Thank you for your advice @Nicolas et @Vonasi
I hope I won’t make too many mistakes because English is not my natural language.

With no optimization, and setting the starting date at the beginning of January, we see that the robots had similar performance in terms of yield and drawdown, even during the crash of March.
I will do more testing on different timeframe and different assets.

And I will see how it behaves in live testing (with simulation account first). I will not use algos with a drawdown greater than 10% (eliminatory with most prop trading companies) and / or a squid ratio less than 3 (Paul Tudor Jones considers that a good trader must have a squid ratio greater than 2.5 and a drawdown whose duration between the 2 peaks (and not only between the trough and the peak) is less than 4 months.

The only optimization I had done was the SMA20 Daily.
As this is not a criteria I use in my manual trading, I tried the SMA 15/ 20/ 25 and 30.
In most cases it was the 20 or 30 that gave the best results. I chose the SMA20 which is more used by traders and is more likely to give robust strategies than an SMA30.

Too much optimization can generate nice equity curves in backtest / walk forward, but the actual results of data mining based strategies are in 99% of the cases disappointing.
I prefer not to do optimization if these are strategies I already use in manual trading. Why change something that works at the risk of creating a losing strategy?

For my new strategies I mainly optimize the triggers. I use python-coded genetic algorithms to choose the best entry, filter, exit and risk management triggers (keep only the strategies that look strongest and most robust with IS / OOS, walf forward analysis , Monte Carlo simulation, different departure date, different timeframe, different assets …)

I hope that with PRT we will soon be able to use genetic algorithms and
Monte Carlo simulations. 😉

NO-OPTIMISATION_JAN-2020.png

I have also tested on other assets and other TFs.
Here are 3 results on a bond future (Bund), a commodity future (oil) and a cryptocurrency future (Bitcoin).

1. The Bund with the 15min Timeframe breakout strategy is + 37% in 1 year when its benchmark is + 3%. The equity curve is not subject to the erratic fluctuations of the benchmark prices.

2. Bitcoin with an H1 squeeze strategy makes + 145% return with -40% drawdown. Using the strategy that for purchases we get a return of +148 with a drawdown of only -10% (Calmar ratio of 145). During this period Bitcoin experienced a drawdown of 84%

3. Oil with the mean reversing strategy (modified) is + 1025% with a drawdown of 14% (the PRT report notes -6.78% of Drawdown but on the graph we see that the DD is 14%)
The Calmar ratio is 73!!

These results were achieved with an average risk of 1% per trade.

For the mean reversion on oil, I modified a trigger in the strategy that generates frequent small losses and very large gains.

BUND-37-3-pour-le-bench.png BTC-145-et-10-avec-SQUEEZE.png BRENT-1025-et-6.7-BIS.png BRENT-1-million-with-1-per-trade-and-6-DD.png