w h e r e I S p r e a d t > 0

( r e s p . I S p r e a d t < 0

) are indicator variables for whether the spread is positive

(resp. negative), θ_{i }is a state fixed effect, and δ_{t }is a year fixed effect. Avgrowth_{bit }is the smoothed growth rate for the state in which the headquarters of bank b are located. Standard errors are adjusted for serial correlation.

The regression controls for state and year fixed effects. While we see that the C/D is higher in states with more favorable growth rates, we are most interested in the coefficients γ^{− }and γ^{+}, which measure how banks in different growth environments differentially react to changes in the spread between the commercial lending rate. Because a negative spread occurs only twice, and is a quite particular situation (in a perfectly flexible market, banks facing a negative spread should eliminate all credit from their portfolios), we allow a separate coefficient

on (Spread_{t }∗ avgrowth_{bit}) when the spread is negative. The negative and marginally statistically significant coefficient on γ^{+ }

suggests that banks

in high-growth environments substitute towards government securities (away from loans) less when the spread falls. We interpret this to mean that banks in low growth states are more sensitive to government interest rates: because they face less attractive projects to finance, they are more likely to park money in government securities when government securities become more attractive. However, since the number of states in which a bank is headquartered is relatively low, we have relatively low power once we account for serial correlation at the state level.

To achieve more precise estimates, we estimate the same equation, except that instead of measuring growth only in the states in which commercial banks are headquartered, we use the synthetic index described above, which takes into account all the states in which the bank is active. Columns (3) and (4) present results from:

ln(CD_{bit})

= α + β ∗ bkgrowth_{bit }

+ γ + ( S p r e a d t ∗ b k g r o w t h b i t ) ∗ I S p r e a d t > 0

(9)

+ γ − ( S p r e a d t ∗ b k g r o w t h b i t ) ∗ I S p r e a d t < 0

+

θ

_{i }+ ψ_{b }+ δ_{t }+ ε

bit

where bkgrowth_{bit }is the growth index, and ψ_{b }is a bank fixed effect. Column (3) represents the entire sample, while Column (4) represents the post-reform period. The results in columns (3) and (4) are similar in sign to (1) and (2), and this time we may say with some confidence that they are statistically significant.

33