A free data mining report on your trades worth $1500 +

Dear trading friends,

I have invested close to 5000 on data mining knowledge and programming plus another 2500 on processing power and more than 6 years on learning (and another 12,500) on quantitative financial analysis techniques. As a result I made a big return on my investment (although outside of forex world - I am now applying some of this knowledge to forex).

I would like to offer a totally free, 100% no marketing, sales, upgrades, signup etc analysis of your live trading over the last 3-6 months, if you think that will be valuable. There are no catches. I will do the work. If you want to offer in return, I am looking for an expert in MQL for some easy partnership (very basic requests since I don't code in MQL4). This is not neccessary though, just if you are feeling generous also.

If you have:

= Traded more than 200 trades on ONE specific pair over any time frame (if you have traded 1000 times it is better - more data the better).
= Trade off M5 or M15 charts generally
= Doesn't matter if you win or lose overall, so long as there are at least 30% positive pip trades
= Volume does not matter, I am only interested in pips
= And can prepare for me the detailed MT4 report of your trades
= Trade in a discretionary or systematic way, but manually rather than using an EA.

I can:

= Perform a data mining analysis of what predicts your successful trades, using indicators of my choice.

The reason I am asking is that although I trade a lot (I have around 400 trades from the last 2 months), including 200 on one specific pair, it is a small amount of data.

However, using data mining on an 8-core machine with 16GB of RAM (I know I should upgrade to 32GB of RAM), I am starting to achieve up to 75% accuracy in predicting whether or not my trade will be successful.

Interpreting the data is not easy, but it could lead to being able to create an expert which literally signals a positive trade using YOUR trading style.

If you want to discuss this with me, questions will be answered here, or via PM. All data we exchange will be 100% confidential, I won't expose your trades to others. I am only interested in analysing the data for you and seeing whether we can find patterns in your trading which will make you more successful.

Because of the time intensive nature of carrying out this kind of data preparation, cleaning and analysis, I will limit the offer of cooperation only to the first THREE traders who are able to submit a valid dataset according to the above conditions.

Hope that is a helpful contribution to the forum. I'm not sure what else I can offer at the moment other than that.

About me: I have a PhD using quantitative disciplines. I have been in financial markets for nearly 10 years on and off. I have a tiny but successful (so far) hedge fund start up and companies registered in the UK.

Looking forward to see if anyone sees value in this thread and offer. I don't promise to spend hours answering questions I don't deem too relevant. I'm basically offering something for nothing, so if you want to take further advantage, I am not willing to engage. If you are proactive and seeking some new insight and possibly partnership in your trading, great. Just a Sjunday thought.

Best regards,

Sam.

Hi Sam,
In a nutshell, I train a HMM (hidden Markov model) with the history of my trades to predict the probability of the success of my next trade. What is your methodology?

P.S. I have a good knowledge of MQL language and linking R language with MQL programs.

Best,
Math

Dear Math,

I am data mining, so I am throwing in variables, a small number of lags, their combinations and their transforms and testing several thousand model specifications and methods on several training and validation datasets during training and then seeing how models perform on a separate test dataset. I mainly use STATA, although it is hopeless for neural networks, so I use a purchased kit probably which relies on MATLAB or similar.

I am working on defining 'good' trades using a much bigger dataset than my own trades as well.

Take care,

Sam

//----

hey sam..... here's 3972 trades from a laid back yahoo trader...... the explorer has always been fully public..... 86% win rate..... data mine away......

my 100% win rate explorers are private..... but can say in advance, my next 27 trades will be no doubt winners...... otherwise they wouldn't be 100% ya know..........h

Hi Sam,

My brother and I are programmers and have developed a scalping EA that trades 1 minute charts.
We wanted the robot to be fully automatic but from our tests we could not find consistency. With that, we saw it fit to make the robot be like an "indicator service" where if the trade entry conditions are met, the EA sends a notification and email to us and so with human discretion, we can take the trade or not. We have been used to this method but we still lack one thing... Proper Optimization of our entry parameters.
The MT4 strategy tester is just not fit for our robot. To be more specific we believe we need data mining expertise where we have coded the robot to collect the data when it meets the entry conditions in .csv file. We believe if we share the data with a data scientist they will be able to help.
If your offer is still on (since you gave your offer to the first 3 traders and I can see from this thread I am the 3rd person), maybe I can give you more information on how the csv is structured and you can tell us if you can be of assistance and maybe we can move on from there?
Thank you.

Hi Sam can you assist. I am not able to private message you. Can we talk on email. My email is blaisekinyua[at]gmail[.]com

I really need data mining assistance.

Thanks for this data. Because it is multiple pairs, much harder to work with, but if there are say 1-2000 of a single pair it might be alright to look at. Cheers.

Hi, I've been working on machine learning based strategies for several months and have had trouble getting over the hump to where I'd really feel comfortable to start live trading. I've recently arrived at the conclusion that the difficulty in approaching trading from the ML angle appears to properly valuing each tradable point in time (e.g., the closing moment of each bar) in order to properly train a model. The issue is that models have to be trained in a way that aligns with actual trading rules and then the model inputs have to align with both the valuation method and the rules. Additionally, as the time frames get longer, more data is required, which is more easily overcome for backtesting rule-based strategies (i.e., the entire dataset can be tested rather than just a test set).

So I'm considering attempting to create a framework for evaluating different indicators for a rule-based strategy. The idea would be to perhaps create a matrix of sign-based entry indicators (i.e. 1=long, -1=short, 0=neither), general trade/abstain indicators (1=ok to trade, 0=no trade), and exit indicators (could be binary or ternary) and then evaluating different combinations of each.

I'm curious if you're doing anything that resembles this or have any suggestions.