"Term premia and interest rate forecasts in affine models"
Forthcoming, JF
Supplementary information
Posted January 26, 2001

1.  The data

I spliced data from McCulloch and Kwon and Bliss.  The entire
data set I used, including the out-of-sample period 1995 through 1998,


The structure is

                     ---  yields by maturity ----
year   month   3m  6m  1y  2y  5y  10y

Yields are %/year.

The data through February 1991 are from McCulloch's home page


After February 1991, the data are from Rob Bliss.  I believe you can order,
for a small charge, a CD containing a large panel data set and programs used
for interpolating the data from

Dan Waggoner
Research Department
Federal Reserve Bank of Atlanta
104 Marietta Street NW
Atlanta, GA  30303-2713

2.  Parameter estimates

The only parameter estimates I report in the paper are for "preferred"
specifications, which set a bunch of parameters to zero.  An ASCII file
containing parameter estimates for unconstrained specifications (all
seven of them) is


3.  Cross-sectional fitting errors

The estimation technique assumes that yields on 6-month, 2-year, and 10-year
bonds are measured without error.  Therefore these yields can be inverted to
determine the state vector.  Yields on 3-month, 1-year, and 5-year bonds
are measured with error.  The implied cross-sectional errors in fitting these latter
three yields are reported in the attached files.  The only files here are those
for the preferred model specifications in the paper; the essentially affine
A_0(3) and A_1(3) models and the completely affine A_2(3) model.

These are in-sample cross-sectional errors for January 1952 through December 1994.
The ASCII files do not have dates; row 1 corresponds to 1/1952 and row 516
corresponds to 12/1994.  The ordering of the columns is 3-month error,
1-year error, and 5-year error.  The errors are in decimal form (i.e., multiply by
10000 to get basis point errors).

A_0(3) essentially affine

A_1(3) essentially affine

A_2(3) completely affine