The supersmoother algorithm computes three separate smooth curves from
the input data with symmetric spans of 0.05*n, 0.2*n and 0.5*n, where n
is the number of data points. The best of the three smooth curves is
chosen for each predicted point using leave-one-out cross validation. The
best spans are then smoothed by a fixed-span smoother (span = 0.2*n) and
the prediction is computed by linearly interpolating between the three
smooth curves. This final smooth curve is then smmothed again with a
fixed-span smoother (span = 0.05*n).

According to comments by Friedman, "For small samples (n < 40) or if
there are substantial serial correlations between observations close in
x-value, then a prespecified fixed span smoother (span > 0) should be
used. Reasonable span values are 0.2 to 0.4."
This function is basically equal to the
function supsmu written in MALTAB by Douglas M. Schwarz.

Email: dmschwarz=ieee*org, dmschwarz=urgrad*rochester*edu
Real_email = regexprep(Email,{'=','*'},{'@','.'})
The following modifications with respect to the original function have
been made:

[

**1**] In case of constant values of x over a span program was producng
NA. Modifications have been done in the subroutinesto cope with this
case.

[

**2**] The formula for the cross validation residuals in presence of
constant values has been introduced.

[

**3**] All the Mlint suggestions have been incorporated.

[

**4**] The help has been put inside the FSDA style.

[

**5**] A series of examples which explore the different options has been
added.