Skip to content

Glossary

A quick reference for the abbreviations, metrics, and statistical methods that appear throughout this documentation and the alpha-forge CLI output (synced with alpha-forge issue #1192).

Performance metrics

Term Field / full name Meaning
Sharpe sharpe_ratio Risk-adjusted return = return divided by volatility (standard deviation). Higher is better. The default objective for optimization and WFT.
Sortino sortino_ratio Risk-adjusted return that only divides by downside volatility (variation of negative returns). A downside-only variant of Sharpe.
Calmar calmar_ratio CAGR ÷ max drawdown. Return efficiency relative to drawdown resilience. Higher is better.
MDD (Max Drawdown) max_drawdown_pct The largest peak-to-trough decline of the equity curve (%). Smaller is better.
CAGR cagr_pct Compound Annual Growth Rate (%). Lets you compare returns regardless of period length.
PF (Profit Factor) profit_factor Gross profit ÷ gross loss. Above 1.0 means net profit. null when there are no losses (gross loss = 0).
Win rate win_rate_pct Percentage of trades that closed with a profit.
tail_ratio 95th-percentile gain ÷ |5th-percentile loss| of daily returns. Above 1.0 means a fatter upside tail.
VaR / CVaR var_95 / cvar_95 Value at Risk (worst expected loss at 95% confidence) and the conditional expected loss beyond that threshold.

Validation and statistics

Term Full name Meaning
IS In-Sample The period used for parameter optimization / training — the first half of backtest run --split.
OOS Out-of-Sample The period held out for validation only — the second half of --split. A small IS↔OOS performance gap indicates low overfitting.
WFT Walk-Forward Test Robustness check that repeatedly optimizes on IS then validates on OOS across N rolling windows (optimize walk-forward). Detects overfitting.
DSR Deflated Sharpe Ratio A significance measure that deflates the Sharpe ratio to correct for "lucky" results from running many optimization trials (Bailey & López de Prado, 2014). The more trials (n_trials), the stricter the correction.
MC Monte Carlo Runs many randomized simulations (trade reordering / resampling) to estimate the final-equity distribution, ruin probability (ruin_probability_pct), 95th-percentile MDD, and more (backtest monte-carlo).
PSR Probabilistic Sharpe Ratio The probability that the observed Sharpe exceeds a benchmark value. The statistical basis of DSR.

Exploration and regimes

Term Full name Meaning
HMM Hidden Markov Model A probabilistic model that classifies the market into hidden "regimes" (e.g. Bull / Range / Bear) so signals can switch per state. Used via the HMM indicator.
Regime A market state such as bullish, bearish, or range-bound. Gated by HMM or macro indicators to control entries per state.
near_pass rescue zone A mechanism in the explore pre_filter that lets "almost-passing" strategies proceed to the optimizer (pre_filter.near_pass in goals.yaml).
pre_filter A pre-check in explore run that aborts a strategy right after the backtest if it fails minimum Sharpe / MDD / trade-count thresholds, skipping optimization and WFT to save compute.