Try backtesting it for 1 year exactly.
Although I gotta say, in some results it seems a bit weird to me too. I feel like annual return should have been more. Jesse uses this package for calculating annual return:
Which is developed by Quantopian and I assume it's well-tested.