I encountered several wrong values in the historical data, especially when using Minute Resolution.

The values are so frequent and distant from reality that they mess up my backtests.

In order to have more realistic outputs from my backtests I would like to have a way to detect these outliers.

Here is an example of minute data consolidated into 15 minutes bars:

156158_1644044712.jpg

The same data in trading view looks like this:

156158_1644044973.jpg

I would assume the wrong data points are usually (but not always) a High or Low, by the higher chance for those to be wrong being watermarks.

As for now I had to change some of my strategies to somehow avoid using High and Low values, and rather create “my own” consolidated bar High and Low using the minute's min and max Close values.

I was thinking of other solutions including looking at the latest average variance/candle sizes and mark a trade bar as “suspect” if it is outside two standard deviations from it, and deny opening positions whenever we have “suspect” bars as part of the data for our decision. But this of course is too aggressive and makes me miss good opportunities.

Has anybody found a solution or is willing to share their current approach or ideas to solve the problem?

..and, has anybody experience this problem also live?

I will let the thread open for ideas and suggestions :)