I used the tensorflow time series forecasting from this page: https://www.tensorflow.org/tutorials/structured_data/time_series with a few other datasets. It is working fine, except for the multi-step and auto recursive models. With the tensorflow example dataset (jena_climate_2009_2016), I get T(deg) predictions like this:
With all stock OHLC times series I tried, I get Close predictions like this:
The (val) loss are similar in each case.
I can not figure out why the predictions are so "linear" with OHLC datasets.