Quote from austinp:
Excellent question, not really one good umbrella answer. I'll say it this way: most systems are written by traders who already have a profitable edge identified. From there they meld it into language a computer can understand and not negate the edge due to discretionary factors.
That's the really hard part, imo.
If you are focusing on FX trading alone, I would suggest working with strategies based on volatility breakout approach. Spend your time back-testing data on periods of time in the past, optimize that data and then apply it to the last two years or so. Walked forward, virtually.
Do not always pick the most optimal settings... select your criteria from the middle of a positive curve of settings. In other words, a total outlier will always be the best performing dataset from any test. But that's a siren song. You need to find the fattest curve of acceptable data results to work with... i.e. smoothest curve of performance.
I have written volatility breakout systems for GBPUSD (best pair) and EURJPY, EURUSD that are all profitable and robust thru the past four years untouched. Sometimes they have monstrously profitable months. Sometimes they have very painful months. Overall they are solid performers if given time, enough capital and free rein to run.
One last suggestion: create a family of systems with somewhat different parameters for different symbol pairs. Run them as a basket, spread your capital across two to four semi-correlated systems. That should result in the smoothest overall performance curve possible.
It may take you years to get where you want to be in system building. Not an easy path, but there are no easy paths in our profession. That's why the reward at end is unlimited... journey is hard but the reward is infinite potential.
Thanks again. The three systems I've written so far have all been systems identified by the traders I'm working with as potentially profitable strategies. All three have shown to be "maybe" profitable sometimes. However, I'm not sure if the things stopping them from being profitable enough for me to run are:
1. My expectations of profitability may be too high. I think right now that if a system is making me, say, 50 pips per month profit in forward testing then that's too close to unprofitability. I'd like to find something with more buffer between where I am and break-even. I just have no idea what's reasonable. I don't know if people have automated systems out there that are pulling in 2000 pips a month profit, or whether that's just ridiculous. Then of course there's drawdown and predictability to add in there.
2. I need better backtesting and optimization tools. My current testing methodology is extremely tedious and time consuming. My algorithms may very well be fundamentally sound but without the right input data sets I won't have a winning strategy. However with my current set of tools it will take me months and months to get an answer to that. This is why I'm on the verge of writing/acquiring better tools but that in itself is an endeavour that is pointless if there's nothing behind all this.
3. Maybe my systems are no good.
I agree about not choosing the optimal settings. My plan was to take several years of data and backtest and walk-forward test through it to see if I can figure out if an input set that backtests profitably and forward tests profitably has any differences in characteristics than an input set that backtests profitably but does not forward test profitably. If option (a) is a backtest that shows 200% profit in a month but a forward test that loses 5%, and option (b) is a backtest that shows 15% profit in a month and a forward test that gains 5%, option (b) is the obvious choice. I just have to figure out if there's a way of statistically differentiating between the two backtests beforehand.
I've been given another system by my trading partners that has been shown in backtesting to be consistently profitable over the last 2 years. OK, ranging from mildly profitable to very profitable, depending on the month. In live forward testing it also seems to be working quite well. So my next step will be to automate that, then to thoroughly test it. I'm TRYING to not just shoot around in the dark as a programmer hoping to hit profitability, but instead rely on people who have actual trading experience. Between the two of them they have over 20 years of experience in this market, so even though it hasn't turned either of them into a millionaire, they still have good experience.
One last question, and it's sort of personal and maybe impossible to answer. What are people with automated trading systems getting in terms of average monthly or yearly returns? I don't know whether I should be ecstatic to receive 2% per month, or bummed if I don't get at least 20%. I'd start with a small account balance ($1000-5000) and probably risk around 2.5% per trade, +- depending on my calculations of risk for a particular trade. Of course I'd love it if I made millions and millions of dollars, but at least I'd like to see this being worth the effort I put into it vs. just spending more time on my other job. I just want to be more diversified financially, at least. If I get a system that over a couple years is outperforming what I'd expect to get from my other investments, I don't have a problem putting more money into my trading account. I'd just want to see some solid predictability first. Anyway, if anyone has any comments they feel to share about this, either on the forum or via PM, I'd greatly appreciate it.
Thanks again,
- Andrew.
So computers help,but theyre not the key to Fort KNOX.
