Update May 12 2021
Thought I'd look at how this algorithm has done since publication in 2016. Annualized return was 12.4% compared with buy-hold annualized return of 3.9%.
You would probably have re-parameterized or seeked for a new algorithm over that timeframe, but its nice to know the original did OK.
This algorithm continues to perform well, outperforming buy-hold by 30.4% annualized over the last 10 months:
Notice this had a very low drawdown (1.2%).