for each quarter) and the discount rate in order to verify to what extent appropriate proxies for the
category-specific risk, for the relative market power of lenders and borrowers, size dummies for lenders
and loans, geographic dummies can explain heterogeneity in these categories of lenders. Most
contributions in the literature briefly described in section 2 are based on a priori specifications or
sometimes apply the “general-to-specific” methodology in a simplistic way by determining the lag
structure of the model only on the basis of the t-statistics and without performing appropriate joint test for
linear restrictions on a general “unrestricted” model. In this paper, we attempt to apply in a rigorous way
Hendry (1985, 1988) and Harvey (1989) methodology: the final “parsimonious” specifications are strictly
determined by applying joint “variable deletion tests” in a “general unrestricted” model with four time
lags: The variables appearing in the estimates with lower levels of significance have not been eliminated
because the zero-restrictions on their coefficient have been rejected in preliminary general “variable
deletion tests”. This does not strictly apply to the dummies, which have been tested jointly for the sake of
our comments, in order to confront their values and level of significance. The appendix contains therefore
the “general unrestricted” specifications of the models and the “variable deletion tests” performed to
obtain the final specifications. Of course, by assessing the level of significance of the various regressors
one has to consider that we are dealing with very volatile (and heterogeneous) financial variables. Also for
this reason all the tests have been performed in the White “robust to heteroskedasticity” version (White,
1980, 1984). In a sense, the results of sub-section might be considered an extension of the descriptive
analysis of section 3, since the point is not to identify a precise behaviour equation, but rather to measure
and compare the impact of some theoretical and structural variables on the spread between lending rate
and monetary policy rate.
The second equation estimated is meant to detect the possible existence of a hierarchy in lending
behaviour and again is based on a very simple consideration. While it is certainly not possible to associate
the loan size class to the size of the borrower, the largest loan size class is certainly more affected than the
other two by the behaviour of large and powerful companies with higher market power. Therefore it
seems reasonable to ask ourselves whether the share (and not the absolute amount) of loans granted to the
largest class is demand determined. If so, for a given level of loans supplied by the banking system the
share of loans to the whole of the other size classes is necessarily residually determined and cannot be
demand-determined. For the sake of our estimates, we seek to verify whether the share of loans to the
largest size class is negatively correlated to its price, positively correlated to the bad debts (since
borrowers have obviously incentives to renegotiate them, or, at least, to seek an accommodative behaviour
from the bank) and positively correlated to the demand expectations. This last point, while being
consistent with a “demand-determined” lending behaviour, is not consistent with the phenomenon of
“flight-to-quality”.
20