hmmmm... instead of evaluating pattern's predictive power, you should evaluate it with a strategy. Predictive power is only about win rate but you need expected value to make it work.
1
(binned per thread starter's request) Researching and testing
- #42
- Edited 4:31am Jun 2, 2024 4:18am | Edited 4:31am
- Joined Sep 2008 | Status: Trader | 2,275 Posts
Moin. I've added 'reversal detection' - or name it better, - based on Stochastic, because I saw that condition somewhere. lw - lower wick uw - upper wick bd - body sk - StochasticK with lenght=14 lw > (bd*2) && bd < 50 && uw < 2 && sk < 20 → (lower wick is 2 or more times bigger then a body, body occupies less then 50% of a range, upper wick is less then 2% of a range) sk = 100 * (close - lowest(low, length)) / (highest(high, length) - lowest(low, length)) Final dataset was reduced from 496 to 72 casesQuantiles. {image} Random bear...
Is it really true that we have a hammer and stochastic "oversold" sk<20:
50% 0.0107 to 50% -0.0137 ??
also invertet hammer with context sk<20:
50% 0.0132 to 50% -0.0148
These figures are horrible.
What supports the thesis that banks and marketmakers "use" old (working) strategies out of trading books to suck people into market who apply it and take their money. I think the easier the signal is generated, the easier it is possible to reverse it (costs of reversing is little compared to the money you earn). That would explain why signals like hammer or inverted hammer are not working solely anymore - or even are abused to suck people onto a wrong path.
Now the question is: can you reduce the negative outcomes of hammers by using a SL to avoid max. DD? Here in the test it is only the highest point and lowest point after a certain amount of time used. A SL could change that outcome. Due to my experience a SL is useless. You reduce your max. DD but you also have more losing trades with reduces the max. avg. profit a lot.
What we can see here is using another technique to avoid bad trades:
T7 21 doesn't enter on the signal but later, when the signal get confirmed. - Of course the main signal than is already gone but the power of it could last longer like in the example from T7 21. So you use the signal only after a pullback. That is a very common entry strategy for breakouts. You wait for the confirmation. But this is also a good example how complex a backtest can get if you really want to backtest it automatically. Due to my experience I can say that T7 21's approach of backtest will probably on the limit to test with one computer (even with the power of today's cpu) on 1 minutes base for one year for one currency.
- #44
- Jun 2, 2024 5:06am Jun 2, 2024 5:06am
- Joined Dec 2020 | Status: Re█████ | 2,873 Posts
Important reference post
Stats for GU, hourly
If one decides to go short after any bearish candle and wait for (t) <=100 hours, there is 80% chance to gain 35 pips (green)
If one decides to go long after any bearish candle and wait for (t) <=100 hours, there is also 80% chance to gain ~35 pips (yellow)
Or go long/short after any bullish candle
If you see an arrogant 'experienced' trader, who has ben trading patterns for ~20 years, show him this.
Stats for GU, hourly
If one decides to go short after any bearish candle and wait for (t) <=100 hours, there is 80% chance to gain 35 pips (green)
If one decides to go long after any bearish candle and wait for (t) <=100 hours, there is also 80% chance to gain ~35 pips (yellow)
Or go long/short after any bullish candle
If you see an arrogant 'experienced' trader, who has ben trading patterns for ~20 years, show him this.
Air force
If one decides to go short after any bearish candle and wait for (t) <=100 hours, there is 80% chance to gain 35 pips (green) If one decides to go long after any bearish candle and wait for (t) <=100 hours, there is also 80% chance to gain ~35 pips (yellow)
If you see an arrogant 'experienced' trader, who has ben trading patterns for ~20 years, show him this.
- #46
- Edited 7:01am Jun 2, 2024 6:00am | Edited 7:01am
Suppose we go long and set a TP at +35 pips. In those 20% of the cases the TP was not reached, what prices did it touch?
In those 80% of the cases the TP was reached, what prices did it touch before reaching the TP?
It is this last graph that shows why it is not trivial to make money even if prices reach +35 pips in 80% of the cases. Even if you set a very wide SL at -140 pips (=Break even, (80%/20%)x35 pips), it will be touched in 2.5% of the cases that would otherwise eventually have hit TP.
In those 80% of the cases the TP was reached, what prices did it touch before reaching the TP?
It is this last graph that shows why it is not trivial to make money even if prices reach +35 pips in 80% of the cases. Even if you set a very wide SL at -140 pips (=Break even, (80%/20%)x35 pips), it will be touched in 2.5% of the cases that would otherwise eventually have hit TP.
