Apply half-spread and commission on both open/close trade

[lgrawet]

Enhancement description

Hi,

Full spread is currently applied to buy price on trade open and open+close commission only on trade close. This doesn't reflect real life and affects backtesting stop loss/trailing stop loss trigger level.
Also,, I'm using trade.pl_pct for trailing stop loss calculation and I have to subtract commission manually.
It also leads to bad optimization.

The situation should be :

• trade open: apply half-spread on buy price and also apply commission
• trade close: apply half-spread on sell price and also apply commission

Thanks a lot!

Code sample

Additional info, images

No response

[charlieyanhx]

I hit this while building an exit-policy library against 0.6.6, and I think the report is right but for a slightly different reason than "the spread is too big" — the total is fine, the placement is wrong.

Minimal reproduction on a perfectly flat market, so any deviation from 100 is friction and nothing else:

import pandas as pd
from backtesting import Backtest, Strategy

n = 40
df = pd.DataFrame({"Open":[100.]*n,"High":[100.]*n,"Low":[100.]*n,
                   "Close":[100.]*n,"Volume":[1e6]*n},
                  index=pd.date_range("2024-01-01", periods=n, freq="D"))

class OneTrade(Strategy):
    def init(self): pass
    def next(self):
        if len(self.data) == 5 and not self.position:
            self.buy(size=10)
        elif len(self.data) == 20 and self.position:
            self.position.close()

st = Backtest(df, OneTrade, cash=100_000, spread=0.01, commission=0,
              finalize_trades=True).run()
t = st["_trades"].iloc[0]
print(t.EntryPrice, t.ExitPrice)   # 101.0  100.0

	expected (half each side)	actual
entry	100.50	101.00
exit	99.50	100.00

Round-trip cost is 1.00 either way, which is presumably why this has been stable for so long. The difference is where it sits.

Cause looks like two places:

• _adjusted_price applies price * (1 + copysign(self._spread, size)) — the full spread, on the entry fill.
• At close, closed_trade = trade._replace(exit_price=price, ...) records the raw price, so the exit is marked at a level the trade could not have received.

Why it matters beyond bookkeeping, which is what I think the reporter is describing: a stop or trailing stop expressed as a percentage from entry is measured from a reference that sits a full spread away from the mid, and pl_pct compares that inflated entry against an optimistic exit. So the trigger levels drift with the spread parameter even though the strategy logic didn't change — and it drifts asymmetrically for longs and shorts. Total P&L looks right, so it doesn't show up in reconciliation.

Before anyone writes a patch, the semantics are a maintainer call, because both readings are defensible and they aren't equivalent:

1. Half-spread each side — entry mid * (1 + spread/2), exit mid * (1 - spread/2) for a long. Round-trip cost unchanged, so existing results move only slightly, but stop/TP trigger levels and pl_pct become correct. My reading of the issue title is that this is what's being asked for.
2. Full spread each side — a genuine doubling of friction. Arguably closer to crossing a quoted spread twice, but it would change every published result and every optimisation surface.

Happy to send a PR for whichever you prefer, with tests covering the flat-market case above plus long/short symmetry and the stop-trigger drift. Option 1 is the one I'd expect to be uncontroversial.

(Commission does look correctly applied on both legs — _commissions = commission + trade_open_commission at close — so I think that half of the report is already fine, unless the complaint is about when cash is debited rather than what's counted.)