Adamus
Experienced member
- Messages
- 1,898
- Likes
- 97
more pure mechanical trading research
My process of creating workable trading systems generally follows the following path, as it is following now with NinjaTrader and before with TradeStation and Java.
(1) Find an idea somehow - normally while trying to improve another system. Looking at the price action around the trades in a backtest resultset gives me a ton of ideas, mostly useless but occasionally decent. E.g. yesterday I was looking at GBP-USD 30 min bars and they were going all in one direction for 5 or 6 bars, and then consolidating and chopping around, and then going on another run - sometimes in the same direction, sometimes the opposite. So I figured I needed to see what the average number of consecutive days in one direction was, and when it averages more than 2, I should just enter in that direction after one bar and try to scalp a bar of points, exiting at the end of the bar.
(2) I wrote an indicator to average the number of consecutive bars in the same direction. Long and short.
(3) I tried it out on GBP-USD, EUR-USD and JPY-USD. Useless! It doesn't cover transaction costs which on entries and exits @market could be 4 ticks per round-turn easily compared to the historical data.
(4) Started tweaking it. I made the entries on a limit just below the previous bar's close, and the exits on a bracketed pair either side of the close, so only a 1/3 of the executions would be a stop and so slippage would be less.
(5) Added filters to it, e.g. entering only on the 3rd bar of the run, not the second, or only when a volatility indicator is relatively low.
(6) The more I do this, the higher I get the average profit per trade and the closer to the magic $10 dollars which would make me happy (after transaction costs of $25) - 10 is double figures, you see - but the system makes fewer trades of course. I start off at 18 trades per day but -$20, and by the time I have managed to raise the dollars per trade up to around $0 the number of trades per day has fallen to about 0.1 or 0.2 and the equity curve looks more like a seismograph than a nice account balance.
Plus of course at that point the danger of curve fitting rears it ugly head, so I have to make sure I've got out-of-sample data or other markets to test it on when I think it's ready.
So that's where I am now. I've researched the consecutive trades idea into the ground and I'm looking for the next bright idea, so I figured I'd write about it here and let some inspiration come out of the woodwork while I write. I guess I'll go back to the charts where the consecutive runs idea was actually a bad idea, and see whether I can see a better way of doing it.
My process of creating workable trading systems generally follows the following path, as it is following now with NinjaTrader and before with TradeStation and Java.
(1) Find an idea somehow - normally while trying to improve another system. Looking at the price action around the trades in a backtest resultset gives me a ton of ideas, mostly useless but occasionally decent. E.g. yesterday I was looking at GBP-USD 30 min bars and they were going all in one direction for 5 or 6 bars, and then consolidating and chopping around, and then going on another run - sometimes in the same direction, sometimes the opposite. So I figured I needed to see what the average number of consecutive days in one direction was, and when it averages more than 2, I should just enter in that direction after one bar and try to scalp a bar of points, exiting at the end of the bar.
(2) I wrote an indicator to average the number of consecutive bars in the same direction. Long and short.
(3) I tried it out on GBP-USD, EUR-USD and JPY-USD. Useless! It doesn't cover transaction costs which on entries and exits @market could be 4 ticks per round-turn easily compared to the historical data.
(4) Started tweaking it. I made the entries on a limit just below the previous bar's close, and the exits on a bracketed pair either side of the close, so only a 1/3 of the executions would be a stop and so slippage would be less.
(5) Added filters to it, e.g. entering only on the 3rd bar of the run, not the second, or only when a volatility indicator is relatively low.
(6) The more I do this, the higher I get the average profit per trade and the closer to the magic $10 dollars which would make me happy (after transaction costs of $25) - 10 is double figures, you see - but the system makes fewer trades of course. I start off at 18 trades per day but -$20, and by the time I have managed to raise the dollars per trade up to around $0 the number of trades per day has fallen to about 0.1 or 0.2 and the equity curve looks more like a seismograph than a nice account balance.
Plus of course at that point the danger of curve fitting rears it ugly head, so I have to make sure I've got out-of-sample data or other markets to test it on when I think it's ready.
So that's where I am now. I've researched the consecutive trades idea into the ground and I'm looking for the next bright idea, so I figured I'd write about it here and let some inspiration come out of the woodwork while I write. I guess I'll go back to the charts where the consecutive runs idea was actually a bad idea, and see whether I can see a better way of doing it.