Y1 = Y2
H 0: <λ = λ.
μ1 μ2
The third step forward considers the formation of the groups as endogenous. The idea is that firms
introduce innovations because they are more productive, young or intensive at investing into R&D
activities or innovative capital, or maybe because they have more cash flows. The age profile of the
board members and/or the share of temporary workers might be correlated to the innovativeness of
the firms as well.
The new specification for labor productivity growth can be thought of as a standard case of
switching regression model with endogenous switching (as explained firstly in Maddala 1983). We
want to consistently estimate the parameters in two regimes: whether firms are innovative (regime j
= 1) or non innovative (regime j = 2) over the period of observation. The new model specification is
the following.
(ТетрогагУ
-----------I + ε1it if innovative
L √ it-2
(6)
(Temporary . ð
I + ε2it if noninnovativ
L √ it-2
The marginal distribution of the error terms εjit j=1,2 can be assumed normal with zero mean and
constant variance σj2. We shall modify this strong assumption in the robustness estimations. The
conditional distribution of the error terms are instead different from zero, according to:
Ε(ε1it | innovative) ≠ 0
(6’)
Ε(ε2it | non innovative) ≠ 0
This is because a criterion function determines whether a firm belongs to regime 1 or 2, as in
equation (7):6
'D1 = 1 if δZit + ωit > 0
1
(7)
D1 = 0 otherwise
6 D1 = 0 ⇔ D2 = 1 , meaning that, if a firm has not introduced an innovation in 2001-2003, it is non innovative by
definition.
11