Live large or die young: subsidized loans and firm survival in Brazil

For a review of the financial system in Brazil, the role of BNDES within it and the financial structure of the BNDES itself, see De Mello and Garcia (2012). Regarding the pronounced and highly persistent regional inequality in Brazil, see Ehrl and Monasterio (2019), for example.

The BNDES’ balance sheet relative to the Brazilian GDP has tripled between 2000 and 2015, reaching a new record height equivalent to 15% of GDP.

I chose to analyze only the effects of automatic indirect and non-automatic (direct and indirect) loans. As explained in greater detail in Sect. 3.1, the term “direct” refers to whether the loans are awarded directly through the BNDES or indirectly through an accredited financial intermediary, for example, a commercial bank. These lines are not the only ones offered by BNDES, but those with the largest number of recipients and with the largest average loan volume. I refrain from considering other lines of financing also because they are target at very a specific group of companies (e.g., micro-enterprises) or because the funding is not linked to the purchase of products or inputs (e.g., in the case of exporter aid). Moreover, data access to some of these other credit lines is not public.

According to the parameter estimate of the base hazard p, the exit probability has an increasing, decreasing or constant functional form. A disadvantage of the generalized Gamma distribution is that it cannot derive closed solutions for the risk rate (Kaniovski and Peneder 2008). Their coefficients only indicate whether a given variable has a positive or negative impact on firm survival. According to Manjón-Antolín and Arauzo-Carod (2008), the nonparametric Cox estimation avoids biased \(\beta \) coefficients caused by the imposition of an erroneous distribution assumption. However, if it is possible to approximate the true distribution well with one of the parametric choices, the semi-parametric estimation will be less efficient and therefore not automatically the best choice. Economic theory does not offer much guidance regarding the most appropriate choice for the stochastic error distribution and duration. Following previous works in this line of research, I will compare the distributions cited above.

Note that, either way, informal businesses are not of interest in the present case because they must not to apply for BNDES loans and hence should not be compared with other companies that have obtained BNDES loans.

The difference between the two concepts is obviously that an establishment is the smallest unit of business activity, concentrated in a single address. On the other hand, the firm / company can, but does not need be, a conglomerate of establishments.

Note that information from political donations could not be used in the current context because only a very small fraction of firms appear as donors. Thus in an analysis with the population of firms, the significance of these variables vanishes.

Note that Fig. 2 is not equivalent to the usual representation of survival according to Kaplan and Meier (1958). Here, time in the market as measured by the horizontal axis refers to the years after entry in the sample, i.e., the moment it starts being at risk instead of using the number of years a firm is active in the market. Therefore, the maximum time on the horizontal axis is equal to 12 years.

Note that in the survival analysis, firms are followed over the whole observation period, and not only in the year the firm receives a loan.

These firm-specific determinants still turn out to be significant for firm survival with the expected sign; however, their quantitative effect in the combined model is lower than in the simple duration model, see Table 6.

Unfortunately, the effect of the of ever receiving a BNDES loan cannot be identified in the combined selection-duration model because this variable is equal to the selection indicator \(c_i\) that is required to separate the different parts of the likelihood function (8).