autoinv

Ray Data Co’s Automated Investing toolkit. A small library of reusable utilities for quant research and strategy development, built from the L1-L5 curriculum drills (gemchange roadmap) and Halls-Moore’s Successful Algorithmic Trading.

Design principles

Time-series aware by default. Cross-validation utilities use TimeSeriesSplit, not shuffled KFold. We do not expose shuffle=True on financial data.
Honest metrics. Every performance number is reported alongside the relevant baseline (majority class, buy-and-hold, random permutation).
Bias-aware. The validation module exposes a four-bias audit checklist (optimisation, look-ahead, survivorship, cognitive) that should gate every strategy before it trades real capital.
Small surface area. Each module is focused. Functions, not classes, unless state is genuinely required.

Modules

Module	Purpose
`data`	Price pulls (yfinance), returns, lagged features, Fama-French factors (direct from Dartmouth)
`stats`	Normality tests, Student-t MLE, factor regression with Newey-West HAC, permutation test
`validation`	`TimeSeriesSplit` helper, `score_walk_forward`, `BiasAudit` checklist
`metrics`	Sharpe, drawdown decomposition, Brier score, one-shot strategy report
`portfolio`	Cost-aware Markowitz, efficient frontier, PCA factor decomposition
`pricing`	Black-Scholes, Greeks, Monte Carlo, put-call parity
`engine`	Minimal event-driven backtester skeleton (DataHandler / Strategy / Portfolio / ExecutionHandler)
`polymarket`	Anonymous read-only client for Polymarket Gamma + CLOB + Data APIs (markets, prices-history, orderbook)
`kalshi`	Anonymous read-only client for Kalshi trading API v2 (markets, events, candlesticks, orderbook). Mirrors the polymarket module structure so the two can be used interchangeably for arbitrage scans.

Quickstart

from autoinv import data, stats, validation, metrics, engine

# 1. Pull prices and returns
prices = data.get_prices("SPY", start="2020-01-01", end="2026-04-01")
returns = data.get_returns("SPY", start="2020-01-01", end="2026-04-01")

# 2. Normality audit
report = stats.normality_report(returns.to_numpy())
print(f"Fat tails? {report.rejected_at_5pct}")

# 3. Test a signal with permutation
signal = (returns.shift(1) > 0).fillna(False).to_numpy()
perm = stats.permutation_test(returns.to_numpy(), signal, n_permutations=10_000)
print(f"Permutation p: {perm.p_one_sided:.4f}")

# 4. Run a strategy through the backtester
# (see scripts/demo_buy_after_up_day_v2.py for a full example)

# 5. Performance report
strategy_returns = ...  # from backtest
r = metrics.strategy_report(strategy_returns)
print(f"Sharpe: {r.sharpe:.3f}, MaxDD: {r.max_drawdown:.2%}")

# 6. Gate with the bias audit
audit = validation.BiasAudit(
    optimisation_bias=False,
    look_ahead_bias=False,
    survivorship_bias=False,
    cognitive_bias=False,
)
assert audit.passed, audit.report()

Tests

.venv/bin/python -m pytest tests/

All 30 tests green as of 2026-04-10.

What’s intentionally NOT here