ι ; g)+Dt (zt ; g)- Dt (zt+ɪ ; g)]. (5)
7 All the computations are programmed in Mathematica language with the mathematica 5.0 software.
8 We simplify the notations by posing zt = (xt, yt) .