Skyline:
The results of your independent trial actually exceeds the results of my online program quite dramatically. Using the base criteria for the EUR on my analysis program, you'll see that the expectation is 3.0%
Your Results:
48 short  win 62.16%  lose 37.84%
37 long  win 54.17%  lose 45.83%
Total trades 85  win 57.65%  lose 42.35%
Total win $28,371.05  lose $28,352.94
((28,371.05/28,352.94) * .5765)  .4235 = 15.34% Expectation
My friend, if you can maintain just a 3% expectation you will bust the world eventually. With a 15% expectation, you'll own the universe.
One important aspect that must be factored in your calculations is to adjust the amount wagered (dollars used for purchasing  risk) to reflect the expectation relative to the bankroll. Your analysis essentially uses a flat bet (never changining  winning or losing) over the trial period which could mean over or under betting the bankroll. Check out my post regarding Kelly Criterian, "Never go Busted".
Middle analysis  Taking home the Pips
Moderator: moderators
Please add www.kreslik.com to your ad blocker white list.
Thank you for your support.
Thank you for your support.
Hi PitBoss,
Thx for your reply
uao I'm the master of the universe and I didn't know that eheh
I'm double checking your formula to calculate expectation for my EURUSD test and I barely calculate a 0.02% of Math Exp instead of 15%.
Of course I know that the lot size should be used as per kelly formula but just to do the test I used 1 fixed lot.
Here's my math supported by this site :
http://www.hquotes.com/tradehard/simulator.html
Math Exp = Prob*WL(1WL)
where
 Prob is the number representing probability of a winning trade of your trading system. For example, if you traded 100 times and won in 61 trades, then the probability to win is about 61% or 0.61.
 WL is defined as average winner divided by your average loser
said that in my system I get these figures :
Prob = 57.64% = 0.5764
W/L = Avg Profit trade / Avg Loss trade = 579$ / 787.58$ = 0.7351
thus final Math Exp should be :
Math Exp = Prob*WL(1WL) = 0.5764*0.7351(10.7351) = 0,00028515 = 0.02 %
So in your calculation what's wrong in my humble opinion is that you consider the whole gross profit as average profit and the whole loss profit as average loss.
Am I wrong ?
Thank you.
Cheers,
Skyline
Thx for your reply
uao I'm the master of the universe and I didn't know that eheh
I'm double checking your formula to calculate expectation for my EURUSD test and I barely calculate a 0.02% of Math Exp instead of 15%.
Of course I know that the lot size should be used as per kelly formula but just to do the test I used 1 fixed lot.
Here's my math supported by this site :
http://www.hquotes.com/tradehard/simulator.html
Math Exp = Prob*WL(1WL)
where
 Prob is the number representing probability of a winning trade of your trading system. For example, if you traded 100 times and won in 61 trades, then the probability to win is about 61% or 0.61.
 WL is defined as average winner divided by your average loser
said that in my system I get these figures :
Prob = 57.64% = 0.5764
W/L = Avg Profit trade / Avg Loss trade = 579$ / 787.58$ = 0.7351
thus final Math Exp should be :
Math Exp = Prob*WL(1WL) = 0.5764*0.7351(10.7351) = 0,00028515 = 0.02 %
So in your calculation what's wrong in my humble opinion is that you consider the whole gross profit as average profit and the whole loss profit as average loss.
Am I wrong ?
Thank you.
Cheers,
Skyline
 Attachments

 calc.jpg (74.01 KiB) Viewed 1333 times
Opps, brain fog!
You are correct. Should be:
( ( (579.00/787.58) * .5765 )  .4235 = .0003
Hardly enough to own the universe, lol.
However, the following fact was proven by your short sampling which was the basis of my initially continuing investigation of the phenominon:
If the open is lower than the previous middle, the close will be higher than the previous middle, over 50% of the time, and
If the open is higher than the previous middle, the close will be lower than the previous middle, over 50% of the time.
Now if an incident occurs on a consistant basis over 50% of the time, a clear and distinct bias is prevelent. Hence, the foundation of my concept.
If a person can accept the concept that at any arbritrarilly selected point, on a continigous line of closing prices, the price from the selected point is just as probable to go up as go down  then they'll have the equivilant of a 5050 proposition as to what direction the price will take  it is random.
Similar to throwing a dart to select a pair, then rolling a single die, even number go long  odd number go short. Over a statistically significant sampling size, one will have won as many times as they lost  no measurable advantage or disadvantage exists  it is random.
