We tested whether automatically switching settings according to the market would make you win using five months of signals (positive in 20% of 576 cases).
Hello, I am Nico-tama.
Recently I published the August sign records for my own indicator. Even applying the same settings, the win rate varies quite a bit by currency pair and timeframe. The 1-hour GBPUSD was 70.6%, and the 1-hour EURUSD was 47.1%.
Looking at this, I thought, why not automatically select only the combinations that are doing well? The market changes, so the settings should adapt to the market. I think this is a common idea.
I actually built it and tested it with five months of data. The results were not very good. Since I have the opportunity, I’ll write it exactly as is.
■ What I made
It is a program that runs on MT4. It monitors 10 combinations in total: five currency pairs and two timeframes, from a single chart.
For each, I score simple settings ①–⑥ based on the most recent outcomes, using only the better-performing ones. The 5-minute timeframe is re-scored every 6 hours, and the 1-hour timeframe every day, so as the market changes, the settings and even the currency pairs used will automatically switch.
In scoring, I did not use the win rate as-is; I used the lower bound of the 95% confidence interval. Since the sample size is small, this approach automatically pushes less frequent combinations to the back.
■ How I validated it
This is the most important part: I chose settings using only the information known at that moment.
If you look later, you’ll see that “August’s good one was the 1-hour GBPUSD.” If you apply that to August, you’ll obviously get good results. To prevent that, at each point I select settings only from earlier records, and count the results based on signals that appear after that.
The data covers May to September 2026.
■ Results
| Method | Occurrences | Win rate | Profit/Loss per 10,000 yen per trade |
|---|---|---|---|
| Automatically switch (best-performing settings) | 227 | 59.5% | +147,400 yen |
| Automatically switch (initial settings) | 202 | 55.4% | −12,700 yen |
| Fix with August’s good settings | 813 | 55.7% | −37,200 yen |
| Fix with all combinations to setting ⑤ | 756 | 54.8% | −169,200 yen |
| Fix with all combinations to setting ⑥ | 1,198 | 51.2% | −1,030,900 yen |
The line at the very top showing plus 147,000 yen is from picking the best-performing results among various parameter trials. In other words, this number itself is hindsight, and there is no guarantee it will apply to the future.
In fact, with the initial settings, I had a loss of 27,000 yen. Then after trying 576 parameter combinations, the gain occurred in only20% of cases. The median profit/loss was minus 83,000 yen.
For comparison, there is also a result where the August-good settings are fixed for the entire period. This shows a loss of 37,000 yen. Automatic switching is somewhat better, but neither result is something to be proud of.