Terms and definitions

Alpha

Return in excess of a benchmark after adjusting for beta (market exposure). True alpha is rare; most apparent alpha in backtests is overfitting noise.

See also: Beta, Sharpe Ratio

Anchor walk-forward

A walk-forward mode where the in-sample window always starts from the beginning of the data and expands forward. Contrast with rolling walk-forward, where the window slides.

See also: Walk-Forward Analysis, Rolling walk-forward

Annualised Sharpe ratio

Daily Sharpe ratio scaled by √252 (trading days per year). The scaling assumes i.i.d. returns, which is often violated. Autocorrelated returns require a different scaling factor.

See also: Sharpe Ratio, Deflated Sharpe Ratio

Backtest

Simulation of a trading strategy on historical data. All backtest results on this platform include statistical validity metrics (DSR, PSR) and a deterministic result hash.

Beta

Sensitivity of a strategy's returns to a benchmark (typically the broad market). A beta of 1.2 means the strategy moves 1.2× the market on average.

See also: Alpha

CAGR

Compound Annual Growth Rate. The constant annual return that would produce the same total return over the same period. More meaningful than total return alone because it accounts for time.

Calmar ratio

CAGR divided by maximum drawdown. Measures return per unit of peak-to-trough loss. Values above 0.5 are reasonable; above 1.0 is strong.

See also: Maximum Drawdown, CAGR

CPCV

Combinatorially Purged Cross-Validation. Divides data into N groups and evaluates all C(N,k) IS/OOS splits, producing a Probability of Backtest Overfitting (PBO) score.

See also: PBO, Walk-Forward Analysis

Curve-fitting

Optimising strategy parameters to match historical noise rather than persistent signal. Produces high in-sample metrics that collapse out-of-sample. The core problem of retail backtesting.

See also: Overfitting, DSR

Data snooping bias

The inflation of apparent performance when many strategy variations are tested on the same historical data. Every look at the results consumes degrees of freedom, even informal ones.

See also: Multiple testing, DSR

Deflated Sharpe Ratio (DSR)

A Sharpe ratio adjusted for the number of strategy variations tested and the non-normality of returns. Answers: what is the probability that the true Sharpe is positive after accounting for multiple testing?

See also: PSR, Multiple testing, Sharpe Ratio

Drawdown

The decline in portfolio value from a peak to a subsequent trough, expressed as a percentage of the peak. Maximum drawdown is the largest such decline over the full backtest period.

See also: Maximum Drawdown, Calmar ratio

Embargo

A buffer of bars excluded from the training set immediately after an out-of-sample period. Prevents the model from training on bars that may still carry information from the test window.

See also: CPCV, Purging

Equity curve

A time series of portfolio value over the backtest period. The shape of the equity curve (smoothness, drawdown depth and duration, recovery speed) is often more informative than a single Sharpe ratio.

Expectancy

Average profit or loss per trade, expressed as a percentage. A positive expectancy is a necessary (but not sufficient) condition for a viable strategy. Expectancy = (Win rate × Avg win) − (Loss rate × Avg loss).

Fixed fractional sizing

Position sizing method where each trade risks a fixed fraction of current equity. Compound growth effect means winning positions grow larger and losing positions shrink, producing an asymmetric return profile.

In-sample (IS)

The data used to optimise or fit strategy parameters. In-sample metrics are biased estimates of future performance because the parameters were chosen to fit this specific data.

See also: Out-of-sample, Walk-Forward Analysis

Kelly criterion

A formula for the theoretically optimal fraction of capital to risk per trade: f = edge / odds. In practice, traders use fractional Kelly (25–50%) because the formula is sensitive to estimation error.

Kurtosis

Measure of the 'fat-tailedness' of a return distribution. Excess kurtosis > 0 means more extreme events (large wins and losses) than a normal distribution predicts. The PSR and DSR account for kurtosis.

See also: Skewness, PSR, DSR

Maximum Drawdown (MDD)

The largest peak-to-trough decline in portfolio value over the backtest period. Live drawdowns typically exceed backtested drawdowns by 1.5–2×.

