In backtesting a DAILY strategy (entry and exit on the SAME DAY), would it be more accurate to use historical "adjusted close" values or simply the actual "close" values for my data backfill? Thanks for any input.
Think about it. If you actually traded then you would get filled at actual prices, not adjusted. IMO It only makes sense to adjust for splits.
I agree...But data providers typically provide adjusted. Thanks though, I thought I was missing something obvious.
Most data vendors I know provide the actual prices and ways to link the various series. They are not resposnsible if users don't know what they are doing, which is the case with 95% of backtesting results.
If the entry and exit is in the same day don't use the adjusted close. Since your holding time is entirely intraday, you don't need to adjust for splits or dividends. Also, don't assume you'll be able to get the closing price for the day on an illiquid stock. In real trading it is fairly easy to get the "real" opening price in most stocks without affecting the opening auction price if your order is small enough, but the only way to get the closing price is with a MOC order. At least in my experience with fairly illiquid stocks, the executions are very poor with MOC orders. I bet MOC is fine for something like SPY though. Also, the opening and closing prices from most free data vendors such as finance.yahoo.com are often not the correct opening and closing price....especially for the opening price. The closing price is usually correct I think.
Using Adjusted Close is simpler, but less accurate. I wrote the back-adjuster/stock-adjuster used by CSI. If you want to develop a trading system, then use adjusted data to validate the basic trading rules. Once you are confident in your system, then, to validate your system in a more realistic way, write your own back testing environment so that you can account for dividends/splits (reinvest/withdraw) and futures/options rolling (with slippage) in the manner you think is more accurate. Most people would rather start paper-trading based on the less accurate simulated results followed by real trading.
It depends on your system I think. Right now I'm following a rotation system in which I allocate my funds for a fixed period of time to a certain ETF, with an en entry @ the close of the day and an exit @ the close of the day, several weeks later. I find adjusted closes more appopriate for the backtesting.
You are correct in that it is less accurate but why are you saying it is simpler? I don't get that. I ran accross this problem recently and I was stunned to find out that dividend adjustments distort price series beyond recognition as it is shown in the animated gif in this blog http://www.priceactionlab.com/Blog/...d-backtesting-part-i-close-vs-adjusted-close/ Why backtesting at all using distorted series? I don't see the point.