Beating Best Constant Rebalanced Portfolio

[sword134]

Is it realistic that when running an exponential gradient model I am beating the best constant rebalanced portfolio return on the same timeframe? I thought that the BCRP was the creme de la cremé of investing and it couldn't be better, so how come that I am getting a EG model which has a higher return with the exact same data and timeframe?

[Jackal08]

Hi @sword134,

I invite you to join our Slack Channel, it's a great place to ask questions and get community feedback.

Alternatively, if you dive into the source code and find a bug, please do raise the issue here and we can look into it.

Right now, these types of questions would require a large time investment on our side to get reacquainted with the literature.

[sword134]

@Jackal08 i've found the bug. I use yfinance to get the data and there was an error in one of the tickers where it suddenly dropped 80% or so in value and did a full recovery the next day (didn't happen irl)

[Jackal08]

Ah, great to know its not on our side, thanks.

[sword134]

@Jackal08 actually having sorted out the dataset issue, the best constant rebalanced portfolio returns are still smaller than the ones I get using an EG. Therefore I am reopening the issue.

[sword134]

When looking at the returns it will sometimes return over 100% while there is no where near that big increase in any of the assets its allowed to trade, how can this be?

[sword134]

@Jackal08 for example, when I am looking at the returns and calculate the difference between each day of returns (I am training a daily model for the example) I can see that one of the days it went from 7,37 to 7,05 there by losing -0,31 (31%). When I look at the assets it was in invested in for that day I can see it had 99% of its portfolio in 1 stock, this stock on the same day "only" lost -14% of its value, so how come its showing a completely wrong number in returns?