Hi,

I was just wondering how 1-day return on stake is calculated for the leaderboard?

For example, consider integration_test_7 which has a 2.5% return on stake today (17th March). The live rounds are:

https://numer.ai/integration_test_7/submissions/251

https://numer.ai/integration_test_7/submissions/252

https://numer.ai/integration_test_7/submissions/253

https://numer.ai/integration_test_7/submissions/254

The model is only staked on correlation. So the 1-day-return must be calculated from its correlation on those rounds today, or some calculation of its projected payout today compared to yesterday. But I canâ€™t figure out the exact equation. Can anyone explain how itâ€™s calculated?

Thanks in advance.

1 Like

I think they are using the actual staked amounts for each round (along with mmc if it is turned on), so youâ€™d probably have to look at an account with a more significant stake to make sense of it (or pull that data from api). But then I think you just look a what their projected payouts were today relative to yesterday, or maybe it is just the single day return? (That would make more sense probably.) Now Iâ€™m gonna have to go do the mathâ€¦

To my eyes it looks like they are computing the difference between todayâ€™s daily corr estimate (aka â€śscoreâ€ť) and yesterdayâ€™s daily corr estimate and summing those deltas over all four rounds. Since this model is staked on corr it appears theyâ€™re not bothering including any mmc deltas.

For each of the rounds you linked I calculated the following corr deltas between 3/16 and 3/17:

```
251: 0.0001
252: -0.0013
253: 0.0118
254: 0.0142
sum = 0.0248
```

2 Likes

Thanks for the help guys.

Good eye @profricecake, I tried this today and it worked there too. So seems likely this is the calculation.

Coincidentally today is the start of a new round (18th March). So now weâ€™re using 255 to calculate the score instead of 251.

https://numer.ai/integration_test_7/submissions/255

I naively assumed that given this is the first day of the round, you take the previous days value as 0, which seems to give the right number.

```
252: 0.0055
253: -0.0045
254: -0.0009
255: 0.0103
sum = 0.0104
```

This is quite interesting, as it probably explains why on the 11th there was a huge spike in 1-day returns as the new round started. Worth bearing in mind for the future that 1-day returns on Thursdayâ€™s are probably going to be more volatile.

1 Like