Group by direction

[andreas-vester]

I would like to apply a strategy on a portfolio and backtest/groub_by the results by direction (longonly, shortonly, both).

What I have so far:

• closing prices of all assets in the portfolio
• relevant signals
  • boolean dataframe with long entries
  • boolean dataframe with long exits
  • boolean dataframe with short entries
  • boolean dataframe with short exits
• one size array containing positve target percentages for long exposure and negative target percentages for short exposure
• adjusted _price dataframe that has security weights and parameter values as column headers

So far, I am using vbt.Portfolio.from_orders() to analyze longonly.

vbt.Portfolio.from_orders(
    close=_price,
    size=size,
    size_type="targetpercent",
    price=close_daily,
    call_seq="auto",
    cash_sharing=True,
    group_by=["mlr_window", "sma_window"],    
)

No, I am wondering how to best implement distinct backtests that incorporate longonly, shortonly, and both. I was playing aroung with direction=["longonly", "shortonly", "both"], but this leads to a shape mismatch.

Let's say I have five securities in my portfolio and I am going to test a single parameter combination, i.e. five columns in my _price and size dataframe. Do I need to broadcast them manually to 15 columns? Do I need to insert the direction as another column level (5x "longonly", 5x "shortonly", and 5x "both")?

At the end of the day, I am basically trying to replicate the example from within the docs (here with respect to from_signals), but on a portfolio basis.

>>> pf = vbt.Portfolio.from_signals(
...     close,
...     entries=pd.Series([True, True, True, False, False]),
...     exits=pd.Series([False, False, True, True, True]),
...     direction=[list(Direction)],
...     broadcast_kwargs=dict(columns_from=Direction._fields))
>>> pf.asset_flow()
    Long  Short    All
0  100.0 -100.0  100.0
1    0.0    0.0    0.0
2    0.0    0.0    0.0
3 -100.0   50.0 -200.0
4    0.0    0.0    0.0

[polakowo]

@andreas-vester yes, you need to tile the price DataFrame three times and append a new column level "direction" (you can use price.vbt.tile for this). You can then pass direction as direction=price.columns.get_level_values('direction').values.

[andreas-vester]

Thanks. This is working :-)