Following up on the comment from @of_s, and with the huge caveat that this data is 10+ years old and the below rows are cherry-picked, rank-based predictions that drove an S&P500 market neutral fund for 5+ years. Annual returns averaged 20.2% which used to be among the top of the heap for market-neutral systematic funds.
| Prediction Horizon weeks | 1 | 2 | 4 | 13 |
|---|---|---|---|---|
| Return metric, for horizon | 4.79 | 17.41 | 40.54 | 149.68 |
| Return metric, avg per week | 4.79 | 8.71 | 10.14 | 11.51 |
| Alpha contribution (annualized) | -0.20% | 5.09% | 6.48% | 8.58% |
My intuition is a 2 to 4 week prediction window will yield better results for signals. And the performance gap between main tournament and signals might be supporting my intuition. However, feel free to give me a hard time about this 
as I haven’t done a study with recent data or the signals universe.
I think I heard @mdo say before that targets matter more than modeling techniques – I wonder if numerai has done or will do a study to determine optimal signals target horizon.