I having been building multiple EAs to trade strong trending, sideways trending and ranging markets.
I have been looking at signalling tools on the EA to identify when the market conditions are right for sustainable Edge with any one of the above EAs and their approaches, that is an EA that would (empirically) determine when the market is trending aggressively and switch on the EA with Edge on such trending markets.
My basic approach is to apply each strategy on each currency pair, index, or commodity at the minimum lot size, say 0.01 lots in the case of currencies. Then one can retrieve real time data on the performance of that EA and compute the win rate versus the required win rate for sustainable edge.
One can find the mean PnL (with profits normalised to 1 lot per trade) and Standard Deviation with which to potentially produce an efficient Portfolio of traded instruments; i.e. to select the optimal strategies for the given period based on:
[ul]
[li]Instrument - Forex, Indices, Commodities etc…
[/li][li]Trade Type - Long or Short
[/li][li]EA Market Category - Aggressive Trend, Sideways Trend, Ranging
[/li][/ul]
The premise is. The markets move randomly on the whole and edge in any trading system is created by selectively applying the one approach most suitable for that period.
So the EA or combination of them test the market continuously for the right conditions and then select a portfolio containing the instruments, directions, and market types that offer the best risk adjusted return at any given time.
You are bound to find that one of your EA’s produces better results than the other two. Presumably this will be the one that is kicked into operation by conditions defining the least random of the three market categories. I would suggest this would be the Aggressive Trend category, as this is closest to the opposite of random.
So, as long as you have a sufficiently large number of markets from which to find some that are Aggressively Trending, what’s the point of work on the other two EA’s?
Thanks for the feedback. I guess the market fluctuates between those 3 core categories, either long or short. So at any given time, I do not know which mode it is in. Trading an EA best suited for a sideway ranging market will lose money for me as there will be no edge. The trick in creating that edge would be in classifying the market conditions appropriately and selecting the right EA for the job.
The best way to do this as I can see for now is with empirical observations of the performance of the EAs in question:
If say the EURUSD Long only - Sideway Ranging - EA with an TP:SL 1:1 is performing at a 56% win rate on the last xxx observations, whereas the EURUSD long - Aggressive Trending EA with a TP:SL 3:1 is underperforming the required strike rate.
Clear option 1 is the best at this point in time. I understand your point about perfecting the Trending EA and looking for suitable markets across a wide spectrum. That is what I will have to do initially as I develop a system for all three. But a complete system would exploit opportunities all all three situations.
That’s kind of the golden ticket right there. If you can determine what the next phase of the market will be with decent accuracy (trending, mean reverting, high volatility, low volatility), it doesn’t really matter what your entry strategy is. A regular MA cross strategy will be profitable in trending environments, and a standard RSI strategy will be profitable in mean-reverting markets.
Regime detection/switching is definitely worth reading up on. I’d start with fractal dimension and hidden Markov models and go from there. I find current strategies (including using trading system results) only help with determining what the current environment is and we all know the market can change in an instant; there will always be an element of lag.
I’ve done quite a bit of investigating in this area. I’d be happy to discuss with you in more detail, if you’re interested.
An EA for Sideways Ranging market conditions, r:r of 1:1 and a win rate of 56% just seems to confirm that price movements in such periods are random.
So an Aggressive Trending mode has to have more predictable price movement.
But a 3:1 ratio might just undermine your own good work: there is no TA support for a standard 3:1 or any other ratio, so your profitable exits are therefore random as far as the trend and market are concerned: so naturally your returns will be random too. 3:1 sounds great if it occurs on every winning trade but an essential element of trend following is simply to let the trend run, potentially (though rarely) to much more than 3:1.
Test for statistical significance and get the p-value and you’ll know if it was actually random or not. But knowing just an average win rate isn’t enough to conclude anything.
Yes I would be very interested to hear your thoughts on this. So far I have found the lag in information to be the biggest problem. I’m not sure if Market Portfolio Theory can be the lazy answer to risk allocation in this situation.
I think MPT can be effective once you have determined your “trading action”, and it can help answer the question of “how much”. I have looked into using MPT in this regards only very briefly, so I may have misunderstood.
Perhaps I am late to the party on this thread but it caught my attention with the topic I have been thinking about - classifying market environments.
I saw the mention of Markov models and am curious what other techniques you guys have tried out… what models tend to work well for you?
I have recently began working towards using Gaussian Mixture models to perform classifications … one model for trend and one model for volatility using TA inputs as features… I haven’t gotten to the point where I can test them yet but almost there. Seeing some other traders talk about market classification gives me confidence I am on the right path.
I also plan on building system-specific controls charts (from Production Control subject) that use a rolling window of win percentage of perhaps the last 30 trades or so for each of those systems (trend , range) … assuming this will give me an idea if they are coming into synch with current market conditions or not…
Curious across what time frames you guys have tried application of classifications - with regard to how much they lag ? Any major breakthroughs for you guys in this area?