Pattern recognition / Trading without indicators

[SpiritBeing]

I've not been able to devise an effective mechanical system which uses only indicators for entry and exit, but I can see that MACD has value as part of a confirmation process.

Money management - I used a fixed percentage of equity per trade, initial risk is n ATR away, so position size is %/n. A friend of mine is starting to get into the Kelly Criterion, which tells you the 'optimal' amount to bet, but you can also go to zero as well. So another theory would be to allocate a certain percentage of trading pot to a certain strategy, then apply Kelly to that. There's some stuff about this on the net.

everything you wrote up there is excellent and you are obviously very on-point with it(i imagine)

I have extensively looked into optimal MM %R and im well versed in the Kelly criterion. Im glad you are looking to apply the Kelly value to each of the strategies you use..I think this is a clever approach. I used to design systems back in my mechanical days and my logic was identical to yours....

However....how correlated are your systems? for example....i wouldnt like to apply the Kelly formulas to two systems trading the same instrument!(but thats without much substance on my part, but my intuition brings me to say this).

However it all depends on the numbers and the distribution of the systems results!


My EOD strategy i created also uses a similar ATR type position sizing method..but with a twist

 

[meanreversion]

However....how correlated are your systems? for example....i wouldnt like to apply the Kelly formulas to two systems trading the same instrument!(but thats without much substance on my part, but my intuition brings me to say this).

Spooky .... I've been working on this over the last week.

I run two systems with a correlation of +40%. This in itself is not a problem, as long as the combined Sharpe ratio is higher than either of the individuals, which in my case it's not. So in other words, I'm trading two vaguely similar systems.

However, the starting point for any system for me is that it should have positive expectancy. I've not gotten good results for combining a pos. exp. system with a negatively correlated neg. exp. system --- you end up with a lower Sharpe.

Whether it's return/drawdown or Sharpe, there's not much point in combining two systems unless the combined result has a higher ratio than either of the two individuals. I need to think about this a bit more. Is it possible (for example) to find two systems which are negatively correlated, each with pos. exp., such that combined Sharpe is better than either two? I think the answer is 'yes', I just need to find the systems.

 

[SpiritBeing]

Spooky .... I've been working on this over the last week.

I run two systems with a correlation of +40%. This in itself is not a problem, as long as the combined Sharpe ratio is higher than either of the individuals, which in my case it's not. So in other words, I'm trading two vaguely similar systems.

However, the starting point for any system for me is that it should have positive expectancy. I've not gotten good results for combining a pos. exp. system with a negatively correlated neg. exp. system --- you end up with a lower Sharpe.

Whether it's return/drawdown or Sharpe, there's not much point in combining two systems unless the combined result has a higher ratio than either of the two individuals. I need to think about this a bit more. Is it possible (for example) to find two systems which are negatively correlated, each with pos. exp., such that combined Sharpe is better than either two? I think the answer is 'yes', I just need to find the systems.

Spooky?.....intuition is a powerful tool...we traders use it everyday playing the markets.

Interesting.....interesting....interesting!

I havnt thought about system trading for a while but i love thinking of new ideas!

hmm...I think your correct and i agree with your logic with the sharpe ration but like you said the key is expectancy!!!!. If you have a positive expectancy the key is how your numbers fall in order to determine if pumping the kelly formula to the max is optimal or DANGEROUS.


il will try and give an example....the way i see MM anyways

if you have two systems...S1, S2.

S1 expectancy > S2 expectancy HOWEVER
S2 max drawdown < S1 max drawdown.

So in theory even though S2 has a worse expectancy you could risk more per trade and make more money than S1......

Basically the point im trying to very badly make is....it all depends on the distribution of both systems combined. For example if you had a trend following system, then a counter trending system working on the same instrument then maybe?....Just my thoughts.

Here is a point....since we are talking about positive expectancy so much...would a sortino ratio be more relevant than a sharpe ratio?


ps....check out.... quickly

http://www.trade2win.com/boards/general-trading-chat/116880-current-charts-you-find-interesting.html

I havnt looked into systems for a while but i might want to in the near future. If you have a look at my style roughly....do you think i could code it easily??

GREAT TO SHARE btw

 

[meanreversion]

You'll struggle to code up resistance and support.

Sortino ratio ... zzzzzz .... too tired to think about that today, will look tomorrow. Interesting suggestion though.

The metric I use to compare systems is return/drawdown, or MAR ratio. Generally speaking, the higher the MAR, the higher the Sharpe. It's easier to calculate Sharpe than MAR, so I tend to use the former.

 

[SpiritBeing]

You'll struggle to code up resistance and support.

Sortino ratio ... zzzzzz .... too tired to think about that today, will look tomorrow. Interesting suggestion though.

The metric I use to compare systems is return/drawdown, or MAR ratio. Generally speaking, the higher the MAR, the higher the Sharpe. It's easier to calculate Sharpe than MAR, so I tend to use the former.

FUNNY THING I FOUND!

it seems the worst a systems win ratio....the better the systems expectancy seems to be...well for trend following anyways.

I then read Market wizards 2 again recently and one of the guys in there agreed.

What you rekon:?:

 

[meanreversion]

Hmm, it's possible to have a fairly low win rate, 20-25%, and still make decent coin, although there will be periods of many sequential losses. I think for the psyche you're better off aiming for at least 30% win rate.

