Stochastic Divergence

[garyfritz]

OK, I added the SL and BE.

It's unfortunately very difficult for me to optimize across all pairs, so I picked a pair (AUDCAD) and looked at how that behaved. If this looks promising I'll see about doing a more careful test.

SL = 12 is much too small. 12 pips produces 48% wins, 1.50 avg trade (without costs), $422 profit in the last 6 months. With costs/slippage it would be breakeven at best.

SL = 25 works better: 64% wins, 2.37 avg trade, $630 profit.

NO SL works best: 69% wins, 4.59 avg trade, $1102 profit.

BE: Again, NO BE works best. 15-25 pip BE works about 10-15% worse than no BE, 30 pip BE is almost as good.

So that was on a pair that works reasonably well. I thought I'd compare the results on a pair that doesn't work so well, so I tried USDJPY. It basically broke even over the last 6 months w/o SL or BE. A 5-20 pip SL helped quite a bit this time, but only in the last few months. The avg trade was still too small to be tradeable. Same with BE.

It may have potential at longer TFs. I tried EURUSD on M10 bars, without SL/BE, and it averaged over $50 per trade in the last 6 months, but it did poorly in the 6 months before that. A 20-pip SL improved the equity curve considerably, averaged about 50% more per trade in the last year that it made without an SL, but the win% dropped from 68% to 32% !! This thing obviously does not like a stoploss -- it must often go well in the hole before turning around and winning. BE stops didn't help.

So, bottom line: the system is marginally profitable without SL/BE on 1min, but the trades are so small that it would be breakeven at best in actual trading. I would not consider this a usable system on 1min bars. Higher TFs may work but it's pretty spotty even there.


BTW somewhere along the line I added a rule to exit the trade when the 128%k crosses the 5 (for buy) / 95 (for sell) line. It's not in your rules doc file and I don't remember where I came up with it. That rule seems to hurt results so I removed it.

[Fx93]

Thanks for testing it Gary. I believe the EMA rules help a bit, but likely not enough to make it much better. I'll shelf it for now, I'm not that keen on no SL.

I've got some other mechanical methods that look promising and I'll post them with a backtest in a new thread.

What EAs do you like best?

Edit: I might look at the same method with higher period stochastics at some time to see if that yields anything.

[garyfritz]

Khalid, what's that file? I'm guessing it might be a config file for MetaCharts but the info in it is encoded...