One Way to Develop a Winning System | Page 4 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

lets develop a system which is very very simple but effective. 7 replies

How To Develop Winning Automated Trading Systems 15 replies

Change a losing system into a Winning system 68 replies

Develop a System that Fits You, by Dr Van K. Tharp 3 replies

Need Ea for System I am trying to develop 0 replies

Attachments: One Way to Develop a Winning System
Exit Attachments

One Way to Develop a Winning System

  • #61
  • Sep 25, 2009 7:26pm Sep 25, 2009 7:26pm
Some nice stuff here that you don't usually see on FF, no wonder the thread died
The breaking of a wave cannot explain the whole sea.
  • #62
  • Sep 25, 2009 8:12pm Sep 25, 2009 8:12pm
Quoting Craig
Some nice stuff here that you don't usually see on FF, no wonder the thread died
Agree. Glad to see that you vouced for FXEZ. One more for me and I can pass out them vouches. Great contributions by FXEZ.
  • #63
  • Sep 25, 2009 9:11pm Sep 25, 2009 9:11pm
This is a fantasic thread, although I'm afraid its probably a case of pearls before swine.

If you ever consider a colaborative systems development project send me a PM
  • #64
  • Sep 25, 2009 9:40pm Sep 25, 2009 9:40pm
He wasn't trolling. He was addressing you soliciting your services, which is a violation of Forex Factory rules.

Quoting FXEZ
Bless your heart Mr. cookie monster, I do believe that's your second consecutive off-topic post in this thread. If I didn't know better I'd think you're attempting to sprinkle cookie crumbs all over this thread. Pretty clever trolling for someone who confuses easily.
  • #65
  • Nov 17, 2013 10:25pm Nov 17, 2013 10:25pm
It has been over 6 years since this thread was started. I'm still around, developing trading systems for my personal use only and I wanted to add a couple of things I've learned during that time regarding system development. I'm pleased to say after re-reading the thread that I still agree with just about everything I posted on the topic of system development 6 years ago!
  • #66
  • Nov 17, 2013 10:27pm Nov 17, 2013 10:27pm
As an update to the topic of Drag-Free Position Sizing I've recently been doing some MM simulations using percent of equity sizing.

The Kelly Criterion is a special case of a percent of equity sizing method that also happens to be median log wealth optimal. With percent of equity sizing methods, size is altered with each trade. Trade size will thereby increase/decrease after each respective win/loss. This incremental increase/decrease in size over time creates a negative drag on the trading account akin to paying a (rather large) negative interest rate.

To address this issue I suggested using a plateau method. The obvious drawback of the plateau method is that size never decreases! If a prolonged drawdown occurs, this MM strategy can prove to be devastating, and essentially defeats the purpose of the Kelly sizing algorithm (decreasing risk during drawdown to prevent ruin).

Long story short, I have found from simulations that increasing size immediately after wins, and delaying the decrease in size after a loss, can fix this problem.

I tested two scenarios for wins: increasing size immediately after a win, and increasing size only after an equity high. It didn't appear to make a difference either way. But both methods were effective in creating a positive boost to performance associated with trading a profitable strategy with larger size over time.

For losses it is a different picture. Immediately decreasing size after a loss creates negative drag on system performance (because after a series of trades such as: WLWLWL you lose on larger size than you win on and these losses accumulate over time, thus the negative interest rate).

The solution is to delay decreasing size until a fixed time period / # of trades has elapsed. Depending on the number of trades your strategy takes, a fixed time period could be daily, weekly, every two weeks, monthly, quarterly, a certain # of trades, etc. By delaying the decrease in trade size, it is hoped that the normal wins and losses in the short term will offset, minimizing the need to reduce size in the short term. The effect is that the negative drag on the account from size reduction after a single loss is greatly reduced. At the same time, risk of ruin is minimized because size does decrease after a series of losses.

Combine this trick with using a percent of equity method that only uses a fraction of the entire account, and you have a very powerful money management strategy! Rather than repeat myself I'll just link to some previously posted content on this subject:

Kelly continuous expected growth rate
continuous Kelly Criterion
why continuous Kelly and not discrete Kelly

Note: the second link describes the sub-account method for predetermining the maximum size of drawdown by only putting a fraction of the total account at risk. Also, the attached paper is very interesting.
  • #67
  • Edited 1:23pm Nov 18, 2013 1:21pm | Edited 1:23pm
Very interresting (as usual). I personally thought about an EMA of the bet size for the same reasons. The constant of the EMA would define the rate of adaptation from the current bet size to the target, currently estimated Kelly, bet size. The EMA is symmetrical about the win or loss outcome.

May you detail what you mean by "continuous Kelly Criterion". I compute Kelly using this formula, which is discrete (I hope it's correct):
Attached Image

Where profit(i) is the outcome of the i-th trade (can be negative).

Another question: did you try with other convex (risk averse) utility functions? I'm especially thinking of the square root.
Did you try with concave utility functions (risk seeker)?
No greed. No fear. Just maths.
  • #68
  • Nov 18, 2013 10:13pm Nov 18, 2013 10:13pm
Quoting PipMeUp
Very interresting (as usual).
I hadn't thought of using the EMA of trade size but the idea sounds extremely interesting.

