Edit Your Comment
Humans vs. Robots
![](/userimages.jpg?uid=136673&id=40202&w=100&h=100)
forex_trader_136673
तबसे मेंबर है Jun 28, 2013
852 पोस्टों
Dec 16, 2013 at 07:27
तबसे मेंबर है Jun 28, 2013
852 पोस्टों
@MoneyCode
All indicators move with price, so using one or using five is almost the same because they use same data (price). My view is the same, markets are complicated enough, it os better so simplify.
All indicators move with price, so using one or using five is almost the same because they use same data (price). My view is the same, markets are complicated enough, it os better so simplify.
तबसे मेंबर है Nov 01, 2013
31 पोस्टों
Dec 16, 2013 at 15:08
तबसे मेंबर है Nov 01, 2013
31 पोस्टों
That is a good point, because from a manual trading perspective, the more indicators a trader uses to confirm his idea, then theoretically the greater chance the trade will perform as expected. It's a practice that reputable traders use and recommend, so you are absolutely right in that manual sense.
Think of it from a Learning perspective though. Because training an automated model to think correctly, requires testing and training, in order to learn what works. Example, if I set a trading rule as :
If Low of Bar is above 200 moving average then BUY.
What that one rule just did is eliminate 50% of my training data set (if I'm using a 6 month history of 1 minute charts, then I've just eliminated 3 months of history from being considered by my trading rule.) because half of the chart history will be above the 200 ma and half will be below it (approximately). So, my one trading rule doesn't even look at half of the 6 month period (the half that is BELOW the 200 ma.)
Now, lets say I add another rule. 2nd rule: If the 10 moving average is below the 50 moving average.
So here's the full code now:
1st rule> If Low of Bar is above 200 moving average AND
2nd rule> If the 10 moving average is below the 50 moving average then BUY.
So the first rule eliminated 3 months of data observations and the 2nd rule just eliminated 1/2 of that remaining 3 months.
So now my trading rules are actually only considering 25% of the original 6 month data set.
Each time I add another rule or parameter, then my training data set is in effective shrunk again. Making the results easier and easier to manipulate and 'over fit'. My results potentially become false positive.
The only way around this problem is to enlarge the training data set each time I add a trading rule. That way I don't loose data observations for my code to consider.
So there is a principle difference between a manual Trading approach, and an automated Training approach.
However, once the automated system model is trained to work profitably on 'Live' data then, in theory, the more indicators that were used could actually be better, as you described. The problem is that it is impossible for the system to learn from data when you're preventing it from observing much of the data by too many parameters. This is the degrees of freedom principle (my version). There is a formula for calculating the 'degrees of freedom' which I can't do, but I do grasp the principle of it.
Even manual trading, can be caught in this same trap without knowing it. There is not proof a system works (manual or automated) unless it has worked enough times in a large enough unrestricted field of data observations.
Maybe I haven't explained it perfectly but that is my understanding of it. It explains why so many systems fail once they're put in a live market environment, after appearing to work so well on back testing.
Think of it from a Learning perspective though. Because training an automated model to think correctly, requires testing and training, in order to learn what works. Example, if I set a trading rule as :
If Low of Bar is above 200 moving average then BUY.
What that one rule just did is eliminate 50% of my training data set (if I'm using a 6 month history of 1 minute charts, then I've just eliminated 3 months of history from being considered by my trading rule.) because half of the chart history will be above the 200 ma and half will be below it (approximately). So, my one trading rule doesn't even look at half of the 6 month period (the half that is BELOW the 200 ma.)
Now, lets say I add another rule. 2nd rule: If the 10 moving average is below the 50 moving average.
So here's the full code now:
1st rule> If Low of Bar is above 200 moving average AND
2nd rule> If the 10 moving average is below the 50 moving average then BUY.
So the first rule eliminated 3 months of data observations and the 2nd rule just eliminated 1/2 of that remaining 3 months.
So now my trading rules are actually only considering 25% of the original 6 month data set.
Each time I add another rule or parameter, then my training data set is in effective shrunk again. Making the results easier and easier to manipulate and 'over fit'. My results potentially become false positive.
