Reversion to the mean only trading...

MarkBrown said:
@ES - Converting indicators which often seem useless into a trading model.

a.) it's the logic used to build the model not the indicators that make it successful.

ItnJt1.png


So it looks good but looks can be deceiving why not automated it?

a.) the human eye will mislead you to look for big moves, but chipping is more profitable.

bWlVu4.png


Based on fixed ratio starting with 1 ES and levering with no additional risk up to 10 contracts.

Only wins 57% of the time, run 24hrs a day non stop, 50k to 1,000,000. in 100 days.
1 to 2 risk reward - yes 200 risked for 100 made "embrace the risk" opposite of the norm.

Draw-down was less than 12%, so obviously I was disappointed with that. For years I tried incorporating reversion to the mean with trend trading and found it's better just to park when it's trending and capture the majority of the market action which is choppy action.

Happy to answer questions and give insights to building your own model. Mark

PS I have been running this system for over 15 years so it's not over over optimized, I just added the fixed ratio component to it otherwise it would have just made under 100k over the same 100 days.

By contrast an always in model which incorporates both reversion to the mean and trending modes. Seems at first glance to make more money and it does but when you figure in the slippage and commissions of bidirectional trading you wind up with less. But it looks impressive to the eye.

mi6B3U.png
More...
Hello MarkBrown,

Some of my wording may seem ignorant regarding algo development, so forgive me. Algo development and testing is something I have not done in about 2 years.

I have some questions for you please:

1. Did you back test the algo for historical years results? For example, did you run the algo from 2010-2019 to see the performance? I ask this because when I back test the algo I wrote and programmed, they all showed bad results.

2. Regarding algo development: From your experience, is it common to for alot of algo trading ideas to fail when programmed and then back test?

3. Regarding algo curve fitting and optimization: I read/heard that "over" optimization is a bad thing and will fail in the future trading of the algo. How much of the algo is optimized? Or was optimization even needed?

4. Regarding alo back and walk forward testing: Did you perform all your backtesting, optimization, walk forward testing, experimenting on a certain data set (like from 2010-2015)? And then tested that experimented/tuned algo from 2015-2020? I believe they call his in sample and out of sample data.

5. Is finding a profitable aglo challenging? What can share about the most difficult part of algo development?

6. Why do most algo developers fail at creating a consistent intraday algo?

Thank you sir.
 
Hello MarkBrown,

Question please:

Please talk to me about the importance of tick data (or Tick bar Granularity) with your range bar testing.

How important is it to have reliable and accurate historical data for back testing and development? Especially for the intraday algos.

I recall I lost about $3000 with NT7 5 minute bar algo not knowing their free data only had OHLC data points only for its back testing algorithm.
 
MarkBrown said:
@ES - Converting indicators which often seem useless into a trading model.

a.) it's the logic used to build the model not the indicators that make it successful.

ItnJt1.png


So it looks good but looks can be deceiving why not automated it?

a.) the human eye will mislead you to look for big moves, but chipping is more profitable.

bWlVu4.png


Based on fixed ratio starting with 1 ES and levering with no additional risk up to 10 contracts.

Only wins 57% of the time, run 24hrs a day non stop, 50k to 1,000,000. in 100 days.
1 to 2 risk reward - yes 200 risked for 100 made "embrace the risk" opposite of the norm.

Draw-down was less than 12%, so obviously I was disappointed with that. For years I tried incorporating reversion to the mean with trend trading and found it's better just to park when it's trending and capture the majority of the market action which is choppy action.

Happy to answer questions and give insights to building your own model. Mark

PS I have been running this system for over 15 years so it's not over over optimized, I just added the fixed ratio component to it otherwise it would have just made under 100k over the same 100 days.

By contrast an always in model which incorporates both reversion to the mean and trending modes. Seems at first glance to make more money and it does but when you figure in the slippage and commissions of bidirectional trading you wind up with less. But it looks impressive to the eye.

mi6B3U.png
More...
Hello MarkBrown,

Question please:

1. Is the algo programmed to automatically increased/decrease the position size based on account balance and accounting for drawdown limitations per contract? I think your back test results shows increased position size the more money it makes.

Thanks,
 
SimpleMeLike said:
Hello MarkBrown,

Some of my wording may seem ignorant regarding algo development, so forgive me. Algo development and testing is something I have not done in about 2 years.

I have some questions for you please:

1. Did you back test the algo for historical years results? For example, did you run the algo from 2010-2019 to see the performance? I ask this because when I back test the algo I wrote and programmed, they all showed bad results.

2. Regarding algo development: From your experience, is it common to for alot of algo trading ideas to fail when programmed and then back test?

3. Regarding algo curve fitting and optimization: I read/heard that "over" optimization is a bad thing and will fail in the future trading of the algo. How much of the algo is optimized? Or was optimization even needed?

4. Regarding alo back and walk forward testing: Did you perform all your backtesting, optimization, walk forward testing, experimenting on a certain data set (like from 2010-2015)? And then tested that experimented/tuned algo from 2015-2020? I believe they call his in sample and out of sample data.

5. Is finding a profitable aglo challenging? What can share about the most difficult part of algo development?

6. Why do most algo developers fail at creating a consistent intraday algo?

Thank you sir.
More...


1. yes i run it for 15 years and makes more in the present but not taking into inflation and devaluation of the dollar. typical model that use fixed inputs fall apart over time best to build with percentages or at least some dynamic engine.

2. absolutely, it's about impossible for the average human to completely encompass discretion into code. likewise it's almost impossible for humans to accept the results of an idea back tested because it looks so good to the eye. this impossibility to code and accept results makes most aspiring modelers bitter, rather than self improve.

