Cumulative returns

[Sanjan611]

For a definition like this -

pf = vbt.Portfolio.from_signals(
    close = df['close'],
    entries = entries,
    exits = exits,
    init_cash = init_cash,
)

If I understand this right, pf.cumulative_returns() gives the compounded returns over the period of time. So any profit or loss at a trade is added to the value of the portfolio, which forms the amount being invested for the next trade. Is there a method that gives returns such that the profit/loss doesn't affect the cash invested?

I'm seeking to get a simpler form of returns, where if I invest say $100 and then make a profit of $10 on one trade, I'd like to keep aside that profit and just continue to have $100 in my portfolio rather than $110. So basically, calculate each trade's profit or loss on a fixed capital.

[Sanjan611]

After some exploration and reading, I've found what I needed to be in pf.daily_returns().cumsum().

Using this cumulative approach focuses just on the day to day differences when you're holding an asset (i.e between a buy and a sell). If I'm not wrong, pf.cumulative_returns() does a similar thing but with compounded returns, where the returns made on one day affect the the returns made on future days.

Just putting it out here for anyone else!

[polakowo]

@Sanjan611 why not using pf.returns()? (it's the same as value.pct_change - daily change in equity), while cumulative_returns is compounding. You should never do daily_returns().cumsum(). There is also such thing as active_returns that do not take cash into consideration, only the returns from your current position.

[polakowo]

Yes, both return series should be the same for daily frequency.

You could be correct in assuming cumsum somehow leads to a non-compounded return. But any returns that are based on portfolio value (such as pf.returns()) are taking into account the current cash, which is different across all trades, and so returns can't be compared safely. Consider the following example with two trades that have the same position size and performance but different amount of remaining cash:

>>> import pandas as pd
>>> import vectorbt as vbt

>>> portfolio = vbt.Portfolio.from_orders(
...     [1, 2, 3, 1, 2, 3],
...     [1, 0, -1, 1, 0, -1]
... )
>>> portfolio.trades.pnl.to_matrix()
0    NaN
1    NaN
2    2.0
3    NaN
4    NaN
5    2.0
dtype: float64  # performance is the same

>>> portfolio.returns()
0    0.000000
1    0.010000
2    0.009901
3    0.000000
4    0.009804
5    0.009709
dtype: float64  # returns are different because of cash

>>> portfolio.active_returns()
0    0.0
1    1.0
2    0.5
3    0.0
4    1.0
5    0.5
dtype: float64  # returns the same so you can now sum them up

What active_returns does is to entirely remove the remaining cash out of the equation. I will rename it to position_returns or similar.

[Sanjan611]

Thanks a lot!

[eervin123]

Hi @polakowo did you end up removing the active_returns() method?

[polakowo]

@eervin123 it has been renamed to asset_returns

[eervin123]

Thank you.