See also: Drawdown, Calmar ratio

Mean reversion

The tendency of a price or spread to return to a long-run average. Mean-reversion strategies typically show high win rates and fat-tailed loss distributions — which inflates observed Sharpe ratios.

Multiple testing

The statistical problem that arises when many hypotheses are tested on the same data. The probability of finding a spuriously significant result increases with the number of tests. The DSR corrects for this.

See also: Data snooping bias, DSR

Out-of-sample (OOS)

Data not used in parameter optimisation. OOS performance is a less biased estimate of future performance than in-sample performance, though it can still be gamed by looking at OOS results and adjusting.

See also: In-sample, Walk-Forward Analysis

Overfitting

When a model captures noise rather than signal. In trading, overfitting produces high in-sample Sharpe ratios and poor live performance. The DSR, CPCV, and walk-forward analysis all help detect it.

See also: Curve-fitting, Data snooping bias

PBO (Probability of Backtest Overfitting)

The fraction of CPCV folds where the out-of-sample Sharpe ratio is below zero. PBO > 0.5 means more combinations fail than pass — strong evidence of overfitting.

See also: CPCV

Probabilistic Sharpe Ratio (PSR)

The probability that the true (population) Sharpe ratio is positive, given the observed sample Sharpe, number of observations, return skewness, and kurtosis. Ranges from 0 to 1.

See also: DSR, Sharpe Ratio

Profit factor

Gross profit divided by gross loss. A profit factor of 1.5 means the strategy makes $1.50 for every $1.00 lost. Values below 1.0 indicate a net-losing strategy.

Purging

Removing training observations whose labels overlap with the test period. Prevents the model from training on forward-looking information that was available only in hindsight.

See also: Embargo, CPCV

Recovery factor

Total return divided by maximum drawdown. A recovery factor below 1 means the total profit doesn't cover the maximum loss observed during the period.

See also: Maximum Drawdown

Result hash

A SHA-256 fingerprint of the backtest inputs (engine version, strategy definition, market data). Identical inputs always produce the same hash, enabling independent verification of deterministic results.

Rolling walk-forward

A walk-forward mode where both the start and end of the in-sample window advance by a fixed step. Each window covers the same number of bars. Contrast with anchor walk-forward.

See also: Anchor walk-forward, Walk-Forward Analysis

Sharpe ratio

Annualised excess return divided by annualised volatility. The most widely used risk-adjusted return metric, and one of the most widely abused. It ignores skewness, kurtosis, and autocorrelation.

See also: DSR, PSR, Sortino ratio

Skewness

Measure of return distribution asymmetry. Negative skewness (common in trend-following) means large losses occur more often than large gains of equal size. The PSR and DSR account for this.

See also: Kurtosis, PSR

Slippage

The difference between the expected execution price and the actual fill price. Backtests typically underestimate slippage, particularly for illiquid instruments or high-frequency strategies.

Sortino ratio

Like the Sharpe ratio, but uses only downside deviation (returns below a threshold) rather than total volatility. More sensitive to bad outcomes; less penalised by large positive returns.

See also: Sharpe Ratio

Survivorship bias

The error of backtesting only on assets that still exist, ignoring those that were delisted, went bankrupt, or were removed from an index. Produces systematically overstated returns.

Trend following

Strategies that go long assets in an uptrend and short assets in a downtrend. Typically characterised by low win rates, large average wins, and long drawdown periods with high WFE.

Walk-Forward Analysis

An out-of-sample testing methodology that splits data into sequential in-sample and out-of-sample periods, repeating across the full data range. Produces a distribution of OOS Sharpe ratios.

See also: WFE, In-sample, Out-of-sample

WFE (Walk-Forward Efficiency)

Ratio of out-of-sample Sharpe to in-sample Sharpe, clamped to [−1, 2]. WFE near 1 indicates the strategy generalises well. WFE below 0.5 suggests meaningful parameter overfitting.

See also: Walk-Forward Analysis

Win rate

Percentage of trades that close with a profit. High win rate alone does not indicate a good strategy — a strategy with 70% win rate and 3:1 loss/win ratio is a net loser.