y is always a growth rate, this is, in effect, a triple difference–we can allow small firms and big firms to have different rates of growth, and the rate of growth to differ from year to year, but we assume that there would have been no differential changes in the rate of growth of small and large firms in 1998, absent the change in the priority sector regulation.

Using, respectively, the log of the credit limit and the log of next year’s sales (or profit) in place of y in equation 1, we obtain the first stage and the reduced form of a regression of sales on credit, using the interaction BIG ∗ P OST as an instrument for credit. We will present the corresponding instrumental variable regressions.

Results: The change in the regulation certainly had an impact on who got priority sector credit. The credit limit granted to firms below Rs. 6.5 million in plant and machinery (hence- forth, small firms) grew by 11.1 percent during 1997, while that granted to firms between Rs.6.5 million and Rs. 30 million (henceforth, big firms), grew by 5.4 percent. In 1998, after the change in rules, small firms had 7.6 percent growth while the big firms had 11.3 percent growth. In 1999, both big and small firms had about the same growth, suggesting they had reached the new status quo.

[TABLE 1 ABOUT HERE] This is confirmed when we estimate equation 1 using bank credit as the outcome. The result is presented in column (2) of Table 1 for the entire sample of firms. The coefficient of the interaction term BIG ∗ P OST is 0.95, with a standard error of 0.033. Column (1) estimates the probability that a firms credit limit was changed: the coefficient on BIG ∗ P OST is close to zero and insignificant, suggesting that the reform did not affect which firms limits were changed. This corresponds to the general observations that whether or not a firm’s file is brought out for a change in limit has nothing to do with the needs of the firm, but respond to internal dynamics of the bank. We will make use of this fact to partition the sample into two groups on the basis of whether there was a change in the credit limit or not: the sample where there was no change in limit can be used as a “placebo” group, where we can test our identification assumption. Finally, column (3) gives the estimated impact of the reform on loan size for firms whose limit was changed: the coefficient of the interaction BIG ∗ P OST is 0.27, with a standard error of 0.10.

This increase in credit was not accompanied by a change in the rate of interest (column (4)).

9