ДалееСледующее руководство
Стратегии «один против остальных» и «один против одного»
Технический
Техническое РУКОВОДСТВО
Бэктест моделирует, как торговая стратегия вела бы себя на исторических данных при заданных допущениях; это не реальные результаты инвестирования.
Прослушайте это руководствоОзвучено ИИ · ~3 min
Trying many variations and selecting the best result can overfit the history, so evaluation must account for data leakage, costs, selection and uncertainty.
A backtest replays a trading rule against historical prices or other market data to estimate how it might have behaved. It is useful for debugging, comparing hypotheses and examining drawdowns, but the researcher chooses the strategy and assumptions after seeing some of the same history. This creates opportunity for look-ahead bias, survivorship bias, data snooping, unrealistic execution and parameter tuning. A strong in-sample result can disappear when costs, delays or later market conditions are included. Bailey and coauthors analyze the probability of backtest overfitting and explain why ordinary holdout methods can be unreliable when many investment configurations are tested. Record all trials, reserve genuinely untouched evaluation periods where possible, use time-aware methods, and estimate performance after fees, slippage, liquidity limits and operational constraints. A later test is not fully independent if choices were repeatedly changed after inspecting it. Avoid using future information, and check whether the historical universe includes assets that later disappeared. Report the test window, data source, parameter-selection process, costs, comparison baseline and uncertainty. If results are advertised, U.S. SEC investment-adviser marketing rules impose conditions on hypothetical performance, including information about assumptions and the audience; applicability depends on the communication and adviser. Backtests are not proof of future returns and do not guarantee that a strategy can be implemented. This guide is educational, not investment advice. The estimate is conditional on the data and assumptions used.
A researcher freezes a strategy before testing it on a later period that was not used to tune parameters.
A backtest includes transaction costs, slippage and realistic position constraints instead of assuming free execution.
An analyst records every strategy variant tried before reporting the best historical Sharpe ratio.
An adviser labels hypothetical performance and provides the assumptions and limitations required for its intended audience.
Markets and trading infrastructure change, making historical results fragile when strategy logic or costs shift. Researchers continue to develop methods for assessing selection bias and robustness, but no diagnostic certifies future profitability. Keep research records complete, evaluate realistic implementation constraints and treat hypothetical results carefully when communicating them. Re-test when the data universe, execution venue or assumptions change. Historical markets, instruments and execution venues change. Review a strategy’s capacity, data provenance, fees and drawdowns before relying on simulations. If hypothetical results are shared with clients, follow applicable disclosure and audience requirements. No validation metric removes investment risk.
Бесплатная рассылка
Три проверенных AI-истории каждое утро буднего дня, написанные простым английским. Бесплатно навсегда, без рекламы.
Одно электронное письмо каждый будний день. Отписаться в один клик. Мы никогда не продаем и не передаем ваш адрес.
Проверьте себя
Мгновенная обратная связь по каждому ответу и общий сертификат с проверяемым идентификатором после прохождения курса.
Поддержите бесплатное образование в области искусственного интеллекта. AI Understanding — это некоммерческая организация 501(c)(3). Никакой рекламы и платного доступа. Сделать пожертвование
A backtest simulates how a trading strategy would have behaved on historical data under stated assumptions; it is not live investment performance. Trying many variations and selecting the best result can overfit the history, so evaluation must account for data leakage, costs, selection and uncertainty.
The guide defines a backtest as a historical simulation under stated assumptions.
The guide explains that selecting a winner from many trials can overfit historical noise.
The guide lists transaction costs, slippage, liquidity and implementation constraints.
The paper discusses limits of ordinary holdout when many investment configurations are tested.
The guide recommends recording all trials, not only the winner.
Продолжайте учиться
Другие руководства, выбранные по этой теме
ДалееСледующее руководство
Стратегии «один против остальных» и «один против одного»
Технический