next up previous
Next: Parameters of the SSA Up: Choice of parameters Previous: Length and location of

Window width $m$

The choice of $m$ depends on what kind of structural changes we are looking for. If we allow small gradual changes in the time series, then we cannot take $m$ very large. On the contrary, if we take $m$ too small, then an outlier can be recognized as a structural change. A general rule is that the value of $m$ has to be reasonably large. Of course, $m$ should be smaller than $\tau$ and all the distances between sequential change-points (in the case of multiple changes), otherwise we can either miss or smooth out all changes in our time series.