banks and 3441 for regional banks.13 Until 1980, banks reported financial statements twice
a year, in September and March, so we constructed annual data of flow variables as the sum
of the semi-annual data. Data at the end of each fiscal year, i.e., March, are used for the
stock variables.
Descriptive statistics for these variables are shown in Table 2. The results reveal that
city banks are much larger than regional banks. For example, city banks have outstanding
loans and deposits that are 15 times larger, on average, than regional banks. City banks
earn 15 times as much as regional banks and pay 1.3 times the salary to each employee, on
average. Regional banks tend to lend smaller amounts to each borrower, and are more
likely to lend to small- and medium-sized firms.
We use year dummy variables to estimate θt in equation (8). To estimate ɪ in
ηt
equations (8) and (9), we use time dummy variables for every two years.14 We cannot use
time dummy variables for each year because they are linearly dependent with IIPt .
Equations (7), (8), and (9) are simultaneously estimated by multivariate regression
(MVR) and three-stage least squares (3SLS). 15 In 3SLS we use rank variables as
instrumental variables for the terms including qi,t , di,t , Ri,t , ri,dt , Pi,t , and Ci,t , the
13 For 1999 and/or 2000, there are some observations that lack data of ASL and/or OPL. Deletion
of them results in only two observations for city banks in 2000. Therefore, for both city and
regional banks, we complement these data by copying the latest available data to include these
observations in the sample. To check the possible bias due to this expediential method, we conduct
the estimation by deleting the two explanatory variables from the demand function. The estimation
results are essentially unchanged.
14 We use a one-year dummy for 1974.
15 Since equation (7) is not necessary for identifying θ from η , we estimated only equations (8)
and (9) jointly. The results were generally unreasonable, suggesting that equation (7) plays an
11