Do you usually put your minimum volume buy rule in your simulation strategy or do you put in the universe buy rules?
I optimized my strategy using the sim buy rule and now when I switch to putting it in the universe it actually drops the returns by 20%/year which is a bit surprising. The turnover also triples.
Does it make sense to rank against companies you’ll never buy?
I think doing that affects the rank-order of each feature. But differently for each feature. When the final rank for all of the features is created the result is chaotic (worse than you would think intuitively). Only by random chance does it help a backtest and I think even then the in-sample results are fitted to noise and it hurts the out-of-sample results (even if the in-sample results seem better).
I put my volume/liquidity rules in the universe.
As you noted, usually it doesn't make much sense to rank a lot companies you'll never buy...
BUT excluding those stocks can significantly shrink your universe and affect movement in the top ranks of your system.
If you notice that your turnover triples, the first adjustment to get comparability should be to adjust your Rank or RanPos Sell criterion down to match the turnover of your original sim. If then, the underperformance remains, it could be a) that you have a noisy / low-robustness system or b) that your system really benefits from carrying all stocks (getting full information on momentum and relative value across all stocks) or c) stuff I can't think of right now...
I also put volume/liquidity rules in the universe definition and by volume it is the median dollar value traded over a period.