Just received round 270 day 1 performance today, the model CORR and MMC dropped to very low level. I am confused with this resut, considing similar model used to generate predictions in previous rounds, and those old prediction still maintain 0.05+ CORR on 1july.
Why this could happen that same model can create such big difference on the same date in different rounds ?
I check the leader board and saw the similar pattern. Pls suggest your thoughts.
Just another random day, in Numerailand. Tomorrow will be another one.
Thanks for the sharing. Let’s wait for tomorrow.
It’s still 4 weeks till round resolution, day 1 scores are very noisy, don’t be bothered by it.
The problem of a weak signal in a lot of noise. Takes several days to average down the noise and the signal to start to correlate. Good robust models should see a slow solid linear rise over the 4 weeks (unlike mine which are buffeted around by chance and with luck end on a high)
Even “good” models won’t necessarily see a slow solid linear rise over the 4 weeks. I do think there’s still a lot to be learned about the intraround movement (mostly volatility) of a model’s daily scores, but no reason to think that it should be a slow linear rise
Yes I was thinking of a rather over-idealized situation
Where are you seeing these updated values?
Also, wouldn’t the randomness of the stock market cause these issues as well?
after comparing the performance from 1JUL to 6JUL in other models in other round, i start to realize that all the model facing certain down turn in that period. And it lead to my following understanding:
So we need to be patient to wait the market gradually aligned with prediction and score result move up daily.
The final score for a round is NOT a sum of the individual daily scores. Only the score on the last day of round close is what matters.
Please read the post that was linked to above: Relationship of daily round correlations to final round correlations - #10 by jrai