Backtesting Guide

Idealized Close (Default)

Trades fill at the next day's closing price with a flat slippage percentage applied. This is the simplest model and is useful for quick iteration and strategy validation.

• Fill price = close price × (1 + slippage) for buys, × (1 - slippage) for sells
• All orders fill completely (no partial fills)
• No spread or market impact modeling
• Best for: initial strategy validation, comparing strategy variants

Alpaca Daily Realistic

A more realistic execution model that accounts for bid-ask spread, market impact, and liquidity constraints. This produces more conservative (and more accurate) results.

• Spread estimation — Uses the Corwin-Schultz estimator to estimate bid-ask spread from daily high/low prices. Wider spreads for less liquid names.
• Market impact — Square-root impact model based on trade size relative to average daily volume. Larger orders move the price more.
• Partial fills — Orders are capped at the max participation rate (default 10% of daily volume). If your order exceeds this, you get a partial fill.
• Fill rate — Calculated based on order size vs. available liquidity. Reported in execution details for each trade.
• Best for: final validation before deployment, capacity analysis, realistic P&L estimation

Cumulative Return

Total return over the backtest period. Calculated as (final equity - initial capital) / initial capital.

Annualized Return

Cumulative return scaled to a yearly rate. Uses 252 trading days for equities, 365 for crypto. Formula: (1 + cumReturn)^(annualizationFactor / tradingDays) - 1.

Sharpe Ratio

Risk-adjusted return. Measures excess return per unit of total volatility. Uses 5% annual risk-free rate. A Sharpe above 1.0 is generally considered good; above 2.0 is excellent.

Sortino Ratio

Like Sharpe, but only penalizes downside volatility. More relevant for strategies that have asymmetric return distributions (large gains, small losses).

Maximum Drawdown

Worst peak-to-trough decline during the backtest. A drawdown of 20% means the portfolio fell 20% from its highest point before recovering.

Win Rate

Percentage of sell trades that were profitable (sell price > average buy price for that symbol). Calculated using FIFO matching.

Profit Factor

Gross profit divided by gross loss. A profit factor above 1.0 means the strategy is profitable overall. Above 2.0 is strong.

Average Holding Period

Average number of days between buying and selling a position. Calculated using FIFO matching of buy/sell pairs.

Total Fees

Sum of all trading fees incurred. Zero for equities (commission-free via Alpaca), volume-tiered for crypto.

Total Dividends

Total dividend income received during the backtest period. Dividends are credited on ex-dates based on shares held.

Monte Carlo Simulation

Generates 1,000 stochastic simulations by bootstrapping from your strategy's historical daily return distribution. This produces confidence bands showing the range of possible outcomes.

• P5 / P50 / P95 bands — The 5th, 50th, and 95th percentile equity paths. The P5-P95 range represents the 90% confidence interval.
• Probability of profit — Fraction of simulated paths that end above the starting capital.
• Median final equity — The expected outcome under typical conditions.
• Uses a fixed seed (42) for reproducible results across runs.
• Requires at least 20 trading days of data to run.

Walk-Forward Analysis

The most important robustness check for detecting overfitting. Splits the equity curve into multiple train/test windows and measures out-of-sample consistency.

• Window count — Default 5 non-overlapping windows. Each window is split 70% train / 30% test.
• Consistency — Fraction of test windows that were profitable. Above 60% is good.
• Degradation ratio — Average test daily return / average train daily return. Above 0.5 is acceptable; below suggests overfitting.
• Train/test correlation — Do in-sample and out-of-sample returns move together? Above 0.5 is good.
• Per-window metrics — Each window shows train return, test return, train Sharpe, test Sharpe, and test max drawdown.

Market Regime Analysis

Detects market regimes from benchmark data and measures your strategy's performance in each regime. This reveals whether your strategy only works in bull markets or is robust across conditions.

• Regimes detected — Bull, Bear, Sideways, High Volatility, Low Volatility
• Per-regime metrics: total return, Sharpe ratio, max drawdown, trading days
• Uses SPY for equity benchmarking, BTC/USD for crypto

Performance Attribution

Decomposes your strategy's return into its sources to understand what drove performance.

• Alpha return — Return attributable to your strategy's skill (above what the market delivered)
• Beta return — Return from market exposure (what you would have earned just holding the benchmark)
• Residual return — Unexplained return (noise or factors not captured)
• Sector attribution — Return contribution from each sector
• Factor exposure — Exposure to common risk factors

Capacity Estimation

Estimates the maximum AUM (assets under management) your strategy can handle before market impact degrades performance.

• Capacity AUM — Maximum portfolio size before the binding constraint is hit
• Binding symbol — The least liquid symbol that limits capacity
• Binding constraint — Whether capacity is limited by volume, spread, or impact
• Based on the max participation rate (default 10% of daily volume)

Start with idealized, finish with realistic

Use the idealized execution model for rapid iteration while tuning your strategy. Once you're happy with the results, switch to the realistic model for final validation. The realistic model will typically show lower returns due to spread and impact costs.

Watch for overfitting

If your strategy scores well on cumulative return but poorly on walk-forward consistency, it may be overfit to the specific date range. Try different date ranges and check that performance is stable. A walk-forward consistency above 60% and a degradation ratio above 0.5 are good signs.

Use survivorship-free mode for accuracy

Enable survivorship-free mode to include delisted securities in your backtest. Without it, your universe only contains stocks that survived to the present day, which introduces a positive bias. With it enabled, positions in delisted securities are force-liquidated at the last available price.

Check capacity before scaling

The capacity estimation tells you the maximum AUM your strategy can handle. If you plan to increase capital, make sure the capacity estimate supports it. Strategies trading small-cap or low-volume names will have lower capacity.

Realism warnings matter

The backtest results include a warnings section that discloses execution model, calendar mode, shorts policy, corporate actions applied, and any partial fills. Review these to understand the assumptions behind your results.