backtesting question regarding Adjusted Close Price

[sbbi]

Hi,

I am trying to do some backtesting on a stock based on its close price. I understand Adjusted Close Price is better for this purpose because it considers dividend and split. Here is my question:

For example, say, the Jan 4th's close price is $10.5, and Jan 3rd's close price is $10, so the daily return was 5% on 1/4. On 2/1, there was a dividend of $0.5 and it is the only dividend between 1/4 and now. As the result, the Adjusted Close Price of 1/4 is $10, and the Adjusted Close Price of 1/3 is $9.5. So if I calculate the daily return based on Adjusted Close Price, it would become 5.26% (i.e. 10/9.5-1). That means the return calculated based on Adjusted Close Price in backtesting deviates from the real world. If my strategy is to sell the stock when it increases more than 5%, the backtesting will sell, but in the reality i will not sell.

I am very confused.. Can anyone enlighten me and share his approaches/opinions?

Thank you!

 

[amnonco]

simply put, you're absolutely right.

to me the most important thing is implementation, so I test the way it'll be in the real world.
with stocks inter day there's a real problem getting the data (unless you pay for it).

on the other hand, you can check how many of the tested days are relevant to the problem.
i mean, if your example happened twice in a sample of 540 days, it means almost nothing.

so get some proportions of the problem relative to the sample size.

 

[sbbi]

thanks. many stocks have quarterly dividend, so the deviation can be significantly large when backtesting more than 1-2 years..

 

[amnonco]

true, but many stocks also have a pretty constant dividend yield, so you can factor that in your backtest every period of your choice.

 

[numbertea]

I am not convinced that over time your backtesting would be affected at all. You can build your system to close with either, but, the adjusted close is not the eod stock price, it is just theoretical so backtesting with it is even further from reality. Backtest to create your system, the data we see will never be the same in the future but the general patterns will be, therefore you could test with either. My opinions.