Please humor me - auto trading forex techniques with 100% monthly consistency
I have been using auto trade systems for six months now - I have been happy with the performance I have achieved, but having said that I'm yet to find a signal/strategy that has backtested (using full tick data, so bar reconstruction/magnification) with zero or near zero net monthly losses, as in every month showing a profit no matter how small. Now I know this would not be the holy grail of course, because such systems tend to have small profits overall and that on an annual basis, monthly drawdowns can be near irrelevant and indeed most of the signals I run (3 of them) do have irregular monthly losses. However, I have nearly managed it with parabolics on 30min - 4 hr timeframes (May 2012 ruined the whole thing ). I wondered if anyone from experience knew of a general system/indicator combination that had not (yet) produced a monthly loss over recent 5 year data?
I would like to reiterate I am not searching for the holy grail here or chasing dreams - I just want to achieve it out of mathematical curiosity without cheating (i.e. not using a rarely trading system).
I certainly haven't created such a system - that much I can tell you. I have been developing automated fx systems for 12 months now and have made all the usual mistakes that I'm sure nearly everybody makes along the way (e.g. not understanding the differences between backtesting and trading live, dangers of curve fitting etc). I'm only now beginning to experience some success trading my strategies live.
May I ask are you trading any of the automatic systems you have created live? Or are you still at the backtest stage? I wonder because, if you're still at the backtest stage, my advice would be to start trading them on sim or live with low stakes and see if your forward tested results match your backtest results. It's just SO easy to create a strategy that backtests well but just falls apart live.
Thanks - there are always tricky months where a system that profits in others, struggles in those, such as the usual August lull (though 2012 was ok), too many flat days in December etc. Generally ranging is only profitable for manual traders as I'm yet to see a 12/24 hr automated system that can really cope with this and make a regular return since they're so reliant on bigger moves to make annual profits.
I've been manually trading for some years now so always take my systems live rather than simply relying on backtesting. Having said that, I have found the bar magnifier in MC to be a life saver for debunking intra bar 'HFT' style trading (always a disaster unless you pay zero commission).
Problems I have with live auto trading psychology (bit of a paradox...) would be that if you have a good run up in a month and it surpasses the mean total profit you expect to return per month, whether to cease trading that month or go on and try to make more in order to account for less successful months, the problem with this being that it could go on to give back. Unfortunately I don't think the modern world has been in existence for enough years to come to any kind of empirical answer on this, there are months that start bad and end bad, which would prove you should hold on each month to contra previous losses and wait for the good times, but then there are equal months that run up to the average and finish flat or down. Bah!
I do perform walk forwards, but I'm yet to see as much use in using out of sample data as with manually checking months with drawdowns, optimising for those, re-applying to the overall, etc. Perhaps I don't understand them well enough.
I run an exhaustive optimisation every fortnight (which takes forever) and I alter my overall 'market view' as a result of that. For example, entries are skewed to long on the EURJPY right now and have done nicely in January. I am however, unconvinced of the merits of doing this and I generally prefer having equal entry and exit parameters, especially since it can sometimes cause funky reversals in the package I use (Multicharts). I think a consolidation month will hit me regardless of how in tune with recent market movement the system is. Regular optimization MAY increase the profit in +ve months though, but I don't see how I could test this as again the markets haven't been in existence long enough for my satisfaction. I suppose I could start with 3 years data for equal entry and exit params, then walk through 24 months after those 3 years adjusting month by month and see if those 2 years had increased profits. My feeling is it would only make a small difference and there is some comfort in being as willing to long as short a market you feel is near enough random (as I have no interest or ability in fundamentals, macro interpretation of news etc).
Walk forward optimizations are useful for seeing if your algo has any sort of predictive power, but I'm not 100% on board with the idea that it is best practice to reoptimize your parameters every month (or whaterver interval) so they best reflect the current state of the market. I feel like this approach will overly curve fit your strategy to ONLY work in recent market conditions and this leaves you wide open to getting hit hard by a sudden change in market behaviour (which happen often).
So, whilst I use walk forward optimisations during my algo development, the parameter set I go live with is actually determined by an optimisation over my entire data set (usually at least 6 years of data which produces 500-1000 trades during a backtest). If my strategy is unable to handle the multitude of market conditions over that time period without acceptable performance, then I go back to the drawing board and try and improve my strategy.
As I understand it, this is not the methodology of most autotraders on this forum, but for me it gives me much more confidence in the ability of my strategy to endure a range of market conditions and still be profitable.
I 100% agree - I only adjust my parameters based on the aggregate data, not just the recent market conditions, though I do give SOME weight to them, so if they suddenly change I will get hit more than if I hadn't, but the underlying system still produced a profit in 10 out of 12 mths with minimal DD.
I'd rather go live with a set that didn't give the max profit in the past 3 months, but gave a profit and also handled the past 4 years including the summer and December lulls. I can't be confident with anything less.
Are you trading spot forex or currency futures? I'm in the spot market, I currently have my auto strategy running across a fairly wide range of pairs. I designed it primarily against the EURUSD, but it works on most pairs with some modifications to parameters.
Spot - I've never traded futures except for cash indices in my life so stick to what I know and the feeds that I'm used to. My favourite pairs are EURJPY and cable right now because currently I use a method that isn't reliant on MAs. I also like to think (or lie to myself) that cable is a good hedge for EJ, i.e. it might move while EJ doesn't and vice versa. EJ isn't very noisy so you could trade that with MAs, but cable is a nightmare with that 'big bar' ranging effect that busts up MA methods so well. The problem I have with the EURUSD is trying to be too smart for my own good... because ECN spreads on this pair are so tiny, I keep trying to make my own mini HFT and trade it on low ticks during the London - New York session. I keep failing. Honestly, if it I had a wide spread, I probably would have made money from it as just another long term auto pair. I'm just drawn in by the idea of making x money each day. Fool.
Also the psychology of auto trading can't be overlooked... I'm way ahead of my monthly average in January due to all that's happened in the Yen. I use the 15 min timeframe and moves like the ones in Jan are an absolute dream. Another big day today with a large daily drop and a profitable short after a hugely successful long. Every bone in my body wants to exit that short now and call it a month. But the mathematics of my backtesting are only confirmed for being always in the whole month, every month. Argh! I was always quite good at not giving back manually so it's a killer. When the consolidation comes, I'll be in a bad, bad mood.