The only way around this problem is to enlarge the training data set each time I add a trading rule. That way I don't loose data observations for my code to consider.
So there is a principle difference between a manual Trading approach, and an automated Training approach.
However, once the automated system model is trained to work profitably on 'Live' data then, in theory, the more indicators that were used could actually be better, as you described. The problem is that it is impossible for the system to learn from data when you're preventing it from observing much of the data by too many parameters. This is the degrees of freedom principle (my version). There is a formula for calculating the 'degrees of freedom' which I can't do, but I do grasp the principle of it.
Even manual trading, can be caught in this same trap without knowing it. There is not proof a system works (manual or automated) unless it has worked enough times in a large enough unrestricted field of data observations.
Maybe I haven't explained it perfectly but that is my understanding of it. It explains why so many systems fail once they're put in a live market environment, after appearing to work so well on back testing.
![](/userimages.jpg?uid=136673&id=40202&w=100&h=100)
forex_trader_136673
तबसे मेंबर है Jun 28, 2013
852 पोस्टों
Dec 16, 2013 at 15:55
तबसे मेंबर है Jun 28, 2013
852 पोस्टों
Thanks Jorge for contribution, you wont be able to post anymore. Your comment is rude.
Read @MoneyCode again, maybe you will learn not to blow accounts.
Read @MoneyCode again, maybe you will learn not to blow accounts.
![](/userimages.jpg?uid=136673&id=40202&w=100&h=100)
forex_trader_136673
तबसे मेंबर है Jun 28, 2013
852 पोस्टों
Dec 16, 2013 at 16:09
तबसे मेंबर है Jun 28, 2013
852 पोस्टों
@MoneyCode much appreciate your contribution.
Usually 200 MA is used by traders to tell the trend. If price is above 200MA then the market is bullish and if price is below 200MA the market is bearish. That way 200 MA should limit the trader to trade only with the trend and not against it, which is good. Then using faster averages 50 and 10 picks up shorter trends and it is a good strategy.
My understanding is that the strategy should perform only on trends, but it will give false signals when the market turns. Also will under perform when market is in range of an extended period of time.
If using this type of strategy, how do you define stop and target. What do you use as stop indicator, ATR?
Usually 200 MA is used by traders to tell the trend. If price is above 200MA then the market is bullish and if price is below 200MA the market is bearish. That way 200 MA should limit the trader to trade only with the trend and not against it, which is good. Then using faster averages 50 and 10 picks up shorter trends and it is a good strategy.
My understanding is that the strategy should perform only on trends, but it will give false signals when the market turns. Also will under perform when market is in range of an extended period of time.
If using this type of strategy, how do you define stop and target. What do you use as stop indicator, ATR?
तबसे मेंबर है Nov 01, 2013
31 पोस्टों
Dec 16, 2013 at 16:11
तबसे मेंबर है Nov 01, 2013
31 पोस्टों
Think of the training period of a trading plan, like a Drug company testing a new drug for FDA approval.
First they need a pool of participants to take the drug, but they can't use just anybody.
The first pool starts out with 100 people, but they have to exclude anyone who is taking any other drugs.
So that first rule (no drugs in past 2 months) will eliminate 25% of their pool of volunteers.
Next they have to ensure there is no genetically related medical conditions in the remaining pool, that would skew the results of the drug being tested. So that rule, eliminates another 25% of the pool. Now their down to 50 testable people out of the original 100.
Next they can't use diabetics or people with a history of drug abuse, so that eliminates more people. Before long, the original pool of 100 test subjects is down to 10 people.
Will the FDA approve a drug as safe, which was only been tested on 10 people? The answer of course is no. In fact the only time the FDA will do such a thing is if the patient approved for the drug, has a very short life expectancy anyway.
So the only way around these testing volunteers shrinking number, is for the drug company to start with a larger pool to begin with. And for each restriction they apply they must enlarge the original testing pool even more, so that in the end they are testing at least 100 qualified people.
That's the same idea, that I must keep in mind each time I add a condition to a trading plan I am trying to develop. The more rules I add, the less statistically significant the results will be, unless I enlarge the data set so as to not restrict the number of data observations being evaluated.
