(binned per thread starter's request) Researching and testing | Page 2 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Mismatch between live forward testing and back testing 4 replies

Researching Saadscalper ea 3 replies

Researching Litewave expert 12 replies

Researching Admiral Markets Ltd 19 replies

Attachments: (binned per thread starter's request) Researching and testing
Exit Attachments

(binned per thread starter's request) Researching and testing

  • #21
  • May 30, 2024 4:53pm May 30, 2024 4:53pm
Attached Image (click to enlarge)
Click to Enlarge Name: Screen Shot 2022-09-17 at 13.09.14.png Size: 21 KB
  • #22
  • May 30, 2024 4:57pm May 30, 2024 4:57pm
Quoting T721
{image}
I think it is important to note that the shooting star or other candles are more powerful at a specific support/resistance levels and that pullbacks to 50% don;t happen all the time
  • #23
  • May 30, 2024 4:58pm May 30, 2024 4:58pm
Quoting T721
{image}
T721 here has very important share that highlights what is, for me, the most important factor when considering candle.s:

CONTEXT
If I don't mention SL and/or TP = longer-term setup possibly in the works!
  • #24
  • May 30, 2024 7:35pm May 30, 2024 7:35pm
My sincere desire is to save W0lfram from falling into a very deep rabbit hole, while unbeknownst to him, the topic he is pouring his cranium into, is something that has been beat to death by others. No sense in watching someone trying to milk a dry cow.

Probably more electricity has been spent testing trading algorithms than predicting the weather.
That's mean to be figurative, although, ... it might be true.

How about we reverse engineer the top trading algorithms, and maybe try to make one of them even better?
At least you are starting with an idea that is proven to work instead of crawling around in the dark looking for a gold nugget.
There ain't no use in cryin' It doesn't change anything." QOTSA
  • #25
  • May 30, 2024 8:32pm May 30, 2024 8:32pm
Quoting EgoNoBueno
My sincere desire is to save W0lfram from falling into a very deep rabbit hole, while unbeknownst to him, the topic he is pouring his cranium into, is something that has been beat to death by others. No sense in watching someone trying to milk a dry cow. Probably more electricity has been spent testing trading algorithms than predicting the weather. That's mean to be figurative, although, ... it might be true. How about we reverse engineer the top trading algorithms, and maybe try to make one of them even better? At least you are starting with an...
No need to save me. Thank you

I don't search 'holy grail', you probably meant that.
I like to mine the data and extract profit from my findings.
Air force
  • #26
  • May 30, 2024 8:40pm May 30, 2024 8:40pm
Quoting W0lfram
{quote} No need to save me. Thank you I don't search 'holy grail', you probably meant that. I like to mine the data and extract profit from my findings.
Got it. No saving required. No further saving efforts will be made.
I did not mean Holy Grail. I don't believe in the concept.

Watching to see if I can add anything useful.
My sincere hope that you get what you are after.
There ain't no use in cryin' It doesn't change anything." QOTSA
  • #27
  • May 30, 2024 8:50pm May 30, 2024 8:50pm
Quoting yoriz
As a proper scientist I tried to reproduce the results of a peer... (Well, actually I was just curious ;-)) I get the same results on GU. Another way to look at the data is to plot the minimum/maximum excursion of both a random sampling and the hammer pattern on top of each other. There is no significant difference: {image} I wondered whether perhaps the "hammer-effect" was just short-lived. So I did the same exercise on the M5 timeframe (i.e. detect a M5 hammer-candle and measure excursion over 200 M5 bars). Again no significant difference: {image}...
Quote
As a proper scientist I tried to reproduce the results of a peer.
Gut! Es ist echt toll, mit Ihnen zusammenzuarbeiten
Air force
  • #28
  • May 31, 2024 4:08am May 31, 2024 4:08am
https://thepatternsite.com/
  • #29
  • May 31, 2024 5:49am May 31, 2024 5:49am
Quoting mtako
{quote} T721 here has very important share that highlights what is, for me, the most important factor when considering candle.s: CONTEXT
Yes and No. I will explain in a moment because:

Quoting EgoNoBueno
My sincere desire is to save W0lfram from falling into a very deep rabbit hole, while unbeknownst to him, the topic he is pouring his cranium into, is something that has been beat to death by others...
Yes and No also.

Here is what I found out over years of research: IF there is an edge on sth. (e.g. here: Hammer) THAN it is showing in 95% of the cases. - NO MATTER if it is in context, if you use it correctly like T721 - the only difference in backtest is that with CONTEXT the results gets better (which can finally result in curve fitting or real edge).

