RSI calculation

[eugenpro]

Hello. During usage RSI function I've touched that with the same timeperiods e.g. 14 RSI still changes depending on data size. for instance if I calculate RSI from 1 Jan 2023 to 01 Aug 2023 or 1 Jan 2022 to 01 Aug 2023 RSI value on 01 Jun 2023 will be different in both cases that makes me crasy. How to reach RSI cosistensy across the time?

[mrjbq7]

Probably has to do with it's definition in TA-Lib, using smoothing:

https://github.com/TA-Lib/ta-lib/blob/main/src/ta_func/ta_RSI.c#L341

[bttger]

@eugenpro how big is the difference for you?

[mario4tier]

This is how RSI works (and any function designated to have an "unstable period" in TA-Lib).

Because the smoothing is recursive, adding/removing to the length of the input will cause a slightly different output.

This is the nature of how the RSI algo works. There is no fix needed to TA-Lib.

If you want strict consistency, you have to make a choice how to handle it in your app. As an example, you can choose to always start your input on 1 jan 2022 for this time series even if later you have access to the 2021 data.

TA-Lib also offer ways to enforce a minimum of inputs to reduce inconsistency (check for setting the unstable period for individual functions).

[mario4tier]

@bttger the difference depends highly of the input variance and how many data you input to the function. I assume most apps just feed all the data they have, and that is an OK approach to it.