Results discrepancy for same Formula using 5min, 1min, and 15sec Intervals

I have an Formula that gets VERY different results for the EXACT same
formula given different time intervals.

I have run this Formula using 5min, 1min, and 15 sec intervals and its
performance varies considerably depending on the interval.

One consistency I have found is that as the time interval gets smaller, performance
of the algorithm decreases - in some cases, significantly.

One thing about this algorithm is that it buys stocks at the market OPEN and the results
are VERY SENSITIVE to this price given the different time intervals.

Has anybody experienced this? If so, which interval, in practice, did they find to be
the most accurate (i.e., the 5min, 1min, or 15sec interval)?

Any direction and suggestions would be appreciated.

The interval that returned the highest profit :slight_smile:

@pmfiorini, there's really no way to answer your question when you provide no details of your formula and no details of the performance decrease. Perhaps applying your rules to shorter time intervals decreases overall exposure which often leads to lower CAR. Or perhaps the shorter time intervals lead you to exit trades too quickly, decreasing your gain per trade. Or perhaps it's one of dozens of other possible explanations.

You really just need to spend some time with the trade lists and backtest report metrics from each timeframe that you would be willing and able to trade in real time. For example, there's no point in testing 15s bars if you don't intend to fully automate your strategy. Find the underlying reasons for the performance differences, and that should give you some direction on which timeframe to trade and possibly ways to modify the strategy to work better in different timeframes.

Different input equals different output.

Say 50 DAY MA would give completely different results than 50 MINUTE MA. This is not discrepancy. That is obvious and expected result.

Large discrepancies? That is expected? Really?

Thanks for your time and answer. This is useful.