 

[meanreversion]

Good article about Sortino here, fairly technical actually ...

http://www.thisisthegreenroom.com/2010/dont-touch-that-sortino-ratio/

 

[brettus]

I thnk the main issue with S and R is that people see too many lines on a chart. They shouldn't be seeing too many lines on a chart. Also at what times of the day S and R is more effective. I find that it works more effectively in New York trading hours. Sticking to the higher time frames aswell. Even if you're trading once a day, you could be trading too much. The only way you can know when to trade and when not to is just practice.

 

[meanreversion]

Has anyone backtested S+R, is it even possible .. ?

 

[SpiritBeing]

Hmm, it's possible to have a fairly low win rate, 20-25%, and still make decent coin, although there will be periods of many sequential losses. I think for the psyche you're better off aiming for at least 30% win rate.

the best swing and trend following systems are between 25-40% win ratio so i think you can get better than decent coin IMO.

I agree about 30% in general.....

 

[leopardtrader-com]

the best swing and trend following systems are between 25-40% win ratio so i think you can get better than decent coin IMO.

I agree about 30% in general.....

Swing system based on inflexion points have >70% success rate especially if you have lower than intended target. The higher the target the more the time and probability of wiping off gains. Trend folloing systems over time is less than 20%

 

[SpiritBeing]

Swing system based on inflexion points have >70% success rate especially if you have lower than intended target. The higher the target the more the time and probability of wiping off gains. Trend folloing systems over time is less than 20%

Its all a matter of opinion...there are people making millions either way.

 

[brettus]

Has anyone backtested S+R, is it even possible .. ?

I dont think its possible to write a program that can do it aswell as the human eye, and then looking at the candle stick patterns on a lower time frame to work out whether or not you should trade it. It probably is, but not for the average reader.

 

[meanreversion]

Here is a point....since we are talking about positive expectancy so much...would a sortino ratio be more relevant than a sharpe ratio?

Ok, I've been boning up on Sortino.. I'm an expert now, ask me anything

Sharpe = excess return / st dev of all returns

Sortino = excess return / st dev of negative returns only

If the system comprises mainly small losses with the occasional big winners (e.g. trend following), Sortino gives you a better idea of the "bad" risk required to get a good return.

 

[SpiritBeing]

Ok, I've been boning up on Sortino.. I'm an expert now, ask me anything

Sharpe = excess return / st dev of all returns

Sortino = excess return / st dev of negative returns only

If the system comprises mainly small losses with the occasional big winners (e.g. trend following), Sortino gives you a better idea of the "bad" risk required to get a good return.

excellent!

can you use this in your calculations for the application of the kelly formula?

 

[meanreversion]

Kelly requires some input of edge, which can be derived from a backtest. How would you propose using Sortino with Kelly?

 

[SpiritBeing]

Kelly requires some input of edge, which can be derived from a backtest. How would you propose using Sortino with Kelly?

Im not sure yet.

This is off the top of my head and im trading balls deep at the moment so bare with me...i think there might be substance to this thought process.

Since the sortino only incorporates the negative skew of a strategy rather than the positive and negative skew like the sharpe ratio.

Compound interest is the reason for all this enquiry....basically scalability is key to huge sums of wealth when it comes to price speculation IMO. So on this note.....


For me the negative skew of a strategy is the most important factor that you must scrutanize HARD...obviousley you must have a positive expectancy, but im sure you are aware that controlling and managing risk is the most IMPORTANT factor to survival and eventually riches.

Im not sure how exactly you would apply this emperically but the jist of it is this IMO...

If the system comprises mainly small losses with the occasional big winners (e.g. trend following), Sortino gives you a better idea of the "bad" risk required to get a good return.

If you calculate the "bad risk"....then this is key for me. i will give a very bad example below.

If a system has an MAXDD of 10% but when it wins it wins 30% then the sortino ratio will only look at the -10% ( compared to sharpe that would use 10% and 30% in the calculation)....

Once you have the max DD for a system then you can then set you own risk level depending on your appetite.

In the above example if you are happy risking 30% maxDD then then upside would then be +90% on winners(if you trades using fixed fraction money management)

Hope this makes some sort of sense

 

[meanreversion]

Hmmmm, it makes a little sense, but irrespecitve of how much or little you dedicate to a strategy, I wouldn't advocate accepting a drawdown of 70%+.

Where I think Sortino is useful is in trading two strategies, each with Sortino > Sharpe (i.e. occasional big winners/many small losses).. I'm still running some numbers and will let you know.

 

[SpiritBeing]

Hmmmm, it makes a little sense, but irrespecitve of how much or little you dedicate to a strategy, I wouldn't advocate accepting a drawdown of 70%+.

Where I think Sortino is useful is in trading two strategies, each with Sortino > Sharpe (i.e. occasional big winners/many small losses).. I'm still running some numbers and will let you know.

I agree....70% DD is unacceptable lol....it was simply for example purposes.

But cool. Your idea of both systems sortino>sharpe then implamenting it sounds solid.

Definately keep me posted on your findings

I may choose to make some systems in the future so it would be interesting to know.

 

[meanreversion]

The issue would be something like this ...

system A has return of 20% and drawdown of 20%
system B has return of 10% and drawdown of 20%

Is is worth trading both systems? What is the main metric (or metrics) for evaluating the combined system?

Let's say system A+B has return of 20% and drawdown of 10%, then I would argue you should trade them together. But what if A Sharpe = 1.2 and B Sharpe = 1.0, and A+B Sharpe (no music jokes please) = 0.9, is the combined system better?

Maybe you also need to look at Sortino for the combined, especially if both systems have low-ish win rates.

This may all sound a bit academic, but the more I think about it, the more I think this is a sound approach to portfolio management of systems.