You cant bs a bs-er. | Forex Factory

Options

Permalinks
Print Thread

Similar Threads

Cant download Mt4 files to new platform 6 replies

Cant update MT4 history 15M data 1 reply

HELP- I cant trade during NY market hrs. 9 replies

You cant bs a bs-er.

  • #1
  • First Post: Sep 20, 2008 8:21pm Sep 20, 2008 8:21pm
I just came accross this little gem out there in the interwebs.

Correct me if im wrong, but does this dose of verbal diarrhoea mean that they use multiple time frames (MTF)?

Summary:We investigate the effectiveness of a financial time-series forecasting strategy which exploits the multiresolution property of the wavelet transform. A financial series is decomposed into an over complete, shift invariant scale-related representation. In transform space, each individual wavelet series is modeled by a separate multilayer perceptron (MLP). We apply the Bayesian method of automatic relevance determination to choose short past windows (short-term history) for the inputs to the MLPs at lower scales and long past windows (long-term history) at higher scales. To form the overall forecast, the individual forecasts are then recombined by the linear reconstruction property of the inverse transform with the chosen autocorrelation shell representation, or by another perceptron which learns the weight of each scale in the prediction of the original time series. The forecast results are then passed to a money management system to generate trades
  • #2
  • Sep 20, 2008 8:56pm Sep 20, 2008 8:56pm
That is really funny. Thanks I needed a laugh. :nerd:
Forex is a Beast that will happily eat you for breakfast, lunch and dinner.
  • #3
  • Sep 20, 2008 9:03pm Sep 20, 2008 9:03pm
Quoting Ted1983
I just came accross this little gem out there in the interwebs.

Correct me if im wrong, but does this dose of verbal diarrhoea mean that they use multiple time frames (MTF)?
Sad part is that some will bite & be reeled in struggling all the way.
  • #4
  • Sep 20, 2008 9:08pm Sep 20, 2008 9:08pm
Dammmmm!!

They stole my system!!!
Polishing my crystal ball
  • #5
  • Sep 20, 2008 9:11pm Sep 20, 2008 9:11pm
No it doesn't.

If the term "window" was ambiguous, it refers to how many previous points they take into consideration, a window of 100 would mean the previous 100 prices or whatever type of data they use it on.

A wavelet transform takes a time series and discomposes it into continually coarser "levels", the last level being an approximation of the original signal, all the previous levels are details of the signal, think of them as being a type of noise.

You probably noticed I mentioned a wavelet-neuro system in tdion's thread, right? Largely it involves the process you just quoted with several important improvements I made.
  • #6
  • Sep 20, 2008 9:21pm Sep 20, 2008 9:21pm
up or down! ? hmm. thats all i see. If I read it right, also, they use elliot wave patterns to forecast the next pattern to get the best fit!.
Now, I know why the investment banks in the us went under.. they were out thinking everyone. Only problem with the diaherria is that patterns dont' always finish and patterns go whachy!. Think of how many times a fib did not reach the level/...50 % 38% ..78% 23.65 ..THAT CAN T BE COMPUTED.!... iD BE HAPPY IF THEY COULD MANAGE MY 401K! LET ALONE FORECAST!
  • #7
  • Edited 10:17pm Sep 20, 2008 9:26pm | Edited 10:17pm
A wavelet transform has absolutely nothing to do with Elliot Waves. It is a DSP technique somewhat similar to a Fourier transform.

Do you really think someone in an institutional environment would joke around with Elliot Waves? Whoaa, you all thought it was about Elliot Waves? That was the reason for the inside joke I just began to understand?

http://en.wikipedia.org/wiki/Wavelet_transform
  • #8
  • Sep 20, 2008 10:17pm Sep 20, 2008 10:17pm
Being poorly educated but dangerously curious, I felt the compulsion to look this one up:

Perceptron


.
  • #9
  • Sep 20, 2008 11:33pm Sep 20, 2008 11:33pm
From the research I've done NNs tend to draw big crowds, but don't tend to be that effective over time. I'm sure some goes to how they're used, tho.

Anyway, are wavelets accurate? I mean it all goes to the fractal nature of trading, but I've thought about it some and couldn't see how it would provide an edge. In terms of pure price, past history is a poor indicator of future results.
  • #10
  • Sep 20, 2008 11:45pm Sep 20, 2008 11:45pm
Yes, you're right.It depends a lot on how someone decides to use them and what kind of data they use them on.

Probably most people that end up being disappointed in them are the ones that don't know the principle "garbage in, garbage out" or that do not retrain their NNs daily if they want to predict a time series. Or they only try one particular type of NN, training method, performance function, transfer function and so on and then they give up.

Wavelets can be used to preprocess the data or as transfer functions and that's all I'm saying...already said quite enough.Also, I don't forecast prices, I forecast returns which I then convert to specific prices.As you know, returns are stationary.
  • #11
  • Sep 21, 2008 12:45am Sep 21, 2008 12:45am
I am pretty well versed in neural networks, and find that it is hard (if not impossible) to prove their validity.

Case and point: how far back do you trade the NN? What timeframe do you use?

Let's take it a step further.... I always thought the best way to make a forex NN would be to have the price be the output, and use fundamentals as the inputs (with price itself as a small weighted input.)

