variables (vector Z), and their values are taken two years before the occurrence of
the debt rescheduling as well. Notice that, in order to input a value of the control
variables, when either I or C is zero, we calculated the average year of both events
(adoption and concession) in our sample and took the control variables’ values two
years before that year.
4.1. The identi...cation problem
This two simultaneous equations model contains both a reduced form, equation
(4.1), and a structural form, equation (4.2). Notice that while there would be
no problem with the estimation of (4.1) as a univariate probit, we would not
model the impact of the adoption on the rescheduling estimating just a single
probit equation for the probability of the debt rescheduling and adding a dummy
(equal one in case of IMF adoption) to the regressors since this dummy would
be endogenous. More formally, it would be correlated with the error term of the
probit equation (4.2). Thus, unobserved factors in≠uencing both IMF adoption
and debt rescheduling would be interpreted as part of the “IMF adoption” e∏ect.
The structural form is identi...ed if at least one variable in X is not included in
Z:15 This identi.cation problem was not an easy question to solve. To identify the
parameters of the model we use both the dummy “previous Fund arrangement”
(BEF) and the rate of change of the “general government consumption” (GGC).
Our assumption here is that, conditional on the programme adoption, these two
variables do not a∏ect the probability of obtaining a debt rescheduling.
This choice is justi.ed on an economic ground. Regarding the dummy BEF,
it is plausible that countries that have had Fund arrangements in the past will
be more likely than others to enter into an arrangements in the future, because
both the authorities of that countries are already familiar with the Fund operating
procedure, and they have al ready gained a sort of “reputation” with the Fund.
15See Maddala (1983), p.122.
19