This forum was established to help traders (especially futures traders) by openly sharing indicators, strategies, methods, trading journals and discussing the psychology of trading.

We are fundamentally different than most other trading forums:

We work extremely hard to keep things positive on our forums.

We do not tolerate rude behavior, trolling, or vendor advertising in posts.

We firmly believe in openness and encourage sharing. The holy grail is within you, it is not something tangible you can download.

We expect our members to participate and become a part of the community. Help yourself by helping others.

You'll need to register in order to view the content of the threads and start contributing to our community. It's free and simple, and we will never resell your private information.

I view the order book as bounding the next trade. The next price action should not exceed the current ask or be below the current Bid. It is more of a potential trading range, but not a guantee of price action.

I am really curious about , if you can show me the results of orderbook after taking into machine along with price ..
I will be thankful . if you discuss ur observations .

i meant,, did you experiment with orderbook data into SVM .. may be the spread or the order volume , etc . if yes.. did u get any noticeable clear results from the machine.

I don't want to hijack this thread for an SVM discussion as there already exists a thread, but the quick answer is I have not used Order Book to date. I don't think it sends the correct picture to the hyperplane as either Bid or Ask may not ever occur. Also, generalization to other similar situations doesn't not seam logical to me. Maybe order book with some pre-processing, but again, does data that "did not occur" prior to price action event matter to an SVM.

Hi asimsaim, thanks for your input on my thread - much appreciated.

Yes, take your points regarding indicators and correlation. HMM's are something I am playing around with primarily to learn about them and the Matlab platform. I do not expect to discover some secret trading sauce, I am just interested in the concept and can see it may provide a different class of information which I may be able to reference in my manual trading. Time and effort will tell. I expect to kiss a lot of frogs.

Your specific suggestion regarding increasing the number of states is a valid one as HMM's are 'supposed' to become more accurate by increasing the number of states. Computing power is the limiting factor as the iterations grow exponentially as you n+1. It's factorial. My latest runs are with 5 states (5!, 5x4x3x2, 120 options per iteration). I'm sampling 100 days of hourly data points on my latest backtests which takes about 4hrs to complete. I'd love to play around with a higher number of states but there are practical limits.

@liciobruno The attached may be of help, mostly research papers and a list of link, presentations, etc on Markov and Hidden Markov.

And yes some Matlab help would be greatly appreciated (I am assuming you have the HMM functions from the statistics toolbox?).

I have called the hmmviterbi function in my code so it returns the probability for each state as time moves n+1. My current model has 6 states so for each time slot the output is 6 probabilities (summing to 1) and the most probable is the predicted state.

This is a fairly basic point, but essential for proper application to a trading model! I can not work out if the output of hmmviterbi probabilities applies to i) the same data point (hourly time slot in my case), or ii) the next data point. Clearly the former is fine to develop a strategy in a backtesting environment, but worthless for live trading - unless it can offer a 'prediction' for the next data point.

So clearly I want to backtest only using an algorithm that can offer a prediction for the next state in the next time slot (a forward algorithm).

I think I am making some progress overall, but do not want to discovery later down the road that the approach has no practical application!

I will post up the NT and Matlab code shortly - it is a bit of a mess and I am no coder so be prepared for that!

know thyself

The following user says Thank You to mokodo for this post:

I will start to study it, but it's a little hard for me to read in English, so, please, can you say me from which file is better to start and have a first global view of the topic, and at the same time is more useful for your project?

And yes, I have the HMM functions from the statistics toolbox, so I can start to explore them.

I wait for your instructions.

Last edited by liciobruno; January 8th, 2013 at 07:36 PM.