Efthymios G. Tsionas, "Inference in Dynamic Stochastic Frontier Models",
Journal of Applied Econometrics, Vol. 21, 2006, pp. 669-676.

The paper uses data for 128 large U.S. banks, each observed for 12 years.
The total number of observations is 1536.

The data were taken from the commercial bank and bank holding company
database managed by the Federal Reserve Bank of Chicago. It is based on the
Report of Condition and Income (Call Report) for all U.S. commercial banks
that report to the Federal Reserve banks and the FDIC. In this paper I used
the data for the years 1989-2000 and selected the sub-sample of large banks,
namely those whose total assets exceeded one billion dollars for at least
three (not necessarily consecutive) years. This gives a total of 128 banks
observed over 12 years.

The intermediation approach is followed in which banks are viewed as
transforming various financial and physical resources into loans and
investments. The output variables are:  installment loans (to individuals
for personal/household expenses) (y1), real estate loans (y2),
business loans (y3), federal funds sold and securities purchased under
agreements to resell (y4), other assets (assets that cannot be properly
included in any other asset items in the balance sheet) (y5).

The input variables are: labor (x1), capital (x2), purchased funds (x3),
interest-bearing deposits in total transaction accounts (x4) and
interest-bearing deposits in total non-transaction accounts (x5).
For each input the price is obtained by dividing total expenses on it by
the corresponding input quantity. Thus, for example, the price of labor (w1)
is obtained from expenses on salaries and benefits divided by the number
of full time employees (x1). The same approach is used to obtain w2
through w5. Total cost is then defined as the sum of the expenses on these
five inputs.

All data are in the file tsdata.txt, which is an ASCII file in DOS format.
This data file has 1536 rows and 12 columns, each column corresponding to
one variable. It is zipped in the file tsdata.zip. Unix users should use
"unzip -a".

column 1     the log of total cost normalized by price of fifth input
columns 2-5  the logs of relative prices of four inputs
             (relative to the fifth input)
columns 6-10 the five outputs in log terms
column 11    the trend variable (seventh year is normalized to zero)
column 12    the log of total assets.

To impose the linear homogeneity restrictions, I follow the usual practice
and normalize total cost and all the prices with respect to w5.

I wish to thank Subal C. Kumbhakar for permission to use the data.