That means: IF a Hammer has any sign of market advantage it is showing in your backtest (in 95% of the studies) - even if you analyze it out of context. The reason behind that is that in cases where a Hammer is not having an Edge, you have for example 53453 pips of win and for example 53212 pips of loss (for example) or vice versa. - That means: It equals out in the end BUT: The advantage after a candle sequence or whatever will always show in the stats - only the edge which is showing is less but the R:R over time or more instruments will show an edge. If that is not the case the pattern/candle or whatever is useless (as I said in 95% of the studies) - no matter which context.

So Wolframs analyze is very good. It covers the basic candles/patterns to start with. But I wonder that the hammer is neutral. It should at least give 1.5 - 2% of an edge. I remember studies where they showed the best candles/patterns and it was going up to 58% I think (I don't remember anymore which pattern it was, sorry )
  • #30
  • May 31, 2024 5:59am May 31, 2024 5:59am
Quoting Thtpro
https://thepatternsite.com/
I really appreciate your input. The problem with stats like that is the out of date datas tested and of course the gap candles/patterns which you don't really have on Forex - but of course you can eliminate that patters with gaps from end of candle to open of new candle.

What I learned on forex is: You need to be up to date: I give you an example: I traded very successfully EUR/SEK in 2023. - Exactly on end of December 2023 the system stopped to work and I lost about 50% of my earnings in 2023 on Jan 2024 - fair enough but I was wondering. After watching the pair for 2 months after Jan 24 I entered again slowly and I was profitable again BUT: The swings and the intra-day trading changed so strong that my old system stopped to work (!).

So you need much more research to be profitable in my opinion. But again: If there are easy patterns etc. they must work over let's say 30 different pairs if you test them or they will (probably in 95% of the cases) give no edge.
  • #31
  • May 31, 2024 7:26am May 31, 2024 7:26am
Quoting Thtpro
https://thepatternsite.com/
Thanks for the link. Will dive into
Air force
  • #32
  • May 31, 2024 10:22am May 31, 2024 10:22am
As you characterize and fingerprint the context of time based bars, consider that they have been found to resemble pink noise due to their fractal nature.

Tick bars vs Time based bars.
Better capture of market activity, offering more granularity during active market trading.
They are more adaptable to market volatility.
Reduced noise: Time-based bars can generate bars regardless of market activity.
Tick bars equalize market activity, creating bars of roughly equal size bars, which can make it easier to compare bars and analyze price action.
Tick bars are not influenced by time zones.
They are a more consistent measure of market activity.

I'm currently trying to get a tick bar generator indicator demo to work in MQL5. If you would like to be kept updated, let me know.

A Statistical Analysis Of The Predictive Power Of Japanese Candlesticks: https://scholars.fhsu.edu/cgi/viewcontent.cgi?article=1101&context=jiibr
There ain't no use in cryin' It doesn't change anything." QOTSA
  • #33
  • May 31, 2024 12:02pm May 31, 2024 12:02pm
Quoting JackintheBox
{quote} I really appreciate your input. The problem with stats like that is the out of date datas tested and of course the gap candles/patterns which you don't really have on Forex - but of course you can eliminate that patters with gaps from end of candle to open of new candle. What I learned on forex is: You need to be up to date: I give you an example: I traded very successfully EUR/SEK in 2023. - Exactly on end of December 2023 the system stopped to work and I lost about 50% of my earnings in 2023 on Jan 2024 - fair enough but I was wondering....
Yeah some things just work under certain market conditions - I don't look at or trade Candlesticks, I just use simple OHLC price bars

I just posted the web link as I remember the stats have been looked at by the Bulowski
  • #34
  • Edited Jun 1, 2024 5:46am May 31, 2024 4:24pm | Edited Jun 1, 2024 5:46am
Quoting EgoNoBueno
A Statistical Analysis Of The Predictive Power Of Japanese Candlesticks: https://scholars.fhsu.edu/cgi/viewcontent.cgi?article=1101&context=jiibr
Thanks! That was interesting to read. However, their conclusion is not useful for practical applications. They tested two cases.

When a Hammer forms, then...

1. price often still drops below the close-price of the Hammer bar -> no significant effect
2. price rarely drops below the low-price of the Hammer bar -> effect?

In the article they celebrate case #2, but if you think about it, that finding is useless! The low-price is reached at a time when the Hammer-bar is still forming. At that time we don't know whether it will become a Hammer. Only after the Hammer-bar is completed and closed do we know it is a Hammer bar. But then you are no longer at the low-price! Instead, you are dealing with case #1 (close-price) for which they could not find any significant effect.

