t3 contest | Page 2 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Challenge - Oanda contest 130 replies

Lets see who the Boss is - Oanda contest 23 replies

vt trader contest 0 replies

Attachments: t3 contest
Exit Attachments

t3 contest

  • #21
  • Mar 17, 2010 9:16am Mar 17, 2010 9:16am
Quoting cspip
Thats great

But don't you need to know the current price or close price of the pair cuz the T3 uses that for its values?
yes the basic t3 done some calculation on the 14 close price(i say 14 bacause i've good result on that) so if you want to have the next value you have already 13 periods and with some calculations you can say where the t3 goes...i can't say any more....sorry....

i have to say to everyone that my function works with the same result on every tf and every cross...


and with more periods i've more precision...
  • #22
  • Mar 17, 2010 9:20am Mar 17, 2010 9:20am
Quoting andrea182
yes the basic t3 done some calculation on the 14 close price(i say 14 bacause i've good result on that) so if you want to have the next value you have already 13 periods and with some calculations you can say where the t3 goes...i can't say any more....sorry....

i have to say to everyone that my function works with the same result on every tf and every cross...


and with more periods i've more precision...
I understand you don't want to give out code etc.....

But I was thinking if you can calculate the next value of T3, then by going backwards in the T3 indy formula, then can't you get the next value of the price?
  • #23
  • Mar 17, 2010 9:21am Mar 17, 2010 9:21am
Quoting cspip
I understand you don't want to give out code etc.....

But I was thinking if you can calculate the next value of T3, then by going backwards in the T3 indy formula, then can't you get the next value of the price?

This is what I've been saying.

T3 only takes the price as an input ( I think) so logically, it follows that if you can guess the next T3 value, you can use the previous 13 known price values to calculate the 14th (unknown) value.
  • #24
  • Mar 17, 2010 9:24am Mar 17, 2010 9:24am
you can "predict" every MA with such accuracy but it won't really help you. Been there, done that.

try to predict the difference!
  • #25
  • Mar 17, 2010 9:25am Mar 17, 2010 9:25am
Quoting 7bit
you can "predict" every MA with such accuracy but it won't really help you.

try to predict the difference!
could you elaborate?

Difference between what?
  • #26
  • Mar 17, 2010 9:30am Mar 17, 2010 9:30am
Quoting Ronald Raygun
could you elaborate?

Difference between what?
difference between the value...

i've good result too...i'll post the screenshot this evening because i haven't matlab on the office pc....

and for the other questions if you goes backword from t3 to close value you should have more error beacuause the value you want to predict gives about the 7% of the value of the t3 so the error is only inside that so it goes up very high....
  • #27
  • Mar 17, 2010 9:30am Mar 17, 2010 9:30am
Quoting Ronald Raygun
could you elaborate?

Difference between what?
the first difference. if i(t) is your indicator then first calculate the difference

di(t) = i(t) - i(t-1)

and then try to predict this function. This will show you (if successful) when the indicator will turn around.

You can then simply integrate this prediction again to get the original indicator values. It will look less accurate but much more realistic.
  • #28
  • Mar 17, 2010 9:38am Mar 17, 2010 9:38am
I'm probably not understanding this properly, but looking at the graph on the first page, andrea's function is capable of predicting both direction and magnitude.

As I see it, if the predictive powers on this indicator is used on a high enough timeframe, there is potential for a scalping style system. If the next bar's T3 value is known, and the current T3 is far from that T3 value, the price is expected to move in such a way so that the current T3 value is closer to the predicted T3 value.

Let's say the predicted T3 value is 1.50000 (by the close of the bar) and the current value (open of the bar) is 1.45000. In order for the T3 value to move from 1.45000 to 1.50000, the price which the T3 is based off of needs to move up so that the current value (T3 value before the close of the bar) moves closer to 1.50000
  • #29
  • Mar 17, 2010 9:46am Mar 17, 2010 9:46am
Quoting andrea182
difference between the value...

i've good result too...i'll post the screenshot this evening because i haven't matlab on the office pc....

and for the other questions if you goes backword from t3 to close value you should have more error beacuause the value you want to predict gives about the 7% of the value of the t3 so the error is only inside that so it goes up very high....
If that's the case, then all you're left with is a multi-timeframe strategy of some sort.

