Trade Management for 2 Trading Schools (Continuation Vs Mean-Reverting:)

[palm 10]

Hi All,

I am now trading continuation systems and would like to add a mean-reverting system to compliment my portfolio. I compare the returns distribution of the two systems below. (pls see attachment)

 

The red curve is for MR while the green one is for CON systems. Both have positive expectations. You can see that the big chunk of returns is cluttered at small gains level for MR. My question is on exit strategies. Given these two systems, how would you design exit strategies differently.

 

Exit strategies issues are

1.) Stop-loss

-(Fix or volatility-based)

-(Tight or loose)

2.) Profit-target

-To have or not to have

-Tight-loose

3.) Time-Exit

-To have or not to have

 

Anyone has experience dealing with MR type systems?

Any other thoughts are welcomed. Thank you

[zdo 20]

The terms MR and CON alone are still too general to give specific answers to your questions.

 

A precision MR system, (for example one with a high hit rate that is only implemented under certain market types and built around patterns ‘in’ a proprietary envelope) would use tight ‘volatility’ based stops, would have tight (% of ‘reversion’) targets, and most likely would test better with timed exits (and with restrictions on time allowed to elapse between trigger and entry/fill)

 

The overall optimum for less ‘presise’ MR systems, like those built on over bought or sold ‘indicators’ , etc, etc. or high sigmas, etc etc. is harder to specify generally. Stop placements test all over the board and a few test best with seemingly ridiculous stop distances.

Almost all MR systems test out best with targets. That is the closest we can come to making a generalized answer to any of your questions. For looser systems, time exits are generally not indicated… Exploring the nature of the ‘central tendency’ you are using is one way to ferret out answers to some of these questions.

 

With CON, stops can test out all over the board. With CON targets, you generally want to stay in as long as possible. But we have to be careful with that generalization because certain CON systems test out better with ‘percent of margin’ targets…

 

Hopefully this will give you some good ‘first guess’ starting places for your testing

[palm 10]

Thank you, Zdo

I only tested CON systems in the past and profit targets always worsen the system performance Even Con systems with time-exit.

 

Since last post, i came back and test profit-target on my new MR (plain RSI with price starting to reverse). And at Stop:Profit ratio of around 1:3 to 1:4 shows improvements in the system. So thanks for your comments.

 

BTW, when you said precision systems, you mean %win >60%? My system's %win dropped to 35% after applying tight stops. I know this is a personal preference, but in your opinion, an MR system with win rate this low is going to fly?

 

Thanks

Palm

[zdo 20]

palm,

 

re: “when you said precision systems, you mean %win >60%?”

My threshold for that term ‘precision’ would start at %win >80%. Without market typing to turn the implementation of such systems on and off, most of these way high hit rate systems tend to be near a wash when costs are included.

re: “ I know this is a personal preference, but in your opinion, an MR system with win rate this low is going to fly?” One of my biggest trading ‘challenges’ to overcome was to learn to ignore my personal preferences and go close to the testing. The system’s expectancy, as mean R-multiple, is much more important than hit rate.

 

Are your CON and MR systems correlated?

 

All the best,

 

zdo

[palm 10]

80% win is unheard-of for me. To achieve that, you are using filters, right? Do you use a lot of filters (such as volume, volatility, etc...) or just a few but only trade on extreme values. (for example, use only volatility filter, but enter only at 3 standard deviation conditions.

 

Also, are MR systems you are talking about traded on relative values or outright directional. (Pair trading Vs buy on price retracement)

 

Also, with 80% rates, the number of trades will be quite low, right? I'm having a hard time trading-off between number of trades and expected profit/trades. I mean i have no good way of balancing the 2.

 

They are inversely correlated. This is the reason i try to develop MR systems. My current portfolio is packed with trend following/break-out systems. I only trade one market, so i have to make the most out of as many market conditions as possible.

 

Thanks again Zdo. Big help for me.

[zdo 20]

Re “80% win is unheard-of for me. To achieve that, you are using filters, right?”

