Okay, been backtesting my system some more, and refining it here and there (anyone else notice their systems do better when getting simpler? 
). Thing is, I've hit a snag that's bothering me.
My system appears to randomly break for a month, then hop back into profitability as if nothing happens. Losses aren't catastrophic, and are made back over time, but... it makes me worry. I've got 6 months good quality data, and have been optimising on the first 3, testing on the second 3 (trading on 3 minute bars). Over 3 month periods, it works great. However, doing a by month breakdown, it just seems to explode for a bit, then pull itself together and keep going.
If I work on smaller timeframes, optimise on two weeks at a time and then test on the following week, this appears to not be such a big issue, so I think it's a change in the nature of the market which my algorithm is too simple to handle currently.
I suppose I'm just tired and looking for reassurance that as long as it continues working long term, relatively short term hiccups aren't a major issue. Given that the algorithm seems to work or fail in patches, paper trading it when it slips out of profitibility for a while, until it starts making a profit again, should also reduce the impact of this.
Thoughts?
Edit: Worth noting, the month it breaks on moves around depending on what I'm testing against. This a currency- focused trading strategy, and I'm using the same parameters optimised against multiple pairs to reduce the influence of curve fitting. The two most promising so far are GBP/USD (nice and sane) and EUR/AUD (err...). It does fantastically on EUR/NZD and GBP/JPY, but like hell am I trying to run an automated trader on those two.
). Thing is, I've hit a snag that's bothering me.My system appears to randomly break for a month, then hop back into profitability as if nothing happens. Losses aren't catastrophic, and are made back over time, but... it makes me worry. I've got 6 months good quality data, and have been optimising on the first 3, testing on the second 3 (trading on 3 minute bars). Over 3 month periods, it works great. However, doing a by month breakdown, it just seems to explode for a bit, then pull itself together and keep going.
If I work on smaller timeframes, optimise on two weeks at a time and then test on the following week, this appears to not be such a big issue, so I think it's a change in the nature of the market which my algorithm is too simple to handle currently.
I suppose I'm just tired and looking for reassurance that as long as it continues working long term, relatively short term hiccups aren't a major issue. Given that the algorithm seems to work or fail in patches, paper trading it when it slips out of profitibility for a while, until it starts making a profit again, should also reduce the impact of this.
Thoughts?
Edit: Worth noting, the month it breaks on moves around depending on what I'm testing against. This a currency- focused trading strategy, and I'm using the same parameters optimised against multiple pairs to reduce the influence of curve fitting. The two most promising so far are GBP/USD (nice and sane) and EUR/AUD (err...). It does fantastically on EUR/NZD and GBP/JPY, but like hell am I trying to run an automated trader on those two.
Last edited:
