Skip to content
All library documents

Choosing Adjusted Close Data for Return Modeling

Article Quant Q&A · Author: Eka
Read the original

Summary

The document discusses whether machine learning models built from historical stock prices should use raw closing prices or adjusted closing prices. Adjusted closes account for corporate actions such as splits and dividends, which can make historical returns more comparable and prevent split-related distortions from creating misleading features or signals.

The answers raise a caveat about dividend adjustment: embedding dividends in the price series can create a move that a model predicting daily price changes may not be designed to anticipate. One proposed compromise is to adjust historical prices for splits while keeping dividend payments separate. The discussion offers practical considerations rather than comparative tests, and it does not specify a forecasting target, data-processing convention, or model evaluation procedure; the appropriate series depends on how returns and dividends are represented in the task.

Key ideas

  • Adjusted closing prices account for corporate actions such as splits and dividends.
  • Raw closes can create misleading return measures around corporate actions.
  • Dividend-adjusted prices may introduce changes that complicate daily price-move prediction.
  • A possible approach is to adjust for splits while modeling dividend payments separately.
  • The data choice should match the prediction target and treatment of total returns.

Tags

EquitiesMachine learningBacktestingStatistics
Full text
# Which close price should we use for machine learning?


# Which close price should we use for machine learning?












I am building a machine learning model using historical prices and I am using data from yahoo finance. Currently yahoo finance data have two close prices one normal close price(close) and other adjusted close price(Adj close). My question is which close price should I take for teaching my ML model. Is there any (dis)advantage of using adjusted close price instead of close price?

## Answer by HyperVol (score 2)

https://quant.stackexchange.com/a/25725

You should use Adj close price

Using Adj close price gives you the adjusted values of close price, hence the fair picture in case of off-beat events like splits and dividends. Using close price instead of adj close price provides unrealistic and false values of metrics like returns, which could generate false signals in your ML model.

## Answer by Jacques Joubert (score 0)

https://quant.stackexchange.com/a/44877

Hmm I don't know if I agree that the adjusted price should be used. On the one hand the adjusted prices do help to avoid the impact of corporate actions like stock splits.

On the other hand using the adjusted price incorporates dividends into the price and that will cause a unpredictable shock to price. Lets take an example: you are making a prediction on each days move. On the 4th day a dividend is payed and there are capital gains. Your model wont be able to account for the dividend.

I think the best thing to do is to take the raw data, adjusted it for corporate actions, and not include dividend payments in the price.

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.