# Look-ahead bias in backtesting

How can a Pine Script backtest use a daily close before that close exists?

A request.security call to the daily timeframe with lookahead set to barmerge.lookahead_on returns the final daily value on the intraday bars of that day. Requesting close[1] instead returns the previous completed day, which a live chart also had.

Look-ahead bias appears when a backtest uses information that was not available at the moment the trading decision was made.

## Where look-ahead bias comes from

Filling an order at the close of the bar whose close generated the signal. Using an indicator computed on the full series, such as a normalization over the full history. Reading economic data at its reference date when it was published weeks later, or after a revision.

Each of these leaks makes the backtest trade on the future, and the result cannot be reproduced live.

[Screenshot: Publication calendar of a macro series in Stochastly: the delay between each observation date and the date the value was published.](https://stochastly.com/screenshots-current/learn-look-ahead-bias--publication-calendar--ml.png)

Publication calendar of a macro series in Stochastly: the delay between each observation date and the date the value was published.

## The Pine Script case

In TradingView, calling request.security on a higher timeframe with lookahead set to barmerge.lookahead_on returns the final value of the higher-timeframe bar while it is still forming. Historical bars then show values that a live chart did not have.

The usual fix is to request the previous completed value, for example the series shifted by one bar, when lookahead is on.

```
request.security(syminfo.tickerid, "D", close[1], lookahead = barmerge.lookahead_on)
```

“The most reliable approach to achieve non-repainting results is to use an expression argument that only references past bars (e.g., close[1]) while using barmerge.lookahead_on as the lookahead value.”

TradingView (n.d.), [Other timeframes and data](https://www.tradingview.com/pine-script-docs/concepts/other-timeframes-and-data/)

## How to detect it

A causal strategy gives the same output for past bars when future bars are removed. Truncating the series at a cut, replaying the strategy and comparing the overlap bar by bar reveals any output that depended on data after the cut.

Stochastly runs this truncation test before a search and stops the run when past output changes.

“Traders can often confirm whether a strategy has lookahead bias by forward testing it on realtime data, where no known data exists beyond the latest bar.”

TradingView (n.d.), [Strategies](https://www.tradingview.com/pine-script-docs/concepts/strategies/)

## Frequently asked questions

### What is an example of look-ahead bias?

Buying at the open of a day using that day's closing price, or using a quarterly figure at its period end when it was published later.

### Is repainting the same as look-ahead bias?

Repainting indicators change their past values as new data arrives. When a backtest reads those revised values, the result carries look-ahead bias.

### How is look-ahead bias different from survivorship bias?

Look-ahead bias uses future information in time. Survivorship bias uses a universe filtered by the future, such as testing on the assets that still exist today.

## Sources

TradingView (n.d.). [Other timeframes and data](https://www.tradingview.com/pine-script-docs/concepts/other-timeframes-and-data/). Pine Script v6 User Manual, Concepts.

TradingView (n.d.). [Strategies](https://www.tradingview.com/pine-script-docs/concepts/strategies/). Pine Script v6 User Manual, Concepts.

Arnott, Harvey and Markowitz (2019). [A Backtesting Protocol in the Era of Machine Learning](https://doi.org/10.3905/jfds.2019.1.064). Journal of Financial Data Science 1(1), 64-74.

Brown, Goetzmann, Ibbotson and Ross (1992). [Survivorship Bias in Performance Studies](https://doi.org/10.1093/rfs/5.4.553). Review of Financial Studies 5(4), 553-580.

[Primary source for Look-ahead bias in backtesting](https://www.tradingview.com/pine-script-docs/concepts/other-timeframes-and-data/)

## Related

[The causality check in Stochastly](https://stochastly.com/research/causality)

[How to backtest a trading strategy](https://stochastly.com/learn/how-to-backtest-a-trading-strategy)

[Point-in-time data and vintages](https://stochastly.com/learn/point-in-time-data-and-vintages)

[Signal timestamps and execution delay](https://stochastly.com/learn/signal-timestamp-and-execution-delay)

[Export to Pine Script](https://stochastly.com/product/exports)

[Pine Script look-ahead checker](https://stochastly.com/tools/pine-script-lookahead-checker)

---
Source: https://stochastly.com/learn/look-ahead-bias