- #50
- Jun 2, 2024 7:38am Jun 2, 2024 7:38am
- Joined Jun 2013 | Status: Trader | 14,594 Posts
Important reference post Stats for GU, hourly If one decides to go short after any bearish candle and wait for (t) <=100 hours, there is 80% chance to gain 35 pips (green) If one decides to go long after any bearish candle and wait for (t) <=100 hours, there is also 80% chance to gain ~35 pips (yellow) {image} Or go long/short after any bullish candle {image} If you see an arrogant 'experienced' trader, who has ben trading patterns for ~20 years, show him this.
like previous daily candle or more, bearish or bullish to now the internal hourly structure.
i saw once stats of something similar, if last 2 daily candles are green, place buy after bearish hour
- #51
- Jun 2, 2024 7:44am Jun 2, 2024 7:44am
- Joined Dec 2020 | Status: Re█████ | 2,873 Posts
{quote} Omg, the results are even negativ if I interprete it correctly. Is it really true that we have a hammer and stochastic "oversold" sk<20: 50% 0.0107 to 50% -0.0137 ?? also invertet hammer with context sk<20: 50% 0.0132 to 50% -0.0148 These figures are horrible. What supports the thesis that banks and marketmakers "use" old (working) strategies out of trading books to suck people into market who apply it and take their money. I think the easier the signal is generated, the easier it is possible to reverse it (costs of reversing is little compared...
Is it really true that we have a hammer and stochastic "oversold" sk<20:
50% 0.0107 to 50% -0.0137 ??
also invertet hammer with context sk<20:
50% 0.0132 to 50% -0.0148
it's a horror
but I'll continue to test other patterns this way, even if its showing no_better_than_random results so far.Now the question is: can you reduce the negative outcomes of hammers by using a SL to avoid max. DD?
Off-topic: similar idea with high probability but requires very high DD tolerance https://www.forexfactory.com/thread/post/14821764#post14821764
Air force
- #52
- Edited 10:38am Jun 2, 2024 10:28am | Edited 10:38am
- Joined Dec 2020 | Status: Re█████ | 2,873 Posts
Suppose we go long and set a TP at +35 pips. In those 20% of the cases the TP was not reached, what prices did it touch? {image} In those 80% of the cases the TP was reached, what prices did it touch before reaching the TP? {image} It is this last graph that shows why it is not trivial to make money even if prices reach +35 pips in 80% of the cases. Even if you set a very wide SL at -140 pips (=Break even, (80%/20%)x35 pips), it will be touched in 2.5% of the cases that would otherwise eventually have hit TP.
- Let's take TP {20 < x < 35} pips
- analyze positions in a sequence at which TP was reached and position of min_price
- on a scatter plot we would like to see points are clustered above the line of equality (x=y)
Under this conditions price hits tp in 85% of the cases before going too deep: {-0.0102, -0.01, -0.0154, -0.0119, -0.0109, -0.0085}
Please check for 154 pips with your algorithm. Is it the same thing?
In post#2 there was a table attached for a sequence of 200 candles
Here I used data from 100
Need to check how it goes with random candles...
Air force
- #53
- Edited 1:40pm Jun 2, 2024 1:27pm | Edited 1:40pm
- Joined Dec 2020 | Status: Re█████ | 2,873 Posts
Dark cloud
1. Without swing detection, only candle shape
pc - previous candle
cc - current candle
[n] - history reference operator, i.e. close[2] - close 2 bars back
(*fail in comment, cc close < pc body mean)
Found 1647 cases =~5.3%
Because pattern is bearish, values < 0 is profit (green table)
Air force
- #54
- Edited 3:50pm Jun 2, 2024 1:43pm | Edited 3:50pm
Please check for 154 pips with your algorithm. Is it the same thing?
For example, here I made a scatter plot with TP=35 pips with both winning trades (the rightmost dots) and trades that never hit TP in 100 hours:
Your scatter plot looks different. Can you please elaborate?
- #55
- Jun 2, 2024 5:29pm Jun 2, 2024 5:29pm
- Joined Dec 2020 | Status: Re█████ | 2,873 Posts
{quote} Happy to check, but I am not sure how to read your scatter plot. If you put the TP at 35 pips, and the dots represent all cases were the TP was hit, then I expected all the dots to be on a vertical line at PosMax=35 pips with various PosMin pips.
PosMax is a position in a sequence where the maximum high occurred. PosMin - position of lowest low
I set tp in the interval {0.002 < x < 0.0035}, otherwise x>=0.0035 all the extreme DD comes out and plot becomes noisy
For example:
and that case on scatter plot:
Dots above equality line indicates that profit achieved first
Air force
TP = 154 is too much for only 100 H1 bars. But with TP = 35 you can see clear patterns similar to your plot in post #55 (https://www.forexfactory.com/thread/post/14887349#post14887349). Unfortunately, a lot of dots are near the leftmost border which means price started to go south right after opening the Buy position.
- #58
- Jun 3, 2024 12:07pm Jun 3, 2024 12:07pm
- Joined Dec 2010 | Status: Foook Bollinger-dr.Kegel knows! | 9,794 Posts