Backing up a little bit - my whole approach is sort of ‘filter’ first. I call it MarketTyping and it ‘types the auctions’ and conditionally gives fuzzy allocation weighting and sizes to each member of a portfolio of systems. This MarketTyping was consistent with my original visions of trading and after several frustrating regressions with the typical approaches, I made the ‘never go back’ choice. It’s completely bassackwards logic to the vast majority of system development… but it was the only way I personally could figure out how to both survive and thrive intraday and to automate trading.

 

Re “Do you use a lot of filters (such as volume, volatility, etc...)”

No. However, typical ‘volatility’ is a small aspect of the MarketTyping. Related to ‘volatility’ and more important in the typing is measures of pressure and pressure releases, etc.

 

Re “or just a few but only trade on extreme values. (for example, use only volatility filter, but enter only at 3 standard deviation conditions.”

The only system I run with a +80% hit rate is envelope based and its accuracy is a factor of the adaptive fit of the envelopes, not on exclusivity in setup conditions. Under the MarketTyping, its weighting is going to be extremely low at 3 sigma. Actually the only time I’m ever active at 3 sigma is at very short time frame discretionary intraday trading in certain conditions.

 

Re “Also, with 80% rates, the number of trades will be quite low, right? “ On its own, the envelope system is very active… many setups. Maybe something more relevant to your system below...

 

“Also, are MR systems you are talking about traded on relative values or outright directional. (Pair trading Vs buy on price retracement)”

Outright directional. Not sure I fully understand the question... will say though, in my experience, (excepting fx pairs of course) spreads are best done based on seasonals not technicals.

 

re “They are inversely correlated. This is the reason i try to develop MR systems. My current portfolio is packed with trend following/break-out systems. I only trade one market, so i have to make the most out of as many market conditions as possible.”

Once you get these systems settled in, look for a system to include that has very low (neither + or - ) corre to either of them…

 

Begin Intermission

Just skimmed the thread again and am going off on a short rant on testing vs “personal preferences” (see post #3) here for the general readers and noobies. If there's one, there are 640,000 posts about how traders are applying stops, targets, allowable drawdowns, etc. because that's what they 'prefer'. Most don't know or try to find out what is optimal for the system(s) they are trading. While many of them may be 'passing' for now, they're certainly bringing down the "class average". Usually their whole process doesn't have the vibratory congruence required for staying power. Imo, any trader who insists on a 'preference' must also be willing to get the hammer, tongs, bellows, and fire and literally beat the system's properties to where its 'shape' / properties matches those 'preferences'. It requires the persistence to endure 1000 ++ failures and requires a lot of work and time. Otherwise go with close to the best testing parameters.

End Intermission.

 

I’m having a very similar conversation elsewhere regarding oscillator based 'MR' systems and the issue of target /risk ratio and hit rate so I thought I would share some of my thoughts here too. Palm, being that ‘the slightest change to any parameter of a system is actually creates a whole new system’ (paraphrasing my own experiences in testing, Mathemagician,etc), let me know if it’s high jacking your thread or too far off topic and I’ll butt out.

Anyway, the significant drop off in W/L hit rate when target size is increased with an oscillator based MR system ( rsi in this case) arises from the rsi’s tendencies to ‘top out’ many more times in a move before price tops and really ‘reverts’. To reveal this tendency of the osc hooking 2,3,4,5,6,7… times before the ‘real’ price reversion takes, set the length parameter of an rsi to a radically short length and observe a bunch of moves and subsequent reversions.

One approach I’ve used is to deal with this is to start with small targets (and in some cases smaller stops also) and increment them to a larger % of what meets your criteria for ‘reversion’ (ie the central tendency) with each trigger in a move. Unfortunately, this really complicates the code . Also, don’t be at all surprised or discouraged if your test results go off in the wrong direction at first. Persist with no guarantees - That is the only thing I can say here with certainty. “Great things are not done by impulse but by a series of small things brought together” Vincent Van Gogh. In this case tweaking the price aspects of the patterns may be enough. More likely other adjustments will be necessary. One or more additional relationship factors may need to be included with the ones stated so far by op. of “plain RSI with price starting to reverse” etc etc., Or, you may have to get creative with how you vary the targets (and stops), etc etc (As my sizing allocations come prior to this stage, I can’t really speak with any authority about it; but you could conceivably also increment size in these progressions. ) hth

Edited February 23, 2011 by zdo

[palm 10]

Thanks for your recommendations on observing no. of times price exceed extreme reading before making actual revert and on stepping parameter values incrementally. I'll do those. For now i'm several step behind. I still have to convince myself that RSI works or not. For one thing, both RSI <20 and RSI >80 shows upward bias of subsequent price movements. Also, my out sample's performance does not quite match up with my insample's. Lastly, since no. of occurance is quite low (RSI<20 occurs 4% of the time), tweaking price patterns will further reduce no. of trades which i fear will not be representative of the "RSI effect" i'm after.

