riskkit

A framework-agnostic risk-management toolkit for systematic traders.

Most open-source trading tools focus on the fun part — signals, indicators, backtesting engines. They leave the part that actually decides whether you survive thin or absent: how big a position to take, when to cut size, and when to stop trading altogether. That's what blows up retail algo traders, not a bad entry signal.

riskkit is that missing layer. The components are pure Python with no dependency on any exchange, data provider, or backtesting framework. They don't know what CCXT is. You feed them numbers; they hand back decisions you can audit. Drop them into backtesting.py, vectorbt, backtrader, freqtrade, or your own loop.

⚠️ Not financial advice. riskkit helps you implement a risk policy you have chosen. It does not choose one for you, and it cannot make a losing strategy profitable. Test everything on paper first.

---

Install

pip install "git+https://github.com/HasibDaddy/riskkit.git"

Zero runtime dependencies. Python 3.9+. (A PyPI release is on the way — until then, install straight from GitHub with the line above.)

---

What's in the box

Component	What it does
PositionSizer	Volatility-adjusted fixed-fractional sizing with an optional half-Kelly ceiling, a reduction ladder for losing streaks / drawdowns, and a hard notional cap.
DrawdownManager	Tracks high-water-mark drawdown, maps it onto a tier ladder (cut size → raise the bar → halt), with a recovery ramp and a rolling weekly-loss pause.
StopEngine	A composable stop stack per position — initial, break-even, ATR trailing, EMA trailing, time, and volatility stops. The tightest one wins; stops only ever move closer.
CorrelationGuard	At most one open position per correlation group. Groups can be static (you define them) or computed dynamically from a rolling return-correlation matrix.
SessionManager	Daily trade/loss caps, profit-taking stops, minimum spacing, escalating cooldowns after losing streaks, and tilt detection.
PreTradeValidator	The composable final gate: runs every rule against a proposed trade and vetoes it if any fails — returning exactly which checks passed and failed.

Every decision is auditable — the sizer returns which multipliers fired, the stop engine logs each adjustment, and the validator returns a pass/fail line for every single check.

---

Quick start

Sizing a trade

from riskkit import PositionSizer, SizingInputs

sizer = PositionSizer(
    base_risk_pct=1.0,      # risk 1% of equity per trade, before adjustments
    max_risk_pct=1.5,       # never risk more than 1.5%
    max_notional_pct=4.0,   # never let a position exceed 4% of equity
)

result = sizer.size(SizingInputs(
    equity=10_000,
    entry_price=100.0,
    stop_price=98.0,        # the stop distance defines your risk per unit
    atr=2.5,                # current volatility
    atr_baseline=2.0,       # "normal" volatility -> scales risk down when choppy
    consecutive_losses=2,   # reduction ladder kicks in
    drawdown_pct=4.0,
))

if result.units > 0:
    print(f"Buy {result.units:.4f} units (risk {result.risk_pct:.2%})")
    print("adjustments:", result.multipliers_applied)
else:
    print("Skip:", result.reason_for_zero)

Adapting to drawdown

from riskkit import DrawdownManager

dm = DrawdownManager(tier1_pct=3, tier2_pct=5, tier3_pct=7, halt_pct=10)

state = dm.update(current_equity)   # call once per equity refresh
if state.halted:
    print("No new trades:", state.reason)
else:
    size = base_size * state.size_multiplier   # scale every position by the tier

The two compose naturally: feed DrawdownManager's drawdown_pct and size_multiplier straight into the sizer.

---

Design principles

• Framework-agnostic. No exchange SDK, no pandas requirement in the core, no global state. Just dataclasses in, dataclasses out.
• Auditable, not magic. Every adjustment is named and returned. You can log the exact reason a trade was sized down or skipped.
• Conservative by default. Floors, ceilings, and hard caps bound every knob. The math can recommend; it can never exceed the limits you set.
• Anti-martingale. Size goes down after losses and during drawdowns, never up. There is no "average down" path anywhere in this library.

---

Integrations

riskkit slots into whatever you already use — the examples are runnable:

• backtesting.py — examples/backtesting_py_strategy.py sizes every entry with PositionSizer. On the bundled GOOG dataset that risk layer keeps max drawdown to ~5% while staying net positive — swap in your own signals, the risk model is unchanged.
• freqtrade — examples/freqtrade_callbacks.py drives custom_stake_amount from your risk model instead of a flat stake.
• your own loop — examples/pipeline.py wires drawdown → sizing → validation into one decide_trade() call.

Full docs (mkdocs): clone the repo, then pip install -e ".[docs]" && mkdocs serve.

Roadmap

riskkit is extracted and generalized from a working risk-first trading bot. The core six components are in place; next up is making them effortless to drop into the popular frameworks:

• PositionSizer, DrawdownManager, StopEngine
• CorrelationGuard, SessionManager, PreTradeValidator
• First-class adapters / examples for backtesting.py, vectorbt, and freqtrade
• A hosted docs site with end-to-end recipes
• A single RiskManager façade that wires all six together with one config

Feedback on the API is genuinely welcome — open an issue. See the full ROADMAP.md, CONTRIBUTING.md, and the examples.

---

License

MIT © 2026 Hasib. See LICENSE.