Basically, their experiment contained snooping bias: they use the low-price of a bar that will only later become a Hammer bar.
  • #35
  • Edited Jun 1, 2024 3:53am May 31, 2024 4:46pm | Edited Jun 1, 2024 3:53am
Quoting Thtpro
I just use simple OHLC price bars I just posted the web link as I remember the stats have been looked at by the Bulowski
It appears to be possible to have an edge by comparing OHLC prices of the recent bars. A person named @neurotrader did some experiments: https://www.youtube.com/watch?v=2XQ3PsZActM Although very educational, you can skip the section about the genetic algorithm and jump to the Walk Forward Analysis at 11:53. Using a portfolio of 5 long and short patterns based on the last 3 bars, he is able to generate a bit of profit:
Attached Image (click to enlarge)
Click to Enlarge Name: screenshot.png Size: 203 KB


A few years back there was a group on Internet who built a program to data-mine such OHLC patterns. I forgot the name of that program. Something like "Kandu" or "Kanto", I believe. After a year they suddenly became a private group, named Asimoku of something exotic like that. They did not accept any new joiners anymore. I take that as a signal that they were indeed successul. (Sorry for the vague names. I tried to google these and similar names but could not find it anymore. Perhaps some oldtimer on FF knows what I am talking about).

Anyway, you seem to be able to make signals by comparing the O, H, L, C prices.
  • #36
  • Jun 1, 2024 5:42am Jun 1, 2024 5:42am
The approach by @neurotrader was a bit different from @W0lfram's candlestick patterns. But if we use different numbers instead of O, H, L, C we could try the same. In post #2 (https://www.forexfactory.com/thread/post/14882248#post14882248) @W0lfram described the rules for a Hammer. Also in the paper linked by @EgoNoBueno there is a description. Perhaps we should use these variables instead of comparing O, H, L, C directly:

Inserted Code
lw = min(O, C) - L    // lower wick
uw = H - max(O, C)    // upper wick
bd = abs(O - C)       // body length
range = H - L         // range

Given these variables, we could generate random patterns by chosing the relative sizes. For example, the Hammer would be 0.5, 0.5, 0.02:

Inserted Code
lw > 0.5 * bd  &&  bd < 0.5 * range  &&  uw < 0.02 * range


We could easily try to data-mine random values (or perhaps even do an exhaustive search) and see if we can find statistical significance in single-candle patterns.
  • #37
  • Jun 1, 2024 9:05am Jun 1, 2024 9:05am
Quoting yoriz
{quote} It appears to be possible to have an edge by comparing OHLC prices of the recent bars. A person named @neurotrader did some experiments: https://www.youtube.com/watch?v=2XQ3PsZActM Although very educational, you can skip the section about the genetic algorithm and jump to the Walk Forward Analysis at 11:53. Using a portfolio of 5 long and short patterns based on the last 3 bars, he is able to generate a bit of profit: {image} A few years back there was a group on Internet who built a program to data-mine such OHLC patterns. I...
indeed very interesting. I think you’re referring to OpenKantu, which has a GitHub repository where you’ll find the source code: https://github.com/danielfppps/OpenKantu
Unfortunately for me I can’t read Pascal or whatever it’s written in
Google for “Asirikuy” you’ll find the site
  • #38
  • Edited 11:52am Jun 1, 2024 10:32am | Edited 11:52am
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
Inserted Code
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)

Inserted Code
sk = 100 * (close - lowest(low, length)) / (highest(high, length) - lowest(low, length))

Final dataset was reduced from 496 to 72 cases
Quantiles.
Attached Image (click to enlarge)
Click to Enlarge Name: Mathematica_3eYhltuSo4.png Size: 8 KB


Random bear candle given sk<20

Attached Image (click to enlarge)
Click to Enlarge Name: Mathematica_NPKv2r1yUc.png Size: 8 KB


Attached Image (click to enlarge)
Click to Enlarge Name: Mathematica_LmqrzoK223.png Size: 5 KB


Random bullish candle given sk<20

Attached Image (click to enlarge)
Click to Enlarge Name: Mathematica_0trzreAEqS.png Size: 8 KB

Attached Image (click to enlarge)
Click to Enlarge Name: Mathematica_LmqrzoK223.png Size: 5 KB

Context (or location) doesn't matter and Hammers are no better than any random candle.

Inverted hammer 'anywhere'
Attached Image (click to enlarge)
Click to Enlarge Name: Mathematica_QjO6lw7FZC.png Size: 8 KB

and with "context"
Attached Image (click to enlarge)
Click to Enlarge Name: Mathematica_bEWaXUIRME.png Size: 8 KB

No significant difference from random samples.


Gonna play with number of forward_candles and another location rule, maybe we'll see different results.
Air force
  • #39
  • Jun 1, 2024 11:05am Jun 1, 2024 11:05am
Thanks everyone for links/feedback,.. very interesting content.

Polishing framework and moving further. Dark cloud/piercing line next.
The purpose is not to solve the market, but money management so to speak
Air force
  • #40
  • Jun 1, 2024 4:46pm Jun 1, 2024 4:46pm
Quoting Zennie
I think you’re referring to OpenKantu [...] “Asirikuy”
YES! That was it! Thank you for refreshing my memory.
0 traders viewing now