the fact that only one macroeconomic source of shocks is common to heterogeneous agents is rejected by
very extended empirical analyses. In addition, they show that many statistical properties (like, for
instance, Granger causality) existing at the level of individual agents do not survive aggregation, while
aggregated time series show statistical properties that are absent in the individual data that constitute
them. All this suggests that “mesoeconomic” empirical analyses can be as informative as the more
conventional microeconometric and aggregate time series analyses, by providing a different perspective to
investigate the interaction between macroeconomic policy variables and heterogeneous typologies of
agents.
3. A preliminary descriptive analysis
The graphics reported in the next sub-section and the descriptive statistics of the appendix allow
already to identify some relevant structural phenomena. In the dataset, lenders are divided in three size
classes: large and major banks, average size banks and small and minor banks. Loans are divided into 5
size classes. The smaller categories have been aggregated into the class “C1”, which includes all loans
smaller than 250 millions lira; class “C3” includes loans from 250 to 500 millions lira; class “C4” from
500 millions to 1 billion liras; class “C5” all loans larger than one billion liras. In most cases, the
comparisons will be made among the three larges class sizes because (as explained in the appendix) the
dataset contains many discontinuities in the statistical criteria of sampling and in the definitions of some
relevant variables. Nevertheless, the comparison among the three largest size class is rather informative.
3.1 How heterogeneous are lending rates?
The first structural feature that will be considered here is the different behaviour of lending rates
according to the different kinds of size classes of borrowers and geographic areas, for each given size
class of borrowers. In this case, in order to make the graphics easier to read, we have compared the data
referred to the first observation of each time series, the median observation and the final observation, by
using histograms. This kind of graphical representation (usually not employed for data and observations
of this kind, but rather to represent frequencies), although unusual, allows seeing more directly the
structural differences analysed here. No precise or uniform relation can be detected between the lender
size (for a given class size of loan) in the various areas, by simply observing the data and without
recourse to multivariate analysis, given the potential effects of borrower risk and the demand
expectations.
As one can see from the graphics, a higher interest rate from large and major banks is only
detectable in “area 3” (corresponding to Tuscany, Marche and Umbria) for all the size classes of loans,
while for the loan classes “C3” and “C4” is also detectable in “area 1” (Piedmont, Val d’Aosta and