3. you don't accept the best results, you settle for the 75% area of the results when actually trading. that way you gain some dependability for the sacrifice of profitability. it's foolish to expect optimal results will hold up so if you can be happy with the numbers at 75% it might hold up.

4. i would suggest this for anyone starting out, i have very specific tuning i am doing. i am not building or looking for systems, i am refining and perfecting existing systems that have been used in live trading and show promise. i do specifically omit some events and days out of back testing because i am looking for a pure overall soundness.

5. if i knew then what i know now i would have been the worlds best discretionary trader and not a systems trader. failure to properly detail an idea into code, the time, the money wasted. but it was an educational experience that has given me the knowledge of what all can possibly go wrong. there's nothing as effective to educate one than losing money on a incomplete idea.

6. lack of complete details, inability to express those details into code, being overwhelmed with lust for the unearned rather than being honest with ones self. all trading platforms have optimization pitfalls mostly well documented, yet budding systems traders will put that on ignore. over fitting to the data as was mentioned above accepting less than optimal results as possible reality.

thanks for the questions - m


ps i think most all traders don't spend enough time understanding where they are in the playing field of trading. they don't know the rhythm of the markets they are trading, broad market grounding is essential. forest for the trees sort of thing.
 
Last edited:
SimpleMeLike said:
Hello MarkBrown,

Question please:

1. Is the algo programmed to automatically increased/decrease the position size based on account balance and accounting for drawdown limitations per contract? I think your back test results shows increased position size the more money it makes.

Thanks,
More...


1. yes - there are many schemes. making your current size earn the right to add contracts is highly recommended, also backing down that size quicker than it was added has shown good results.
 
Last edited:
MarkBrown said:
1. yes i run it for 15 years and makes more in the present but not taking into inflation and devaluation of the dollar. typical model that use fixed inputs fall apart over time best to build with percentages or at least some dynamic engine.

2. absolutely, it's about impossible for the average human to completely encompass discretion into code. likewise it's almost impossible for humans to accept the results of an idea back tested because it looks so good to the eye. this impossibility to code and accept results makes most aspiring modelers bitter, rather than self improve.

3. you don't accept the best results, you settle for the 75% area of the results when actually trading. that way you gain some dependability for the sacrifice of profitability. it's foolish to expect optimal results will hold up so if you can be happy with the numbers at 75% it might hold up.

4. i would suggest this for anyone starting out, i have very specific tuning i am doing. i am not building or looking for systems, i am refining and perfecting existing systems that have been used in live trading and show promise. i do specifically omit some events and days out of back testing because i am looking for a pure overall soundness.

5. if i knew then what i know now i would have been the worlds best discretionary trader and not a systems trader. failure to properly detail an idea into code, the time, the money wasted. but it was an educational experience that has given me the knowledge of what all can possibly go wrong. there's nothing as effective to educate one than losing money on a incomplete idea.

6. lack of complete details, inability to express those details into code, being overwhelmed with lust for the unearned rather than being honest with ones self. all trading platforms have optimization pitfalls mostly well documented, yet budding systems traders will put that on ignore. over fitting to the data as was mentioned above accepting less than optimal results as possible reality.

thanks for the questions - m


ps i think most all traders don't spend enough time understanding where they are in the playing field of trading. they don't know the rhythm of the markets they are trading, broad market grounding is essential. forest for the trees sort of thing.
More...

Thank you for your detailed post.
 
MarkBrown said:
ps i think most all traders don't spend enough time understanding where they are in the playing field of trading. they don't know the rhythm of the markets they are trading, broad market grounding is essential. forest for the trees sort of thing.
More...

Thanks,
I agree with this. And this is why 2 years ago I stopped algo development and clicking backtest 100 times per day. It was indeed fun, but I felt something was missing like I was just pulling stuff out of air based on what I knew regarding my own knowledge of trading. I needed some skills.

I felt that starring at the chart (price only) everyday could give me more hands on experiment. There just something about starring at charts all day. Seeing repeated things and managing those repeated things without outside influence. Sort of like a boxer in the boxing ring.
 
MarkBrown said:
5. if i knew then what i know now i would have been the worlds best discretionary trader and not a systems trader. failure to properly detail an idea into code, the time, the money wasted. but it was an educational experience that has given me the knowledge of what all can possibly go wrong. there's nothing as effective to educate one than losing money on a incomplete idea.
More...

Hello MarkBrown

Can you please clarify what you mean by the high lighted section?

Thanks
 
SimpleMeLike said:
Hello MarkBrown

Can you please clarify what you mean by the high lighted section?

Thanks
More...

i learned to build my own computers, run and make rj45 cable, install computer flooring, run a cray, hp ux10, program in various languages, built my own trading software "traderware" and and provided input for tradestation, tradevec and pioneered the multi data feed capability that multicharts uses, install satellite data feeds and run fiber up and down elevator shafts, install 4.7 miles of fiber to my office, run my own dns, web, mail, trading servers, a member of ARIN with my own ip blocks you name it in the trading industry and related fields and in over 30 years i have done it hands on.

i started out building trading systems, never did anything else but focus on trading systems. i did not want to manually make the decisions of where to buy and sell. i wanted a computer to dictate to me what to do. i did that very well to this day. but i wish i had focused as hard on just discretionary trading as my mentor was. i never embraced taking responsibility for trades on the fly with my mind. mostly because i did not trust my eyes, the human eye tends to see what it wants and ignore the rest.

but had i laid aside the pursuit or automated trading systems and instead concentrated my skills as a discretionary trader i would have surely been the best there ever was considering the effort i put in and the money i have spent.
 
You must log in or register to reply here.
Back
Top