However, if an event occurs on a consistant basis over an extended period and it repeats itself virtually identically regardless of the pair without any types of "if this then do that" interpretative senerios, then it is beyond coincidence and becomes fact.
I found that this (over 50%) bias exits, for the most part, in the time frames of 15min, 30min, 60min, 4hrs and daily but not so in weekly or monthly (suspect insufficient number of datas) for each of the major pairs individually and collectively. Since it will occur on a consistant, noninterpretive basis, a positive expectation can be dirived.
I discovered that using the traditional tp and sl askewed the averages and that what ever it went up it went down an equal amount (or viceversa). There was no difinitive, consistant advantage created as a result.
Next I measured the average percentage from the middle to it's high (same percentage as it's low). What I discovered was that if the open was higher than that percentage * the open, then there was a greater propensity for the close to be below the middle and viceversa for the opposite direction. So what I had was a very high advantage but the numerical number of times the opportunity presented itself was too severly limited to effectively generate a regular profit.
After trying numerous experimental trials and beating up the math, I found that the optimum spread was open greater than the open price * 1.002 for the short entry and open less than the open * .998 for the long entry. It gave me a sufficient number of "betting" opportunities to maximize the potential advantage available.
Now a few points for you to consider in your evaluations:
1. Run same trial criteria for GBP, CHF and JPY to determine if basis is consistant.
2. Run trial for a much longer statistical period of time to determine if basis is consistant.
3. Run trial from 00:00 to 23:59 as opposed to 22:00.
4. Run trial with spreads as I defined just above.
5. Run trial with adjustments to bets reflecting Kelly Criterian.
6. Run trials on all 4 major pairs at the same time, measuring each day collectively, then the next and so on until the end of the trial period  you will be greatly surprised at your results  much better than individually.
Let me know what happens.
FrankV
You are correct. Should be:
( ( (579.00/787.58) * .5765 )  .4235 = .0003
Hardly enough to own the universe, lol.
However, the following fact was proven by your short sampling which was the basis of my initially continuing investigation of the phenominon:
If the open is lower than the previous middle, the close will be higher than the previous middle, over 50% of the time, and
If the open is higher than the previous middle, the close will be lower than the previous middle, over 50% of the time.
Now if an incident occurs on a consistant basis over 50% of the time, a clear and distinct bias is prevelent. Hence, the foundation of my concept.
If a person can accept the concept that at any arbritrarilly selected point, on a continigous line of closing prices, the price from the selected point is just as probable to go up as go down  then they'll have the equivilant of a 5050 proposition as to what direction the price will take  it is random.
Similar to throwing a dart to select a pair, then rolling a single die, even number go long  odd number go short. Over a statistically significant sampling size, one will have won as many times as they lost  no measurable advantage or disadvantage exists  it is random.
However, if an event occurs on a consistant basis over an extended period and it repeats itself virtually identically regardless of the pair without any types of "if this then do that" interpretative senerios, then it is beyond coincidence and becomes fact.
I found that this (over 50%) bias exits, for the most part, in the time frames of 15min, 30min, 60min, 4hrs and daily but not so in weekly or monthly (suspect insufficient number of datas) for each of the major pairs individually and collectively. Since it will occur on a consistant, noninterpretive basis, a positive expectation can be dirived.
I discovered that using the traditional tp and sl askewed the averages and that what ever it went up it went down an equal amount (or viceversa). There was no difinitive, consistant advantage created as a result.
Next I measured the average percentage from the middle to it's high (same percentage as it's low). What I discovered was that if the open was higher than that percentage * the open, then there was a greater propensity for the close to be below the middle and viceversa for the opposite direction. So what I had was a very high advantage but the numerical number of times the opportunity presented itself was too severly limited to effectively generate a regular profit.
After trying numerous experimental trials and beating up the math, I found that the optimum spread was open greater than the open price * 1.002 for the short entry and open less than the open * .998 for the long entry. It gave me a sufficient number of "betting" opportunities to maximize the potential advantage available.
Now a few points for you to consider in your evaluations:
1. Run same trial criteria for GBP, CHF and JPY to determine if basis is consistant.
2. Run trial for a much longer statistical period of time to determine if basis is consistant.
3. Run trial from 00:00 to 23:59 as opposed to 22:00.
4. Run trial with spreads as I defined just above.
5. Run trial with adjustments to bets reflecting Kelly Criterian.
6. Run trials on all 4 major pairs at the same time, measuring each day collectively, then the next and so on until the end of the trial period  you will be greatly surprised at your results  much better than individually.
