Quant finance glossary: the terms used on Out of Sample

Quantitative finance borrows words from statistics, computer science and the fund industry, and uses some of them in its own way. This glossary gives a short definition of the terms that come up most often on Out of Sample, with a link to the article where each one is explained properly. It is a reading aid, not a textbook: definitions are kept short, and the linked articles give the formulas, the sources and the caveats.

Educational reference only; nothing here is investment advice. Last updated 22 November 2026.

Research and testing

Backtest
A simulation of how a trading rule would have performed on historical data, with its trades, costs and positions reconstructed month by month or day by day. A backtest describes the past under assumptions; it is evidence, not proof. See a moving-average crossover built in a spreadsheet.
Out of sample
Data that played no part in designing or tuning a strategy. Performance measured in sample, on the data used to build the rule, is almost always flattering; out-of-sample results are the honest test, and the reason for this site's name.
Overfitting
Fitting a model so closely to past noise that it fails on new data. It grows with the number of variants tried, parameters tuned and features added. Our French-language guide to backtests and overfitting covers the main traps.
Multiple testing
Running many trials and keeping the best one. The more trials, the higher the best result will be by luck alone, which is why the number of trials must be reported alongside the winner.
Sharpe ratio
Average excess return divided by the volatility of returns, usually annualised. A Sharpe ratio of 0.5 is typical of a broad equity market over long periods; a backtest showing 3 deserves suspicion before celebration.
Deflated Sharpe ratio
A version of the Sharpe ratio corrected for the number of trials, the length of the sample and non-normal returns, proposed by Bailey and López de Prado. Try it with our deflated Sharpe ratio calculator.
Walk-forward test
A test that repeatedly fits a model on one window of data and evaluates it on the next, rolling forward through time, so that every evaluation is out of sample.

Factors and strategies

Factor
A characteristic shared by many securities that helps explain their returns, such as size, value or momentum, usually measured by a long-short portfolio. See the Fama-French five-factor model.
HML and SMB
The value and size factors of Fama and French: high minus low book-to-market, and small minus big market capitalisation. Their history since 1927 is analysed in the value premium, decade by decade.
Cross-sectional momentum
Buying the assets that rose most over a past window (often twelve months, skipping the latest one) and selling those that fell most, relative to each other. See momentum strategies.
Time-series momentum
Going long an asset whose own past return is positive and short or flat when it is negative, without comparing it to other assets. Trend-following funds use versions of it.
Dual momentum
A combination of relative and absolute momentum: choose the strongest asset, but hold it only if its own trend is positive. See momentum on dual momentum portfolios.
Momentum crash
A sudden, severe loss of a long-short momentum portfolio, typically when beaten-down stocks rebound sharply after a bear market, as in 1932 and 2009. See momentum crashes.
Alpha
The part of a strategy's return not explained by its exposure to the market or to known factors. Much apparent alpha turns out to be factor exposure once the right factors are included.

Financial machine learning

Labelling
Deciding what a model should predict for each observation, for example whether a price hit a profit target before a stop. The triple-barrier method is a common choice. See a laboratory for machine learning in finance.
Meta-labelling
A second model that decides whether to act on, and how much to bet on, signals produced by a first model. See meta-labelling: a toy example.
Fractional differentiation
Differencing a price series by a fractional order so it becomes stationary while keeping part of its memory. See fractional differentiation.
Purged cross-validation
Cross-validation that removes training observations overlapping in time with the test set, so information does not leak from the future into the model.

Risk

Drawdown
The fall from a previous peak in the value of a portfolio, expressed as a percentage. The maximum drawdown is the largest such fall over a period; its length matters as much as its depth.
Volatility targeting
Scaling positions so that expected portfolio volatility stays near a chosen level, cutting exposure when markets become turbulent. See choosing your risk.
Value at risk (VaR)
A loss threshold that a portfolio should exceed only with a given probability over a given horizon, for example 1% over one day. See an introduction to value at risk.
Regime model
A model in which markets switch between hidden states, such as calm and stressed, each with its own return behaviour. See hidden Markov models for market regimes.

Inside a fund

Prime broker
A bank or broker that provides a hedge fund with financing, securities lending, trade settlement and custody in one relationship. See prime brokers.
Custodian
The institution that holds a fund's securities in safekeeping and settles its trades. See asset custodians.
Seed capital
Early money given to a new fund manager, usually in exchange for a share of fees or revenue. See seed capital for hedge funds.
Clawback
A clause that lets a fund or firm take back pay or profits that later prove not to have been earned. See clawback clauses.
Accredited investor
In the United States, a person or entity meeting income, wealth or professional tests that allow them to invest in private funds. See accredited investors and qualified purchasers.
High-frequency trading
Automated trading that holds positions for very short periods and competes on speed. See the debate over high-frequency trading.

Missing a term? The full list of articles is grouped by topic, and the resources page lists free data and open-source tools.