Interesting discussion Maciej Janowicz and Antonberlin.

To make the backtesting even more effective we would also have to use out of sample data and perhaps also use a Monte Carlo simulation to backtest different scenarios.

Even using 40 years of backtesting might not be enough as we saw in the LTCM case.

Anyway, writing an article where I properly backtest the strategy would make it very long and not readable to most people.

The goal is just to show what's possible to do nowadays.