Subject: Re: question about momentum screen
Anecdotally, I fine proximity to 52 week high works better, but I couldn't provide you with hard numerical evidence of that. Just years of testing screen variations.

That's because a one year return is sometimes a bit quirky...if there was a sudden short market spike or plunge exactly a year before, it will have no particular predictive power today but really changes a one-year-return figure a lot. Conversely nearness to 52 week high catches things with short, medium, or longer term momentum all at the same time: it's at the top of whatever rally it was in.

FWIW, I sometimes find nearness to a "stale" 52 week high works better.
e.g., compare today's price to the high price price 1-13 months ago, or something like that.
It's possible you want an "older" 52 week high for more value/dividend oriented stocks and a more up-to-date one for more momentum/growth oriented stocks with higher turnover.

As to how to combine different momentum measures, assuming you're looking at a fairly conventional MI screen construction, I recommend trying this approach

Screen1:
[usual steps, filtering]
[usual screen steps, top N by some metric]
Final sort on momentum metric 1

Screen2:
[usual steps, filtering, exactly as in screen1]
[usual screen steps, top N by some metric, exactly as in screen1]
Final sort on momentum metric 2

Final screen:
Do SOS (screen-of-screens) of the top N of screen1 results and screen2 results.
i.e., take each stock from screen1 and screen2 and sum their final sort ranks. Pick the stocks with the best sum of ranks.

You can do this with three or four different momentum sorts.

The reasoning is that the market has different rhythms at different times: sometimes there is a 3 month rally, sometimes a one year rally. By summing the momentum ranks this way your screen isn't critically dependent on any one time frame of history.

For example, during a strong market rally nearness to 52 week high works very well, as mentioned above. Great for 2006, but not 2009. But right after a crash and on the rebound, distance from 52 week LOW works very well. Great for 2009, but not 2006. You can do two variations of your screen using those two sorts, and end up with a screen that is a bit more "versatile", backtesting not badly in both good and bad market years.

As mentioned, I don't have good numerical justification for these suggestions, just seat of the pants experience.

Jim