Forex trading from engineer with good background in mathematics and neural networks 8

[Konstantinos88]

Hi all.

I am engineer with a good mathematical and neural networks background:

In Mathematics I know very well all the aspects of Mathematical Analysis (including Lebesgue Measure and Integration), Numerical Analysis (including numerical methods of Differential Equations), Linear and non Linear Algebra (with applications in Signals and Systems, Discrete Time Signal Processing, Control Systems Design, Computer Vision) and Mathematical Logic. Moreover I know the most popular optimization algorithms.

In Neural Networks and Artificial Intelligence, I know very well all their aspects, for example Dimensionality Reduction, Multivariate Visualization, Classification Algorithms, Shallow Networks (Classification, Regression, and Clustering), Generalized Models, Mixed-Effects Models, Analysis of Variance, Fitting Distributions, Deep Learning Training, Pretrained Models, Accelerated Training with GPUs, and Preprocessing and Postprocessing.

Also, I know how to apply all of them in Matlab and Python, and the Computational Finance toolboxes of Matlab are 95% understandable by me:

https://www.mathworks.com/solutions/finance-and-risk-management.html

So my question is the following: I have decided to get into the forex online trading. How to do so? What are my first steps? What are my first expectations? For example if I start tomorrow, what I have to expect the first month, the first 6 months and the first year?

All the google is full of advice for people with Middle School Mathematical level and it kind of confuses me. The first advice is to focus on indicators (like RSI, MACD etc), to create automated trading robots (Expert Advisors etc) etc. And it is expected to have profits. Also metatrader 5 has a manual neural network choice, in which you can optimize your strategy (training) and then you can validate it (forward)

About me, I have to focus on this advice, or is there a different way to apply my knowledge in that field? For example can I take the historical data of eurusd and apply to it all the signal processing and neural networks toolboxes of Matlab and Python?

Simply said, what steps I have to follow on the beginning and what my expectations should be?

Thanks in advance.

 

[IRstuff]

Take money, place in toilet, flush -- rinse and repeat.

There have been far smarter people than you or me that have gotten burned. Why would you expect something different?



TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[Compositepro]

I just read something relevant yesterday about the topic of your question.

http://www.collaborativefund.com/bl...cket&utm_medium=email&utm_campaign=pockethits

 

[IRstuff]

Thanks, that was a great read.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[Konstantinos88]

IRstuff, I am looking for a serious answer. Trolling and spamming is never good.

For example, Matlab has many tools in

Financial Toolbox:

https://www.mathworks.com/products/finance.html

Financial Instruments:

https://www.mathworks.com/products/financial-instruments.html

Risk Management:

https://www.mathworks.com/products/risk-management.html

Trading:

https://www.mathworks.com/products/trading.html

All of these help you "throw your money in toilet"?

I am looking for guidelines for my first steps, for my starting point. For example, someone would say, that in my first steps I have to focus on eurousd with H1 or M5 time frames, and I should try to train a neural network in the last 3 years, using this training method, using real ticks and I should expect a mean monthly return of 1.015 implying an annual return of 1.2. Or that I should use the available indicators:

https://www.mql5.com/en/code/mt5/indicators

and I should create my own a strategy. After that I should optimize it the last 3 years (2 years training and the last year validation - forward), manually by using the available optimization techniques from brokers or trading platforms. Or in my first steps, I just have to focus my strategy to be a kind of coin toss, not losing money (50% profit trades - 50% loss trades) and the 1.2 annual return is more advanced?

I mean, we can rent supercomputers or other similar services from amazon or google, and we can train - optimize - validate a neural network using all the historical ticks from the past 10 years, so our neural network will be ready to predict the future of eurusd. But I believe, my i7 core with 8 threads and 16 GB ram is enough for my first fixed income.

Please, I hope there is someone who can guide me.

 

[TheTick]

Two words: black swan

 

[IRstuff]

Therein lies the problem; I was deadly serious, as was the article citation from CP.

The fact that you think past history is an analytical predictor for the future is problematic.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[HamburgerHelper]

This is a trap for smart people. Too smart to understand how dumb it is. Companies got Ph.D quants to do modeling and such and they still get it wrong, sometimes. Look how many banks got shredded 10 years ago.

I had a realtor act all proud of models that she and her engineer partner put together to model the real estate market. It was some complex thing that you could tell they put a lot of work into it. She didn't have much of an answer when I brought up that her models were consistently overshooting what the actual sale price was on almost every property I looked at. I had a better feel for the market just watching closing prices and looking at houses every weekend for 6 months than that dumb model.

------------------------------------------------------------------------------------------
If you can't explain it to a six year old, you don't understand it yourself.

 

[IRstuff]

Zillow has the same sort of problem with their pricing model, as do all the real estate sites. Zillow was optimistic when we were actively selling, but once the sale became pending, the price dropped by $100k.

As for the OP, I suggest reading

https://en.wikipedia.org/wiki/Long-Term_Capital_Management,

which was founded by possibly the best minds in the business at the time, including Scholes and Merton, who won a Nobel prize for their derivatives model. They even did a Nova special on the collapse.



TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[ornerynorsk]

A star for IRstuff and Hamburger. Don't play with any money that you cannot afford to lose, because you are figuratively on the verge of flushing money at any given moment, whether you are trading Forex, futures, any commodity or pseudo-commodity.

20 years ago when I was seemingly very smart and confident, I day traded futures, including currencies. I no longer do. There is a very long story behind why I no longer do, but the cost was the equivalent of a semester at a very good university. It's a bloody minefield. Oh yeah, historicals are great when the market is playing by the "rules". Only thing is, it most often never does. Patterns are a trap that will chew you up and spit you out.

