exitkit

Swap your exit policy the way you swap your entry signal.

Twenty-seven exit models in six families behind one interface. Entry logic is well served by open-source backtesting libraries; exit logic usually is not — most ship one stop and one target and leave the rest to you.

One entry rule (10/30 SMA crossover), five exit policies, backtesting.py's sample data. The 30-day time limit holds a 22% drawdown through 2008–09 where the others reach 65%. Regenerate it with examples/plot_exit_policies.py.

pip install exitkit

Thirty seconds

import time
from exitkit import StopLossExitModel, Position, SignalOutput

position = Position(
    position_id="p1",
    entry_time=time.time() - 3600,
    entry_signal=SignalOutput(direction=1, meta={"implied_vol": 0.18}),
    entry_price=400.0,
    quantity=10,
)

model = StopLossExitModel(stop_loss_pct=0.02, trailing=True)

for signal in model.generate_exit_signals(
    [position], {"spot_price": 391.0, "implied_vol": 0.21}
):
    print(signal.exit_reason, signal.meta["loss_pct"], signal.confidence)

stop_loss -0.0225 1.0

Every model takes (positions, market_data) and returns SignalOutput objects carrying the position they close and why. Swapping policy is swapping the constructor.

Works with backtesting.py

pip install exitkit[backtesting]

from backtesting import Backtest, Strategy
from exitkit import StopLossExitModel, FixedTimeExitModel
from exitkit.adapters.backtesting_py import ExitMixin

class SmaCross(ExitMixin, Strategy):
    exit_models = [StopLossExitModel(0.05), FixedTimeExitModel(24 * 30)]

    def next(self):
        self.apply_exits()          # close whatever the policy says to close
        if crossover(self.s1, self.s2) and not self.position:
            self.buy()

Same entry signal, six exit policies, on backtesting.py's own sample data (examples/compare_exit_policies.py):

exit policy           return %  trades  max DD %  Sharpe
none (hold)              326.1       1     -65.3    0.47
stop 2%                   47.6       8     -64.6    0.15
stop 5%                  283.9       2     -65.3    0.44
take profit 10%          187.5      10     -64.0    0.39
time limit 30d           208.3      31     -22.3    0.68
stop 5% + tp 10%         100.0      29     -28.3    0.49

One dataset and one entry rule, so read it as an illustration rather than a finding — but it is the comparison the library exists to make cheap. Holding time is measured against the bar's clock, not the wall clock, so a replay ages positions by simulated time rather than by whenever you happened to run it.

The six families

from exitkit import FAMILIES

for name, models in FAMILIES.items():
    print(name, [m.__name__ for m in models])

FAMILIES is also how the test suite exercises every model uniformly — adding a model puts it under the whole battery automatically.

Missing market data raises

The one opinion this library holds. Required fields are checked at the boundary and name what is absent:

model.generate_exit_signals([position], {"implied_vol": 0.21})

MissingMarketData: market data is missing 'spot_price'; got: implied_vol.
Exit models require this field - supply it rather than letting a default stand
in, which silently fabricates the decision.

None, NaN and unparseable values count as missing. 0.0 does not.

Where this fits

exitkit decides when to close. It does not fetch data, route orders, or run a backtest loop — hand it positions and market data from whatever you already use.

SignalOutput is a plain dataclass, so wiring it into an existing engine is a translation layer, not an adoption.

Why this exists

The catalogue was extracted from a private options-research program. Writing the test suite surfaced three defects that had survived in running code, all fixed here with regression tests named after them.

Thirty-six fabricated market-data fallbacks. Every model read its inputs as market_data.get('spot_price', 350.0) or .get('implied_vol', 0.2). A caller who omitted a field did not get an error — they got an exit decision computed against an invented price. The volatility default is quieter still: it appears in ratio denominators, so a missing value produces a vol ratio of exactly 1.0, which reads as "no change" rather than "no data". That is why the boundary check above exists.

Time-based exits could not fire. check_time_exit read position.get_holding_hours(), which divided holding_period — a field only ever assigned inside Position.update_pnl(). A model that did not first mark the position saw zero hours held, so a position held nine hours against a two-hour limit did not exit. Holding time is now derived from entry_time; a derived quantity should not depend on another call's side effect.

MarketHoursExitModel had never run. The module called time.time() without importing time, so every invocation raised NameError. It also defaulted its timestamp to the wall clock, which meant a backtest evaluated market hours against whenever you happened to run it — a run at 02:00 would hold everything. That one was found by CI, which runs in a different timezone than the author's laptop, and is why the suite now pins fixed instants.

The feature-window argument was also removed from the required position in the signature: it was the first parameter of every model and not one of them read it.

Tests

pip install -e ".[test]"
pytest -q

141 tests. Four are parametrized across all twenty-seven models, so each must construct, refuse to decide on empty market data, run on complete data, and return nothing when there are no positions.

Licence

MIT. See CHANGELOG.md and CONTRIBUTING.md.