I’m noticing some issues with the performance of the USD/JPY algos, not lining up with the backtesting trades.
I’m investigating as I type this, but I wanted to leave some notes below in case I forget later.
Notes
For future reference, the 3 MACDH algorithms are running on the 1 hour, which means running them every minute might be disadvantageous. I’m changing the live USD/JPY MACD3H Algo to run once every 15 minutes. This might still not be ideal, but we’ll see how it goes this week.
I’ve enabled the USD/JPY Elephant Bar Live-2 now, and modified all elephant algos to be using close – open rather than high – low on candle ranges.
The backtesting algos were checking if the 5 min close – open was greater than 30 pips for elephant bars. When it was too big, it would skip them. However, in the live demos, it was checking for the high and low range, and then skipping those. This has resulted in skipping a lot more trades, and in turn losing good profits while taking poorer quality trades during those larger trends that would have been taken if we went with the body.
Time will tell whether this is an improvement, but for now all elephant bars will now run on this new system. The new live account with $25 will be watched closely.
Calculating Risk
When calculating the units to use per trade my old formula was not accurate. It was occasionally trying to open trades with over 200k units when the balance offered only about 170k on certain pairs. This was complicated to decipher, but ultimately it comes down to the pair organization itself. When USD comes first, I need to use the following formula:
units = balance * 2 * risk * close / profit_target
if (units / leverage >= balance):
units = balance * leverage
When USD comes second, like with EUR/USD, I need to use this instead:
units = balance * 2 * risk / profit_target
if (units / leverage >= balance):
units = balance * leverage / close
It still seems imperfect, because this may calculate something like 178,441 units, when there are only 178,440 available, due to the bid vs. ask prices that might not be accurate enough in my formula. I suspect they are calculating based on ask and I am using bid everywhere.
Conclusion
To sum it up, I have a few new tasks for the upcoming weeks:
- Follow USD/JPY-Live 2 closely as it gets ready to use the full $50 deposit and is trading on the new candle body formula.
- Check at least twice a day to see if any algos are having rejected orders still based on unit size.
- Compare the backtesting results to the live trading this week on Friday afternoon to ensure things are more lined up now.
- Test out whether 2.4 as a profit multiplier is ideal for both USD/CAD and GBP/USD the way it seems to be for EUR/USD.
That is all for the week of July 24, 2023.