I don't believe this is possible.
In the case that the average length is greater than that calculated by the heuristic but still less than 4,000 bytes it is possible to massage this upwards with a
CAST but even then you might need some hacking about with the query syntax to encourage the
CAST to happen before any memory consuming operator that depends on the estimated row size.
In the case that it is less the
CAST to a shorter size obviously won't work without the possibility of truncating data.
The possible down side of a mis-estimate is that the memory grant may be calculated based on incorrect information.
I've wondered before why the statistics on string columns don't keep track of this. Maybe in the scale of things it simply isn't regarded as objectively that important.