Best practices for dealing with nan assets (e.g., IPOs) about cvxportfolio HOT 3 CLOSED

Ok, you ran into a temporary patch. Right now (assuming you're using the repo code) it drops all assets that don't have at least min_history_for_inclusion days of non-missing values at the start date of the backtest (default 252). They way it should work is by breaking the backtest into parts any time a new asset appears (after its min_history...), because each backtest needs to have the same number of assets (but we can concatenate them). All forecasters are robust against nans so they're not an issue. I'm refactoring the code of the simulator this week, not sure if I'll finish that feature soon but definitely in the next 2 weeks. You can try by reducing min_history... and starting the backtest after uber starts trading + min_history...

from cvxportfolio.

Follow-on comment:
The simulator doesn't fail if only internal pricing is used, but the new assets are not represented in the ptf at all.

Code snippet (using the sample code in readme):

simulator = cp.MarketSimulator(['AAPL', 'AMZN', 'TSLA', 'GM', 'CVX', 'UBER'])
backtest = simulator.backtest(policy, start_time='2019-01-01', end_time='2021-04-01')  # UBER went public in mid-2019

Results (UBER not represented at all):

from cvxportfolio.

This should work now, you can test with the hello_world.py in the repo (I added $UBER as you suggested). It's quite difficult to do in fact, but at least for IPOs it should already be correct. Each stock enters 1 year after its IPO (you can change it with the min_history keyword).

from cvxportfolio.