Quote
May you detail what you mean by "continuous Kelly Criterion".

If you take a look at the paper in the 2nd link: "Betting with the Kelly Criterion", there is a derivation of the continuous formula in section 3 (starting on page 3). Take a look at page 5 for the final formula. My version (below) is the same but leaves out the risk free rate (which I am assuming is zero).
Inserted Code
Average(returns) / Variance(returns)
The reason for using the continuous formula as stated in the paper (page 3), "Since there is not a finite number of outcomes of a bet on a security, we must use continuous probability distributions." If a given strategy used the same fixed SL and TP on every trade, it could be argued that the discrete formula is appropriate. For all other strategies where outcomes vary, the continuous formula should be used.

As I'm also not a mathematician, I'm not able to tell if your formula is correct. Perhaps you simplified a more common form of the formula such as off Wikipedia?

Quote
Another question: did you try with other convex (risk averse) utility functions? I'm especially thinking of the square root. Did you try with concave utility functions (risk seeker)?

By "try" do you mean mixing a % of equity method with risk averse utility functions such as square root / log in some way? I'm not sure how I would go about that (so obviously no). I have experimented with many different algorithms but beating Kelly (even fractional Kelly as a risk-averse alternative) is a <hard> problem. It can be done in the short term but in the long term? And can it be done systematically in the short term? If so then perhaps a Nobel prize would be in your future!

For my purposes I'm happy with the solution described before because: it allows a maximum (preset) % of account drawdown, reduces (substantially) the negative drag associated with % of equity methods, and produces geometric growth over the long term. I'm not sure how I could improve on this.
  • #69
  • Nov 19, 2013 1:13am Nov 19, 2013 1:13am
Quote
By "try" do you mean mixing a % of equity method with risk averse utility functions such as square root / log in some way? I'm not sure how I would go about that (so obviously no). I have experimented with many different algorithms but beating Kelly (even fractional Kelly as a risk-averse alternative) is a <hard> problem. It can be done in the short term but in the long term? And can it be done systematically in the short term? If so then perhaps a Nobel prize would be in your future!
No Nobel for me. I'm afraid a lot of smarter people already thought about all of this. I just meant that Kelly is the optimal solution using U(x)=ln(x) as the utility function. You can optimize using other functions like U(x)=sqrt(x). None of them will beat Kelly nor Kelly will beat any of them. There is no order relation, they will all diverge to +infinity if the system is winning and towards zero otherwise. The way they do it may suite you risk aversion better.
No greed. No fear. Just maths.
  • #70
  • Dec 2, 2013 10:01am Dec 2, 2013 10:01am
I read lot of academic papers and some excellent discussions about money managment but there are very limited information about
the risk managment .Apart from MAE/MFE anaysis(MAE to profit ratio etc..) entry exit efficiency ratio I havent seeen anything good.Your feedbacks appreciated
  • #71
  • Jan 2, 2014 12:03pm Jan 2, 2014 12:03pm
Quoting FXEZ
...I prefer a SAR method, where longs are closed out by initiating shorts. ...
You asked for feedback so I would like to write my 2 cents if you don't mind. I do not like SAR method because you are always in the market (there are times to go long, there are times to go short and there are times to go fishing :-) ).
The more you stay in the market, the more you are exposing you open position to unknown risks. I prefer to define a good exit criteria (which is not a simple task) and exit as soon as the related criteria is met.
  • #72
  • Edited 11:07pm Jan 2, 2014 10:52pm | Edited 11:07pm
Quoting olsen-yersen
I read lot of academic papers and some excellent discussions about money managment but there are very limited information about the risk managment .Apart from MAE/MFE anaysis(MAE to profit ratio etc..) entry exit efficiency ratio I havent seeen anything good.Your feedbacks appreciated
Apologies for the late reply. I don't consider myself to be an expert on any of these topics. However, I have found that simulation is one of the better ways to discover what works and what doesn't. In particular, running different MM simulations with regards to a particular system can be useful to discern what works for a particular strategy and what doesn't. Like you, I haven't found much in the way of practical papers on the subject. I also don't think it is possible to generalize all MM observations to all systems.

In that spirit, the comments related to post #66 were in relation to a particular equity curve. My views have evolved / become clarified a bit based on new evidence in that I feel that increasing size only after equity highs is superior to just increasing size after each trade. This also makes logical sense to me and is simpler to implement so better all around.
  • #73
  • Jan 2, 2014 11:05pm Jan 2, 2014 11:05pm
Quoting mahdiquant
{quote} You asked for feedback so I would like to write my 2 cents if you don't mind. I do not like SAR method because you are always in the market (there are times to go long, there are times to go short and there are times to go fishing :-) ). The more you stay in the market, the more you are exposing you open position to unknown risks. I prefer to define a good exit criteria (which is not a simple task) and exit as soon as the related criteria is met.

