Data integrity issues

When I switch between AFL files, I cannot replicate results. As an example, I'll load AFL script 1 and run a backtest. Then I'll load AFL script 2 and run a backtest. I'll go back to AFL script 1 and get different results from the first time I ran AFL script 1 with the exact same parameters on the exact same database. The data somehow gets skewed. I'm running DTN IQFeed intraday with 5 minute bars with mixed EOD/Intraday data. What more information is needed, if any, to answer how I can resolve this? Thanks.

That's not a lot of details Ed.

So things to look for are things like, is this during trading hours? From my previous experience with the IQ feed, albeit long ago was that bad prints would come through and then later get fixed. Another is check how and when corporate events like splits, dividends are handled both at EOD and intra-day. Another thing that comes to mind is that if the price history length has changed between back-tests, and at the end of the test, it assumes closing open positions at end of data history, then they would not match just on that alone. You also have a chicken or the end situation where AB gets it's last EOD bar has completed versus intra-day data. In general terms with respect to data feeds and their personal judgement of what is "Included", if you were to study the NYSE TAQ, you would understand the many flags and filtering switches involved for quote and trade inclusion, some providers are clueless and don't even have consistency across their numerous servers, where you can get different data depending on what server you connect to and what resolution or trading session(s) you are including. Just some thoughts.

When you are connected to IQFeed real time data source, the data CHANGE. There is REAL-TIME data fed constantly AND on-going backfills that might include corrections that IQfeed adds AFTER THE FACT (when user report data errors).

There can be other reasons too, such as use of static variables, composites or anything that can store persistent state and that state affecting backtest results due to how formula is written (using state from previous run) - as explained in the KB.

Please read the Knowledge Base:

It is strongly advised to read entire Knowledge Base. Believe or not, but most questions that you might ever have are already answered.