Kernel Regression Smoothing: for use with time-series data. FORTRAN algorithm

Kernel.exe right click to download executable

raw.dat - in this sample input datafile, 16 is the number of columns to be smoothed, but the first column is the order of the time series and is not counted in the other 16. Smoothed output will be found in a new file, est.dat. The derivative and second derivate curves are output in der.dat and 2nd.dat. The fortran program is adapted from those at Kernel regression smoothing subroutines for biostatistics, which optimize the local and global bandwidth, give the optimally smoothed curve and its first and second derivatives. The additional option is provided to adjust the local bandwith: lower is more detail is required, or higher for excess smoothing.

A separate interpolation might be required if additional time points are to be estimaed.

Time-series in an excel spreadsheet can be pasted into raw.dat as follows: the time variable should be at the left; followed by N time-series variables in the columns. Copy and paste these data into WINDOWS (as a table, i.e., a simple paste). Use option Table>Converrt>Table to Text, comma separated. Add a line at the top with the number N. Save this file as a txt file. Then find this file in windows explorer and rename to raw.dat. Now double click the Smoothing.exe file in that same directory and the program will execute. It needs only a few carriage returns ("enter" presses) after data are presented in the DOS window to allow you to check results.

Once the program runs, the results will be in the three *.dat files.

EXPLODE.EXE fortran routines and a single-column version (remove the first line in the raw.dat datafile since there is only one variable)