When i add a parameter (such as stop-loss) I want to make sure that the improvement in the result is not attributable to the new parameter. For example, i found that just by adding 1ATR exit, i was able to improve returns of a random entry system.

Since you mention preference, i have been trying to answer one question for so long but could not find the answer, so i settle with my preference. The question is this, given 2 systems with their expected returns and risks, is there a way to optimally allocate capital between the 2? Let's assume that they are uncorrelated. I try using Kelly's formula because it claims to maximize growth rate, but it requires me to endure something like 70-90% Drawdown. I ran countless number of drawdown simulations. But i still cound not find the answer. So i settle the question using my preference. I just don't want to endure more than -30% maximum DD on my equity while aiming at 15% compounded returns. So if i have a system that has max DD of -60%, i will only trade half of my money on it. Still the question nags me from time to time, had i been willing to take more risk, my equity growth would have been such and such....

I don't want to turn this post into a "rambling of a confused system trader" so i better stop here.

Thanks again for sharing your experience, Zdo. I'll update my finding on RSI system once there's a meaningful progress.

 

Palm

PS: Which markets do you trade in? I trade Thai individual stocks and its index futures.

[zdo 20]

Re “Thanks for your recommendations on observing no. of times price exceed extreme reading before making actual revert and on stepping parameter values incrementally.” Just to make sure you understood me, I was talking about ‘incrementally’ increasing target size (and stop distance (and size possibly)) with each setup within same price swing, as one way of dealing with W/L dropping off unless wide stops, etc.

 

Re; “I still have to convince myself that RSI works or not.” It doesn’t! razz! :rofl: Seriously though, for me indicators like RSI ‘work’ when I answer “when’ (to use) instead of ‘how’ (to apply) questions.

 

Re: “When i add a parameter (such as stop-loss) I want to make sure that the improvement in the result is not attributable to the new parameter.” The result is not attributable to the new parameter. Result is attributable to the whole new combination of all the parameters. Again - the slightest change to any parameter of a system actually creates a whole new system. Pound that into your whole being. Be careful that you are using stats instead of them 'using' you.

 

Re: “tweaking price patterns will further reduce no. of trades which i fear will not be representative of the "RSI effect" i'm after.” I was hinting tweak the price patterns (and other parameters) to increase the no. of trades – not to anything like HFT levels mind you, but to more setups. In my experience with coded trading, far too many cherry setups turn out to be mediocre trades and far too many mediocre setups turn out to be great trades… Good setups are not neat…they are messy. Good system development is not normal … it’s way out at the edge of reasonable.

 

Re: “i have been trying to answer one question for so long but could not find the answer, so i settle with my preference… given 2 systems with their expected returns and risks, is there a way to optimally allocate capital between the 2?” Until I learned to vary allocations in a portfolio of systems via MarketTyping instead of trying to set allocations at optimal fixed levels, I stayed pretty frustrated with that too.

Another approach for dealing with this that you might find helpful is the newer work of Ralph Vince, particularly The Handbook of Portfolio Mathematics; Formulas for Optimal Allocation and Leverage.

Nauzer J. Balsara also has a less complex but acceptable ( in that he doesn't completely ignore risk of ruin ) allocation model based on Geometric Mean in his book Money Management Strategies For Futures Traders

 

Re: Which markets do you trade in? My automated trading uses US index futures, US treasury futures, and high volume FX futures.

 

Re: “I don't want to turn this post into a "rambling of a confused system trader" Most all automated trading posts are the ramblings of a confused system trader – including mine. When I'm no longer challenged and passionate about it, it will be time to move my interests to another field.

 

Good talking with you Palm. When you have time, keep us posted.

 

All the best,

 

zdo

[palm 10]

Thanks, Zdo.

 

Wish you the best.