First they need a pool of participants to take the drug, but they can't use just anybody.
The first pool starts out with 100 people, but they have to exclude anyone who is taking any other drugs.
So that first rule (no drugs in past 2 months) will eliminate 25% of their pool of volunteers.
Next they have to ensure there is no genetically related medical conditions in the remaining pool, that would skew the results of the drug being tested. So that rule, eliminates another 25% of the pool. Now their down to 50 testable people out of the original 100.
Next they can't use diabetics or people with a history of drug abuse, so that eliminates more people. Before long, the original pool of 100 test subjects is down to 10 people.
Will the FDA approve a drug as safe, which was only been tested on 10 people? The answer of course is no. In fact the only time the FDA will do such a thing is if the patient approved for the drug, has a very short life expectancy anyway.
So the only way around these testing volunteers shrinking number, is for the drug company to start with a larger pool to begin with. And for each restriction they apply they must enlarge the original testing pool even more, so that in the end they are testing at least 100 qualified people.
That's the same idea, that I must keep in mind each time I add a condition to a trading plan I am trying to develop. The more rules I add, the less statistically significant the results will be, unless I enlarge the data set so as to not restrict the number of data observations being evaluated.
तबसे मेंबर है Aug 02, 2011
38 पोस्टों
Dec 16, 2013 at 22:05
तबसे मेंबर है Aug 02, 2011
38 पोस्टों
I actually think humans and robots are similar. Both need to be programmed before they can function. In case of trading, both will first need the reasons to act; entry, exit, buy, sell, taking profit, cutting losses, etc.
The real difference in my opinion is that computers are much more efficient.
The real difference in my opinion is that computers are much more efficient.
Wait for it
![](/userimages.jpg?uid=136673&id=40202&w=100&h=100)
forex_trader_136673
तबसे मेंबर है Jun 28, 2013
852 पोस्टों
Dec 24, 2013 at 21:16
तबसे मेंबर है Jun 28, 2013
852 पोस्टों
Merry Christmas to all humans, robots are not excluded 😄
तबसे मेंबर है May 24, 2011
92 पोस्टों
Dec 24, 2013 at 22:07
तबसे मेंबर है May 24, 2011
92 पोस्टों
Merry Christmas
Best rebate for IC Markets 1,60 RTL+ armada markets 10% commision back + next 30 brokers company only in . Who will give you more?
तबसे मेंबर है Jul 02, 2013
57 पोस्टों
Dec 26, 2013 at 07:38
तबसे मेंबर है Jul 02, 2013
57 पोस्टों
Quants: The Alchemists of Wall Street - A Documentary about algorythmic trading
Money & Speed: Inside the Black Box
The Wall Street Code
Happy Holidays 2 All~ Enjoy!
Sleep is for the weak
![](/userimages.jpg?uid=136673&id=40202&w=100&h=100)
forex_trader_136673
तबसे मेंबर है Jun 28, 2013
852 पोस्टों
Jan 10, 2014 at 16:58
(एडिट हो रहा है Jan 10, 2014 at 17:03)
तबसे मेंबर है Jun 28, 2013
852 पोस्टों
I always wondered what robots do on weekend do after a good week of trading.
तबसे मेंबर है Mar 29, 2012
192 पोस्टों
Jan 10, 2014 at 17:02
तबसे मेंबर है Mar 29, 2012
192 पोस्टों
You must be right @ahuruglica ! MDP at the drums!
A smooth sea never made a skillful sailor.
![](/statistics/utv.jpg?id=677650&r=0.01361820838952632)
*व्यवसायिक इस्तेमाल और स्पैम को ब्रदाश नहीं किया जाएगा, और इसका परिणाम खाता को बन्द करना भी हो सकता है.
टिप: किसी चित्र या यूट्यूब या URL को पोस्ट करने से वे अपने आप आपके पोस्ट में आजाएगा!
टिप: @ चिन्ह को टाइप करें उपभोगता के नाम को अपने आप करने के लिए जो इस चर्चा में भाग ले रहा है.