The PSR (originally called the Deflated Sharpe Ratio in the seminal paper by Bailey and Lopez de Prado, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2460551) is an approach to correct for the non-Gaussian Distribution that is typical of trading returns.
Is it possible to get it added to the Metrics?
Is it possible to get it added to the Metrics?
Rename
The finantic.ScoreCard extension does compute a "robust" Sharpe Ratio, that you can use. I'm thinking it's taking the median instead of the mean of the numerator of the Sharpe Ratio calculation, but I could be wrong. I have my own Sharpe Ratio calculation in my library, but it doesn't agree exactly with the WL numbers, although they are close.
I took a brief look at the cited paper above, but it appears to be doing something more complicated than just taking the median of the numerator. An overkill in my opinion for most cases.
What does matter is that one evaluates the Sharpe Ratio for each individual symbol in the dataset (using the Symbol Rankings tab of the WL Strategy Rankings tool). Then one should remove any symbols from their datasets that don't rank high enough by Sharpe Ratio.
I took a brief look at the cited paper above, but it appears to be doing something more complicated than just taking the median of the numerator. An overkill in my opinion for most cases.
What does matter is that one evaluates the Sharpe Ratio for each individual symbol in the dataset (using the Symbol Rankings tab of the WL Strategy Rankings tool). Then one should remove any symbols from their datasets that don't rank high enough by Sharpe Ratio.
superticker,
I would suggest that you take a look at Marcos Lopez de Prado's other papers to see how the PSR fits into a much larger view. The PSR works to correct the curse of overfitting as well as the non-Gaussian distribution (primarily the pronounced kurtosis), both of which lead the trader to erroneous conclusions about the quality of their models.
I would suggest that you take a look at Marcos Lopez de Prado's other papers to see how the PSR fits into a much larger view. The PSR works to correct the curse of overfitting as well as the non-Gaussian distribution (primarily the pronounced kurtosis), both of which lead the trader to erroneous conclusions about the quality of their models.
QUOTE:
What does matter is that one evaluates the Sharpe Ratio for each individual symbol in the dataset (using the Symbol Rankings tab of the WL Strategy Rankings tool). Then one should remove any symbols from their datasets that don't rank high enough by Sharpe Ratio.
This assumes that you have enough trades in each symbol to compute an accurate Sharpe Ratio, which might be difficult. Furthermore, as the paper I referenced points out, in situations where the distribution deviates significantly from Gaussian there is a strong potential for a SR that might not be valid.
QUOTE:
This assumes that you have enough trades in each symbol to compute an accurate Sharpe Ratio, which might be difficult.
I totally agree. I don't enter stocks in a dataset that have less than 4 trades (and even that may not be enough). I thought about having my custom ScoreCard implementation even returning NaN's for many metrics if there were less than 4 trades.
QUOTE:
in situations where the distribution deviates significantly from Gaussian there is a strong potential for a SR that might not be valid.
I agree there as well. That's partly why one wants to take the median instead of the mean of the numerator of the Sharpe Ratio calculation ("robust" Sharpe Ratio). But perhaps if the Sharpe Ratio skewness is too high, the ScoreCard implementation should return an NaN for that situation as well.
We don't want the optimization setting strategy parameters based on metrics that return an NaN value.
Your Response
Post
Edit Post
Login is required