Thanks for your comments. The quote was from 2007 but I still favor systems that use SAR, vs. using a discrete entry and exit in systems I create. I guess the main reason is that SAR systems seem to be easier to create, or tend to be more profitable when created (from my experience). And system development is a hard problem. Diversification also seems to work better when combining equity curves from SAR systems.

While I like the idea of being out of the market with no exposure, it is also true that you can't win if you don't bet. In addition, I saw a paper (searched but can't locate) a while back that made the case that a particular class of systems (described in the paper) performed better generally as SAR systems, rather non-SAR systems. This is also my experience. I suppose the reason has something to do with game theory, but I honestly don't know how it fits in.
  • #74
  • Jan 13, 2014 9:46am Jan 13, 2014 9:46am
Quoting FXEZ
{quote} Apologies for the late reply. I don't consider myself to be an expert on any of these topics. However, I have found that simulation is one of the better ways to discover what works and what doesn't. In particular, running different MM simulations with regards to a particular system can be useful to discern what works for a particular strategy and what doesn't. Like you, I haven't found much in the way of practical papers on the subject. I also don't think it is possible to generalize all MM observations to all systems. In that spirit, the...
Thank you, do you mean monte carlo simulation?

What I am trying to do is along with the market state analysis ,filter the signals with neural network and GA when it is below the certain MAE ratio .Lets say ratio 1(no drawdown) is the optimal signal.So monte carlo simulation can be better substitue to this?

I am trying to learn how to use quant share software currently http://www.quantshare.com/ .They have some advanced MM ,neural network, GA, brute force features.
  • #75
  • Jan 13, 2014 9:55am Jan 13, 2014 9:55am
Addition to SAR , interesting article about using TD Squential numbers as TP/SL targets http://www.asiapacfinance.com/tradin...ing-Strategies
  • #76
  • Jan 13, 2014 6:05pm Jan 13, 2014 6:05pm
Quoting olsen-yersen
{quote} Thank you, do you mean monte carlo simulation?
I suppose the simulation I was doing could be termed monte carlo because it uses randomization so yes though I was not thinking about any sort of strict or even that rigorous of a testing procedure. More like whipping out Python scripts to test various assumptions / rules for increasing / decreasing trade size.

Quote
What I am trying to do is along with the market state analysis ,filter the signals with neural network and GA when it is below the certain MAE ratio .Lets say ratio 1(no drawdown) is the optimal signal.So monte carlo simulation can be better substitue to this? I am trying to learn how to use quant share software currently http://www.quantshare.com/ .They have some advanced MM ,neural network, GA, brute force features.

I'm not familiar with that software. When signal filtering is done based on measures of equity or volatility I consider it to be within the realm of system development rather than money management. Adjusting system parameters to volatility can be a fruitful area if it can be assumed that future excursions will be similar in magnitude to historical excursions.

I'm not sure that I can justify such an assumption. Just as I don't know beforehand which systems will outperform in the future and which will underperform. For this reason I prefer to take a more naive view. I'm not sure if the magnitude of future equity movement is any more predictable based on past numbers. Maybe you've had different experience or maybe it is algorithm specific?
  • #77
  • Jan 2, 2016 4:03pm Jan 2, 2016 4:03pm
Hi FXEZ

As irrelevant as it sounds to this particular thread, it is the only way I could send a message to you.

I thank you for your most selfless contribution and the "pairs with beta" indicator should be in every serious market-neutral traders' pocketbook.

I've been testing a strategy using your great tool, just to let you know that your work made a difference.

It is rather unfortunate to see most people are looking for a trade-easy-system-handout, yet contributors in FF did not get proper respect for their works.

You have mine.

P.S. I'm about to close my "pairs trading" thread for good before turning it into an automated strategy, so I was wondering if you like to share some thoughts?
No Highs? No Lows? Must be Bose!
  • #78
  • Jan 2, 2016 5:26pm Jan 2, 2016 5:26pm
subscribed
have a nice day sir please
focus while relaxed to see better
  • #79
  • Jan 4, 2016 12:36am Jan 4, 2016 12:36am
Quoting Tardigrade
Hi FXEZ As irrelevant as it sounds to this particular thread, it is the only way I could send a message to you. I thank you for your most selfless contribution and the "pairs with beta" indicator should be in every serious market-neutral traders' pocketbook. I've been testing a strategy using your great tool, just to let you know that your work made a difference. It is rather unfortunate to see most people are looking for a trade-easy-system-handout, yet contributors in FF did not get proper respect for their works. You have mine. P.S. I'm about...
Hi Tardigrade, appreciate your comments. I always felt Pairs with Beta was one of my better creations, so I'm glad you're getting some value out of it. I'm not active on the forum much but I'll drop by your thread to make a few comments.
  • #80
  • Mar 3, 2016 10:10am Mar 3, 2016 10:10am
Quoting FXEZ
{quote} Hi Tardigrade, appreciate your comments. I always felt Pairs with Beta was one of my better creations, so I'm glad you're getting some value out of it. I'm not active on the forum much but I'll drop by your thread to make a few comments.

Sir it seems Your works are very complicated but it looks there is hope.
thanks
0 traders viewing now