The idea is to get inside the bankers' heads.... to understand how THEY moved the market based on certain fundamentals.

Do functional dependencies exist between fundamentals and price activity? I think they do, but maybe only to produce a guess of the next month's price range. I think there is too much chaos in the lower timeframes....

And to those that toute Boatwright's NN, or Oleksandre's Better.... both have been sucking wind since their very impressive introductions to the public.
  • #12
  • Edited 1:14am Sep 21, 2008 12:56am | Edited 1:14am
Tdion, after the post exchange earlier I'm not going to actually reply to any of your posts addresses to someone else, or to me like in this case. You can't wait to point out the "mistakes" you think I'm making and quite honestly I've got better things to do with my time. Way too sick of you. I'm also not interested in your advice or exchanging opinions on NNs with you because you have nothing to offer me of use.
I already mentioned the essentials of my system, returns, wavelets, neural networks.You can google any of these for more info.Goodbye.

edit: I was the last poster, exactly why I wasn't supposed to think you were talking to me? Did you mention someone else's username? Did you use other types of pronouns than the ones I noticed?
  • #13
  • Sep 21, 2008 1:00am Sep 21, 2008 1:00am


Considering I didn't mention wavelets or ANYTHING that you brought up for discussion, I don't see how you can think I was addressing anything TO YOU, AT YOU, or AROUND YOU.

You have a complex. I was speaking to the general idea of forex NNs, which was brought up by rabid.

I will continue posting my ideas on every thread that I see fit, regardless of whether you are posting there or not.
  • #14
  • Sep 21, 2008 6:38am Sep 21, 2008 6:38am
Come on guys I was trying to have a serious discussion here so why move the thread to the Rookie section? Just cause its not a serious title doesnt mean it can't generate a decent thread.

Anyway I looked at the Wiki link for the Perceptron which appears to be a piece of 1950s technology. OK I know maths is maths but surely they could have used a more modern variant.

Oh and; tdion and tenantro.........jerry.....jerry....jerry....jerry!
  • #15
  • Sep 21, 2008 6:41am Sep 21, 2008 6:41am

Quoting Ted1983
Oh and; tdion and tenantro.........jerry.....jerry....jerry....jerry!
  • #16
  • Sep 21, 2008 11:06am Sep 21, 2008 11:06am
Quoting Ted1983
Summary:We investigate the effectiveness of a financial time-series forecasting strategy which exploits the multiresolution property of the wavelet transform. A financial series is decomposed into an over complete, shift invariant scale-related representation. In transform space, each individual wavelet series is modeled by a separate multilayer perceptron (MLP). We apply the Bayesian method of automatic relevance determination to choose short past windows (short-term history) for the inputs to the MLPs at lower scales and long past windows (long-term history) at higher scales. To form the overall forecast, the individual forecasts are then recombined by the linear reconstruction property of the inverse transform with the chosen autocorrelation shell representation, or by another perceptron which learns the weight of each scale in the prediction of the original time series. The forecast results are then passed to a money management system to generate trades
I'm glad this made sense to some of you guys. It just reads like a steaming heap of horseshit to me. I understand the difficulty of explaining complex technical concepts in terms the layman can readily grasp but come on! This person obviously didn't even try! Maybe that was the point.
  • #17
  • Sep 21, 2008 11:14am Sep 21, 2008 11:14am
Opie, it's from an article published in the IEEE, it's not targeted to the layman, it's for the scientific community. They have to mention all those terms and use that language, otherwise their article wouldn't be deemded rigurous enough and wouldn't be accepted for publishing. For someone working in the field of finance, AI or DSP, those terms are not unfamiliar.
  • #18
  • Sep 21, 2008 6:00pm Sep 21, 2008 6:00pm
Quoting tenantro
Opie, it's from an article published in the IEEE, it's not targeted to the layman, it's for the scientific community. They have to mention all those terms and use that language, otherwise their article wouldn't be deemded rigurous enough and wouldn't be accepted for publishing. For someone working in the field of finance, AI or DSP, those terms are not unfamiliar.
Thanks, tenantro. That's what I figured. My first clue this was the real deal was when I looked up perceptron and it turned out to be an actual word and that it seemed to be being used correctly within the context of the passage.

Now my brain hurts.
  • #19
  • Sep 21, 2008 6:43pm Sep 21, 2008 6:43pm
I don't know if this goes here or not but ...

I was reading an article in Futures mag this a.m. titled
"How Algo trading is changing the landscape"
I could understand the article but when I googled Algo Trading well that ended my understanding of the subject.

there was a illustration in the article (ha the pictures are easy) that caught my interest.
the caption said:

Slice & Dice
"Algo traders are able to detect and even create patterns that confuse slower traders. In the schematic below, an algo trader has correlated stocks according to a broad range of parameters and is waiting to buy in, attract, then sell and clean out slower traders."

sounds kind of ..like harvesting people Matrix style...

What's up with this stuff? Any suggestions on where to go to learn or should a mere mortal just pass on it.
  • #20
  • Last Post: Sep 21, 2008 7:37pm Sep 21, 2008 7:37pm
Stop hunting, algo style... laff. There's nothing new under the sun it seems.
0 traders viewing now