Let me know what happens.
FrankV
Results on demo through today:
Made 51 trades, averaging 3 daily
Results include commission both ways
26 long  15 win, 11 lose = 57.7% win
25 short  14 win, 11 lose = 56.0% win
Totals 29 win 56.9%  22 lose 43.1%
Total win $5,229.77, average $180.34
Total lose $2,608.28, average $118.33
Net advantage 43.4%
Oh my gosh, I'm just waiting for the other shoe to drop!!
Made 51 trades, averaging 3 daily
Results include commission both ways
26 long  15 win, 11 lose = 57.7% win
25 short  14 win, 11 lose = 56.0% win
Totals 29 win 56.9%  22 lose 43.1%
Total win $5,229.77, average $180.34
Total lose $2,608.28, average $118.33
Net advantage 43.4%
Oh my gosh, I'm just waiting for the other shoe to drop!!
 RicG
 rank: 150+ posts
 Posts: 274
 Joined: Tue Jul 24, 2007 11:44 pm
 Reputation: 0
 Location: Nashville, TN
 Gender:
Hi pitboss,pitboss wrote:Results on demo through today:
Made 51 trades, averaging 3 daily
Results include commission both ways
26 long  15 win, 11 lose = 57.7% win
25 short  14 win, 11 lose = 56.0% win
Totals 29 win 56.9%  22 lose 43.1%
Total win $5,229.77, average $180.34
Total lose $2,608.28, average $118.33
Net advantage 43.4%
Oh my gosh, I'm just waiting for the other shoe to drop!!
I've been reading this thread with much interest. I appreciate you sharing this information.
I paper traded your method starting on Sunday and the first two days were quite impressive (8 winners, 0 losers). So far today the results are mixed. Obviously this is a small sample that doesn't reflect what happens over the longterm as is noted by your statistics.
As far as your comment, "Oh my gosh, I'm just waiting for the other shoe to drop!!", I think the "other shoe" is a trending market. If all four pairs that you are trading are strongly trending then you will suffer an extremely high losing percentage over an extended period of time. At that point it becomes a question of whether or not your bankroll can withstand that trend.
If you can determine when a market is trending then you would want to reverse your rules and:
Go Long when open is above the mid point
Go Short when the open is below the mid point
However, that is much easier said than done.
I think the good news in always using your method is the fact that most statistics I've read show that a market trends less than 50% of the time (30% is what I've heard). Does anybody on the board have reputable statistics concerning trending vs nontrending market percentages?
Anyway, keep up the good work and keep us updated on your progess and statistics.
Ric
(Disclaimer  This post is for educational purposes only. Always consult a licensed investment professional before taking any trade. Any trade you take is at your own risk.)
Please add www.kreslik.com to your ad blocker white list.
Thank you for your support.
Thank you for your support.
pitboss wrote:Opps, brain fog!
You are correct. Should be:
( ( (579.00/787.58) * .5765 )  .4235 = .0003
Hardly enough to own the universe, lol.
Hi Pitboss, thx for you exaustive explanations.
I'll try to perfom other tests but Metatrader backtest environment is really a pain in the ass and data aren't accurate to do that. Moreover it's not possible with Metatrader to run a backtest with more than 1 pair at once
As an example I ran again the backtest (using fix 0.1 lot this time) on EURUSD starting from 2005 up to May 2008 and result are even worst :
Profit Trades (% of total) = 50.06%
Avg Profit Trade = 44.57$
Avg Loss Trade = 49.72$
Exp = 0.5006*0.890.4994 = 5.3%
mostly caused by bad data available
I'll investigate further to see if it's possible to get in someway better historycal data.
Cheers,
Skyline
RicG wrote:
If you can determine when a market is trending then you would want to reverse your rules and:
Go Long when open is above the mid point
Go Short when the open is below the mid point
However, that is much easier said than done.
Ric
Hi RicG,
if you are able to predict when market starts and ends to trend , then you don't need at all any statistical approach, you'll be the owner of the universe in no time
From what I understand the point introduced here by PitBoss is to have an edge that could be statistically consistent to take profit from, optimizing at the same time the profit using the kelly formula.
Cheers,
Skyline
RicG:
Glad ur demo has been a success so far. Ur right in that it requires a much longer time period to realize statistical truths. In any short term period one could flip a coin (or any random method) to select buys and end up with an 80 result.
