As seen in Table the majority

As seen in Table 1, the majority of firms in our sample are not listed on the stock market as well as they are SME. Firms in our sample have around 416 employees where their exports represent only 6% of their revenues. Overall, figures represent what is expected: Firms listed on the stock market are larger expressed by the number of employees. Regarding investment over capital, Brazilian firms in the manufacturing sector invest around a quarter of its capital every year. Moreover, there is no large difference between them, even when considering among the three categories defined above. What is striking is that firms not listed on the Brazilian stock market, SME and low exporters generate on average cash flow around their capital stock. Moreover, large firms generate only 63% cash flow compared to its capital, public-listed firms generate 44% and high exporters, 75%.
Empirical results Our results from specification (1) are presented in Table 2. Using data from 2008 to 2010, three approaches are explored: pooled ordinary least square (OLS); within groups (WG); and SYS-GMM. In the case of GMM approach, instruments are available for 2008 in the case of equation in first difference and 2009 and 2008 for level equation. As said before two variables are applied as covariates in order to capture the sectorial investment opportunities effect: the industry-level of value added growth and the industry-level of investment growth . At the firm level, we impose the firms’ annual sales growth variable in order to control also for investment opportunities. Time dummies and industry dummies interacted with time dummies were included in all the specification. All estimated models have evidenced that firms are credit constrained even after controlling by industry-level variables. However, cash flow coefficients estimated by GMM have a superior impact when compared to other methods. This result may indicate that within groups estimates may still suffer from endogeneity bias. Notice that Sargan tests reveal that lagged explanatory variables are valid instruments for both system-GMM regressions given that we do not reject the null purchase Go 6983 that overidentifying restrictions are valid. As discussed in Section 3, /Kit-2 may also be an additional explanatory variable for explaining firms’ investment. It is possible that previous cash flow may impact directly actual and future firms’ investment decision. At first moment, we did not include lag values of cash flow in Eq. (1) because our database has an insufficient number of years (T=3) for satisfying system GMM moment restriction. However, as an alternative for the absence of cash flow lagged values as instruments for the additional explanatory variable /Kit-2, we include a moment condition in the level equation based on both lagged sectorial variables and firms’ variables. In the system GMM regression, the coefficient of Cash flow in t−1 is not significant at conventional levels and it does not interfere in the results of current cash flow. The choice of instruments is based on the Sargan test. Considering all these aspects, model 6 reveals an estimated coefficient equal to 0.25. Evaluated at the sample mean, this indicates an elasticity of the cash flow to capital ratio correspondent to 0.98. Indeed, this is a striking result: the impact of cash flow on investment is practically equivalent to the unity; in other words, for every increase in cash flow, investments raise at same magnitude. For instance, Carpenter and Guariglia (2008) findings suggest an elasticity of 0.16 for the UK. In other words, credit constraints in Brazil during the investigated period were more than 6 times what was observed in a developed country. These findings suggest that firms in developing countries are indeed more credit constrained than those in the developed world. However, the impact of cash flow and investment opportunities variables on investment may be different among firms given the degree of credit constraint. In this sense we evaluate the impact of cash flow on investment by classifying firms according to categories that may properly proxy for the degree of credit constraint. The first category considered is the firms’ size, as shown in the first part of Table 3. Due to the fact that pooled OLS and within group coefficient may suffer from endogeneity problem, all models from now on are estimated by system GMM.