fastbt-quant 0.6.0

Pandas-native backtesting library with backtest/live parity

fastbt-quant

Pandas-native backtesting library with backtest/live parity.

Distribution name on PyPI is fastbt-quant; the import name stays fastbt.

Install

pip install fastbt-quant                  # core
pip install "fastbt-quant[data]"          # + idata.fyi data fetching
pip install "fastbt-quant[report]"        # + HTML tearsheets via quantstats
pip install "fastbt-quant[signal]"        # + live signal server (aiohttp)
pip install "fastbt-quant[data,report,signal]"   # everything

Requires Python 3.11+.

Design principles

• BYO data. Pandas DataFrame in, BacktestResult out. No data providers bundled.
• One strategy class, two run modes. Same code for backtest and (eventually) live — the engine swaps, the strategy doesn't.
• PaperBroker only in v0.0.1. Real broker adapters (Alpaca, CCXT, Kite) arrive in v0.2+.
• Declarative JSON strategies coming in v0.0.2. The Python subclass API stays as the escape hatch.

Quick start

from fastbt import Strategy, backtest
from fastbt.indicators import sma

class SmaCross(Strategy):
    def on_bar(self, ctx):
        if len(ctx.history) < 30:
            return
        close = ctx.history["close"]
        fast = sma(close, 10).iloc[-1]
        slow = sma(close, 30).iloc[-1]
        prev_fast = sma(close, 10).iloc[-2]
        prev_slow = sma(close, 30).iloc[-2]

        if prev_fast <= prev_slow and fast > slow and ctx.position.is_flat:
            ctx.buy(qty=100)
        elif prev_fast >= prev_slow and fast < slow and ctx.position.is_long:
            ctx.close()

# `data` is any OHLCV DataFrame — bring your own source.
result = backtest(SmaCross(), data=data, initial_cash=100_000)
print(result)
# BacktestResult(trades=7, return=12.40%, max_dd=5.20%, sharpe=1.18, win_rate=57.1%)

JSON strategies (v0.4.0)

Single-symbol and portfolio strategies can be expressed entirely in JSON. One loader handles both:

import json
from fastbt import load_strategy, portfolio_backtest

with open("examples/momentum_portfolio.json") as f:
    strategy = load_strategy(json.load(f))

# `data` is a dict[str, pd.DataFrame] for portfolio mode.
result = portfolio_backtest(strategy, data=universe, initial_cash=1_000_000)

A portfolio JSON spec describes the same lifecycle (universe → rank → select → allocate → schedule) the Python PortfolioStrategy exposes. See examples/momentum_portfolio.json for a complete momentum-rotation example.

OHLCV schema contract

Your DataFrame must have:

• Columns: open, high, low, close, volume (case-insensitive)
• Index: pd.DatetimeIndex, monotonic increasing, no duplicates
• No NaN values
• high >= max(open, close), low <= min(open, close), volume >= 0

Call fastbt.validate_ohlcv(df) at your ingest boundary to fail fast.

Fill policy (important)

The single biggest source of "backtest was profitable but live loses money" is fill timing. fastbt makes this explicit:

• fill_policy="next_open" (default, realistic) — order submitted on bar N fills at bar N+1's open. You only knew close after bar N ended.
• fill_policy="current_close" (optimistic) — fills at bar N's close. Use only when your signal is known intra-bar.

Develop

git clone https://github.com/Vbhadala/fastbt
cd fastbt
uv sync --extra dev
uv run pytest
uv run python examples/sma_crossover.py

Portfolio strategies (v0.3.0)

Built-in allocators and risk controls so portfolio strategies don't hand-code these every time.

from fastbt import PortfolioStrategy, portfolio_backtest, RebalanceSchedule
from fastbt.allocators import (
    annualized_volatility, inverse_volatility, momentum_score,
)
from fastbt.indicators import roc

class MomentumTop3(PortfolioStrategy):
    def rank(self, ctx, universe):
        scored = []
        for sym in universe:
            close = ctx.history[sym]["close"]
            score = momentum_score(
                {"r90d": roc(close, 90).iloc[-1] if len(close) >= 90 else None},
                {"r90d": 1.0},
            )
            scored.append((sym, score))
        return sorted(scored, key=lambda x: x[1], reverse=True)

    def select(self, ctx, ranked):
        return [s for s, _ in ranked[:3]]

    def allocate(self, ctx, selected):
        vols = {s: annualized_volatility(ctx.history[s]["close"], 60) for s in selected}
        return inverse_volatility(selected, vols)

result = portfolio_backtest(
    MomentumTop3(), data,
    schedule=RebalanceSchedule.monthly(day=1),
    min_drift_pct=2.0,                  # skip trades under 2% drift
    vol_trailing_multiplier=2.0,        # vol-adaptive trailing stop
    vol_trailing_fallback_pct=10.0,
    risk_free_rate=6.0,                 # for Sharpe / Sortino
)

print(result.metrics.cagr, result.metrics.sortino)
print(len(result.stop_outs), "stop-outs recorded")

See examples/momentum_portfolio.py for a complete runnable walkthrough.

Roadmap

Version	Scope
v0.0.1	Single-symbol backtest, PaperBroker, 6 indicators, OHLCV validator
v0.0.2	JSON declarative strategies, short positions, 15 more indicators
v0.1.0	Portfolio/universe strategies (rank + select), live mode via async feed
v0.2.0	Live signal engine + webhook server + idata.fyi data module
v0.3.0	Allocators, vol-trailing stops, drift-aware rebalancing, CAGR/Sortino metrics
v0.4.0	JSON portfolio strategy spec, vol_trailing in single-symbol JSON
v0.5.0	First PyPI release as fastbt-quant. Symbol-aware fills (breaking: Bar now requires symbol).
v0.6.0+	Real broker adapters (Alpaca/Kite), walk-forward validation, intrabar stop simulation

License

MIT