U'll find that this method is not influenced by trending. Upon closely analysing you'll see that in a clear and distinct trend of a pair that this method will have you going with the trend just as often as countering the trend. A trend is not a consideration to the method.
Also, support/resistance, volume, oscillators or other types of traditional indicators/considerations are not considered and are meaningless to the method. Now I'm not saying they are not important, nor am I saying that they could not be incorporated into consideration with the method  just at this stage they are not.
What I personally have found to be true with indicators is that they are open to a broad set of interpretation. Three different people can look at an identical indicator and you could get one long, one short and one hold off. The person with the interpretation that succeeded can thump their chest and say that they told you so with the other two pointing out reasons why their "setup" failed. In reality with a 100 given identical situations it could turn out that one of the other prognasticators would be correct!
The failing of any method is the people factor more so than the method itself. Using gambling as an example, if you were to have a positive advantage making a specific play in a specific situation  the problem most people experience is if the play has failed to succeed 10 times in a row, they tend to question the validity of the play, tinker with modifications and /or fail to pursue the correct play. However, the play could succeed the next 10 consecutive plays or fail the next 10 but the fact of the matter is the play will succeed overall if continuiously pursued  regardless of the most immediate results which is short term relative to the long term .
Pls keep me posted.
FrankV
Glad ur demo has been a success so far. Ur right in that it requires a much longer time period to realize statistical truths. In any short term period one could flip a coin (or any random method) to select buys and end up with an 80 result.
U'll find that this method is not influenced by trending. Upon closely analysing you'll see that in a clear and distinct trend of a pair that this method will have you going with the trend just as often as countering the trend. A trend is not a consideration to the method.
Also, support/resistance, volume, oscillators or other types of traditional indicators/considerations are not considered and are meaningless to the method. Now I'm not saying they are not important, nor am I saying that they could not be incorporated into consideration with the method  just at this stage they are not.
What I personally have found to be true with indicators is that they are open to a broad set of interpretation. Three different people can look at an identical indicator and you could get one long, one short and one hold off. The person with the interpretation that succeeded can thump their chest and say that they told you so with the other two pointing out reasons why their "setup" failed. In reality with a 100 given identical situations it could turn out that one of the other prognasticators would be correct!
The failing of any method is the people factor more so than the method itself. Using gambling as an example, if you were to have a positive advantage making a specific play in a specific situation  the problem most people experience is if the play has failed to succeed 10 times in a row, they tend to question the validity of the play, tinker with modifications and /or fail to pursue the correct play. However, the play could succeed the next 10 consecutive plays or fail the next 10 but the fact of the matter is the play will succeed overall if continuiously pursued  regardless of the most immediate results which is short term relative to the long term .
Pls keep me posted.
FrankV
Skyline,
What you've just pointed out is the very reason that I elected to write a stand alone analysis program. I hate to be restricted within the confines of what I can or cannot do by some program (Metatrader, Tradestation, etc). If you can code in PHP and framiliar working with mysql, I'll share my code with you. (Used PHP so I could make available on the net).
I am going to download the pair's historical data from another source into my database and run an analysis on it so I can compare it to the Metatrader data results, just for curiosity sake.
As far as the acuracy of the data is concerned, I really don't believe that it is all that important or essential. It is conceiveable that if one were to take the time to make up and create a year's daily open, high, low and close it would work out the same as actual data. By the same, I mean that a satistical abnormality can be found which can be exploited one way or another.
Anytime an event will occur more than 50% of the time, a method can be developed to exploit the bias to generate a positive expectation. (Make $$, lol).
What you've just pointed out is the very reason that I elected to write a stand alone analysis program. I hate to be restricted within the confines of what I can or cannot do by some program (Metatrader, Tradestation, etc). If you can code in PHP and framiliar working with mysql, I'll share my code with you. (Used PHP so I could make available on the net).
I am going to download the pair's historical data from another source into my database and run an analysis on it so I can compare it to the Metatrader data results, just for curiosity sake.
As far as the acuracy of the data is concerned, I really don't believe that it is all that important or essential. It is conceiveable that if one were to take the time to make up and create a year's daily open, high, low and close it would work out the same as actual data. By the same, I mean that a satistical abnormality can be found which can be exploited one way or another.
Anytime an event will occur more than 50% of the time, a method can be developed to exploit the bias to generate a positive expectation. (Make $$, lol).
Please add www.kreslik.com to your ad blocker white list.
Thank you for your support.
Thank you for your support.