Replies: 1 comment 7 replies
-
Hello, thank you for your feedback and your PR! Indeed, Yes, we could add a utility function that optimally discretizes the assets units. The reason it was not included is that the package was initially used for institutional investment, for which unit rounding is negligible compared to the notional invested. But I recognise that for small investments it may be beneficial. Regarding the |
Beta Was this translation helpful? Give feedback.
-
Hey, thanks for developing this great library! I'm trying to utilize skfolio to build a personal investment portfolio. I'm using the provided example notebooks, specifically the HRP vs HERC notebook. This example uses the
CombinatorialPurgedCV
class to run cross validation with multiple test paths from different training folds combinations.Now for my first question: is
CombinatorialPurgedCV
purely intended to validate model behavior, i.e. see the distributions of different measures (Sharpe, CVaR, etc) to determine the "stability" of the model? Or should I pick one path (the best one?) from the output ofCombinatorialPurgedCV
and use the asset allocation of that path?Then another question / note: I was experimenting with the PyPortfolioOpt lib before finding skfolio. They have a nice
[DiscreteAllocation
](https://pyportfolioopt.readthedocs.io/en/latest/Postprocessing.html?highlight=discrete#pypfopt.discrete_allocation.DiscreteAllocation) helper to convert asset weights, latest price data and target portfolio size into discrete number of assets to buy. Are there any plans of incorporating something like that into skfolio?My final question relates to to the prices_to_returns helper. If I have pricing data such as this one, where stock B has entered the stock market after A
...then
prices_to_returns
will trim both A and B values to start from the first available date where B has data:Is there a way around this? I would like to train a model with a portfolio that includes B, but having the data truncated like that makes it more difficult to build a good model, as so much information is lost.
Beta Was this translation helpful? Give feedback.
All reactions