If your prediction method works on multiple timeframes, (including nonstandard ones 10m, 20m, ticks, etc), then using the predicted values, guess where the T3 will turn. Once you find a turning point at the same time on multiple timeframes, the entry should be accurate enough to catch a breakout in progress, or before it happens.
  • #30
  • Mar 17, 2010 9:56am Mar 17, 2010 9:56am
Quoting Ronald Raygun
If that's the case, then all you're left with is a multi-timeframe strategy of some sort.

If your prediction method works on multiple timeframes, (including nonstandard ones 10m, 20m, ticks, etc)
i don't have tested nonstandard tf because i don't have that data...

but should works well because i've tested a lot on all standard timeframe and a lot of currency...
  • #31
  • Mar 17, 2010 10:01am Mar 17, 2010 10:01am
Quoting Ronald Raygun
I'm probably not understanding this properly, but looking at the graph on the first page, andrea's function is capable of predicting both direction and magnitude.
of course all the values seem to be very close to where they should be. But this is an optical illusion, it looks better than it really is. It is easy to predict that the value will be near the last value (the absolute value). Thats also where these incredible percent numbers come from: it is very likely that the next M1 bar won't be off a few hundred pips but this does not make it 99% accurate. it can be 99% "accurate" by this measurement and still be completely wrong.

Look closer and examine only the differences to the previous value. here is where accuracy is needed! Try to find any spots where it actually predicted a reversal to the bar exact.
  • #32
  • Mar 17, 2010 10:03am Mar 17, 2010 10:03am
andrea182: what model do you use for your prediction?
  • #33
  • Mar 17, 2010 10:06am Mar 17, 2010 10:06am
So predicting price is absolutely out of the question here, that part makes sense.

What about direction?

The prediction seems to be the farthest from the actual on the 3rd and 4th data points from the right. I don't have the raw data to dig deeper into the differences between the derivative of the predicted and actual values, (andrea, if you please), but I'd imagine it could be useful for something somewhere.
  • #34
  • Mar 17, 2010 10:43am Mar 17, 2010 10:43am
i've attached the chart of the difference...

see on le left is less then 1 pip because is at 10*-5....

that's some data of eurusd...
Attached Image(s) (click to enlarge)
Click to Enlarge Name: untitled_diff2.jpg Size: 36 KB
Click to Enlarge Name: untitled_diff.jpg Size: 35 KB
  • #35
  • Mar 17, 2010 10:47am Mar 17, 2010 10:47am
and that's the result of this matlab function corcoeff()( you can find it here:http://www.mathworks.com/access/help.../corrcoef.html)

correlation in differences

Cd =

Column 1

1.0000
0.9990
  • #36
  • Mar 17, 2010 10:59am Mar 17, 2010 10:59am
Quoting andrea182
and that's the result of this matlab function corcoeff()( you can find it here:http://www.mathworks.com/access/help.../corrcoef.html)

correlation in differences

Cd =

Column 1

1.0000
0.9990

Will this need to be retrained as time goes on, and how long will it stay accurate before it needs retraining?
  • #37
  • Mar 17, 2010 11:20am Mar 17, 2010 11:20am
Quoting onsmash
Will this need to be retrained as time goes on, and how long will it stay accurate before it needs retraining?
my code takes about 1000 previous data and gives the next it's very faster to calculate the result( in matlab less than 1 seconds)..

but the thread is started for another scope....

how can i use this prediction??
  • #38
  • Mar 17, 2010 2:24pm Mar 17, 2010 2:24pm
Quoting andrea182
how can i use this prediction??
How can can the community help you completing the other half of your project if you don't share the code you already have? Publicly sharing all code is a requirement for developing software with the community.

Otherwise you should just hire some paid programmers.
  • #39
  • Mar 17, 2010 2:30pm Mar 17, 2010 2:30pm
Quoting 7bit
How can can the community help you completing the other half of your project if you don't share the code you already have? Publicly sharing all code is a requirement for developing software with the community.

Otherwise you should just hire some paid programmers.
yes but what change if i share my code??you cannot improve the prevision...so i'll send only to someone help me to find a high precision strategy...

that's why i have create a thread called t3 "contest" so if someone want my code should give me the strategy...
  • #40
  • Mar 17, 2010 2:49pm Mar 17, 2010 2:49pm
Quoting andrea182
that's why i have create a thread called t3 "contest" so if someone want my code should give me the strategy...
You would have more and better responses and strategies to chose from if you just shared the code and everybody could experiment with it or even improve it. You would not lose anything, you would instead profit from it.
0 traders viewing now