How Backward Price Adjustments Can Create Look-Ahead Bias

Summary

The document contrasts backward-adjusted and forward-adjusted stock price histories, focusing on corporate actions and look-ahead bias. Backward adjustment uses the full history so that the latest adjusted price matches the latest unadjusted price. As a result, observations before a corporate action are rescaled using information from that later event, and historical values can differ from the prices that were available at the time. This can introduce look-ahead bias in a backtest.

Forward adjustment instead keeps the series aligned with unadjusted prices at the start; later corporate actions cause adjusted values to diverge from unadjusted prices, but the response says this avoids the same look-ahead issue. The answer refers readers to a discussion by Ernest Chan and describes the point as straightforward, while noting uncertainty about whether a dedicated paper covers it. It gives no empirical test or detailed treatment of how vendors implement adjustments, so researchers should verify the adjustment conventions in their own data.

Key ideas

• Backward adjustment can revise pre-event prices using later corporate action information.
• A backtest using backward-adjusted historical data may therefore contain look-ahead bias.
• Forward adjustment aligns prices at the start and can avoid that specific bias.
• Adjusted series diverge from unadjusted prices around corporate actions, with the direction depending on the adjustment convention.
• Data vendors may differ, so the stated distinction does not replace checking a dataset's methodology.

Tags

# forward- and backward adjusting stockprices


# forward- and backward adjusting stockprices












Do you guys know if a paper has been published that discusses forward- and backward adjusted stockprices, and the look ahead bias coming from backward adjusted data?

## Answer by Richard Harrison (score 6)

https://quant.stackexchange.com/a/65974

Ernest Chan talks about how backward adjusted prices induce a look-ahead bias in the comments of one of his blog posts, Beware of Low Frequency Data. Scroll down in the comments to when he's responding to Samuel. Essentially when using backward adjusted prices, you must look at the entire dataset and backward adjust everything so that the latest adjusted price will match the latest true price but any price in the adjusted series before a corporate action will deviate from the unadjusted price. With a forward adjusted price series, the unadjusted and adjusted prices will match at the beginning but then after any corporate action, the adjusted and unadjusted will deviate so the latest prices will be different but there's no look-ahead bias. I'm not sure of any paper discussing this but what I've said is basically all there is to it.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.