The impact of public guarantees on credit to SMEs

The average number of employees per business in industry and services was less than 4 in Italy, versus almost 8 in the EU15 in the year 2001 [Istat, 2004]. In 2003, the share of small firms was 99.4%, the same percentage as in 1996. These firms provided 69.5% of total employment, against 71% in 1996.

For a survey of the issues, see in particular Bosworth et al. (1987), Gudger (1998)

This was established in 1996 with the generic mandate of providing guarantees to banks and financial institutions, against their loans to SMEs, as well as against their minority equity participations in small and medium-size companies, and to MGIs, against their guarantees for SMEs’s borrowing. Hence, the SGS offers direct guarantees to lending banks, co-guarantees together with other guarantor institutions, and guarantees of last resort to MGIs

This approach was changed in 2006 to ensure full reimbursement within a short time since insolvency, as requested by the new Basel 2 criteria for bank capitalisation

This shortcoming was corrected only in part in 2005, by setting up a specialized section devoted to the ICT sector

The default loss ratio can be decomposed as the product of the default loan rate, the repayment rate and the reciprocal of the guarantee coverage rate. These ratios are presented in Table 3.

The default rate for banks’ loans to micro enterprises is 9.82%, Banca d’Italia (2005).

This is the ratio of the deficit to the amount of guaranteed loans, and is equal to the product of the deficit ratio by the guarantee coverage ratio (Table 3).

These data originate from the Fund’s books. Information on SMEs’ financial statements was drawn from AIDA balance-sheet data bank. From the latter, a random sample of 11261 SMEs was drawn, including firms that were eligible for the Fund’s guarantee (3952) but did not apply for it, and firms that were not eligible (6066), because of the EU exclusion of some economic sectors from the guarantee. A totoal of 1,243 of sampled firms received the Fund’s guarantee. The sample period is 1999–2004. Financial data comprise financial costs, earnings, net worth, fixed and intangible assets, long/short term bank-related debt, long/short term bonds, long/short term non-bank-related financial debt, sales, number of employees, depreciation allowance, total assets.

Cross section estimates were made but are not reported in the text. They are made available upon request to the interested reader. In these estimates, the δ coefficient linearly increases over time, signalling the possibility that the estimates are affected by temporal variation. This distortion might be due to changing macroeconomic conditions, such as a decrease in official interest rates, or factors that allow firms to systematically save on financial costs over time. For instance, this result might come from improvements in financial management attributable to technological advances. To account for this possibility the authors resort to the DID estimation procedure.

According to Bound et al (1995), instrumental variables estimates may be biased in small samples. A correct practice is to report a statistic that measures this possible bias. Staiger and Stock (1997) prove that when the instrumented variables are no more than 1, the reciprocal of the F-test of the exclusion of the instruments in the first stage approximates the fraction of the OLS bias that is still present in the IV approach to LATE in a finite sample. For an example of the inclusion of the F of the first stage in a LATE estimate, see Ichino and Winter-Ebner (2004).

Proofs are available from the authors upon request.

This figure is obtained by taking the antilog (to base e) of the estimated dummy coefficient, subtracting 1 and multiplying the difference by 100. Since the dependent variable is in log, once taken the antilog one must refer to the median value of the dependent variable, not to the mean value. For more details see p. 321 of Gujarati (2004) and cited references.

On the basis of a Hausman test, we can reject the null hypothesis of consistency of both fixed and random effects.

See previous footnote.

Estimates have also been carried out by replacing total assets with the share of physical over total assets. The results show no appreciable difference from the above estimates