No. of Recommendations: 1
Tight graphs like that are intrinsically interesting, nice work!
I'm not a dog in this hunt, but FWIW, a walk-forward analysis comes to mind:
1) Use, e.g., the first ten years of data to start off. Determine the trend line from just that past ten years of data.
2) Then advance a quarter and redetermine the trend line from just that past data that now includes one additional past quarter.
3) Repeat - keep determining the trend line up to time 't' using only data up to 't' (never using data to the future of 't').
The 'burn-in period' need not be ten years, that just came to mind.
It might be of interest to segment the data into two pieces: pre-1998 and post-1998.
IIRC, Jim used post-1998 data to show show that BV/share was predictive of future smoothed three and five year returns.
What you do with the walk-forward data depends on what your goal is, but at least 'walk-forward' addresses the issue of 'future-peaking' in what you're doing.