It is better to have enough ideas for some of them to be wrong, than to be always right by having no ideas at all.

 

[TehMightyEngineer]

They even did a Nova special on the collapse.

I think this is the documentary you're thinking of. I remember watching it on NOVA as a teen and most of it going way over my head, but some of the ideas stuck because I still remembered it today.

http://watchdocumentaries.com/trillion-dollar-bet/

Ian Riley, PE, SE
Professional Engineer (ME, NH, MA) Structural Engineer (IL)
American Concrete Industries

https://www.facebook.com/AmericanConcrete/

 

[ornerynorsk]

One other thing that I don't think has been discussed. You may, in fact, be very proficient in higher math, neural networks, AI, etc, but when you are planning an electronic trading system such as this, what guarantee do you have as to what prioritization your trades will have? Your choice of broker and how they will prioritize your trades will have everything to do with success or failure where milliseconds and tenths-of-a-cent are your playground. There are potential electronic traffic delays at every interchange of the trading system, up to and including the exchange itself. I see this as the most critical aspect of the whole endeavor, assuming you have everything else dialed in and figured out.


Also - have you seen the warning on every financial product offered? "Past performance is not indicative of future results". Historical data is junk. It happened, it's over, it's done. It may or may not repeat because circumstances and all of the factors that went into producing a certain result in the past are never precisely and exactly duplicated, which is why I say patterns are a trap. IRstuff said it more eloquently that I am saying it - "The fact that you think past history is an analytical predictor for the future is problematic. "


It is better to have enough ideas for some of them to be wrong, than to be always right by having no ideas at all.

 

[IRstuff]

ON, thanks for that; we actually agree on some things ;-)

Patterns are often misleading for other reasons as well; the human brain has adapted to its processing load by taking shortcuts that rely heavily on pattern recognition, which is why we are fooled by optical illusions and why we see hunters in the stars. But, typically, there's confirmation bias built into pattern recognition, because it's essentially supervised learning, so we only see the patterns that correspond to our pre-assigned classes.

Unsupervised learning does have the potential to see patterns that we ignore because of the brain processes noted in the cited article. You may be smart enough to get a SOM to find hidden patterns; if so, kudos and pre-congratulations on your upcoming Nobel Prize. However, Kohonen developed SOMs decades ago, and another fallacy in thinking would be that you are the first person in the entire world to think of use SOMs or Matlab toolkits to make a killing in the financial markets. It's certainly possible that there are a bunch of shadowy figures in the financial markets that are making oodles of money that way, but odds that they are all supremely disciplined at keeping these secrets is a big stretch. And, one should note that despite have truckloads of PhDs and quants, almost all funds tend to do worse than the overall market, i.e., investing in index funds will beat a huge chunk of the quant-driven funds.

I would think that a high percentage of engineers have attempted the same, particularly because we're analytical and see streams of numbers as being highly amenable to analysis, but markets are worse than random; they're PEOPLE-driven, or more precisely, emotion-driven. To that extent, the financial markets contain the equivalent of rogue waves, which were heavily ignored as the drunken ravings of scared or lunatic mariners. However, it turns out that rogue waves are real, and a 90-ft wave popping up in the middle of a calm sea happens way more often than statistical probability says they should. The morale of all of this blathering is that rogue waves are everywhere, and they are not predictable by ordinary means.

On top of that, technical analysis, which is what you are essentially attempting, can only be done on data that you know. But, not everyone in the market is subject or has access to the same data. The missing data, or the unknown unknowns that a certain US government official re-made famous, will often be the demise of any technical analysis approach.

One should also be reminded that futures trading is essentially gambling in its almost purest form; which then should make obvious that the markets themselves are essentially gigantic, global, casinos where there are no minimums nor maximums. You can lose a dollar or lose a billion; the market doesn't care, as long as it gets its cut.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[drawoh]

Konstantinos88,

A teacher asked a little boy, "If you had seven apples and I asked for three, how many apples would you have?" The boy replied "Seven."

Are you solving the right problem?

--
JHG

 

[NRP99]

I will advise you to do fundamental analysis rather than technical analysis/day trading/forex trading/options or futures. No matter how many tools you use, there are always external factors present which will do more harm than profit or loss. Market manipulators such as big institutional investors will decide the price movement up or down in matter of minute. If you happens to in opposite position than their bet, there goes all your money.

If you have deep pockets, go ahead and take the bet.

If you instead use your mathematical analysis for fundamental analysis of company finance, you will come to know actual worth of a company. You will come to know lot of things in the process.
Here is the interesting sppech by Charles Munger-Link

 

[Terratek]

It's hard enough for a business owner to predict the success of his own company, let alone an outsider who may not know anything about the business. You can make money, but don't think it won't be mostly due to luck.

 

[HamburgerHelper]

Most of the time you will never be able to compete against someone that has being doing something for full-time for a living. A pro should be able to out do an amateur on all occasions. Kaggle, this data mining competition website where people who want models put up prize money for the best models, had this competition for forecasting wind generation based on weather data. I didn't win with my submission but I beat out several data scientist and utility forecasters in my first only only foray into data mining/modeling. That though I think is is a little different than market forecasting since the parameters are well understood and there aren't a lot of unknown or loosely understood variables. So, take that for what it is. Occasionally, an amateur will out do a pro. Wind generation forecasting is probably less competitive too than market forecasting, too. If you want to get into data mining/science, Coursera has a series of free online courses that are put together by Andrew Ng, a co-owner of Coursera and a professor at Berkley. The classes are held in high regard and the Netflix 1 million dollar movie recommender model contest a number of years back was won by someone that went through these courses.

------------------------------------------------------------------------------------------
If you can't explain it to a six year old, you don't understand it yourself.

 

[ashtree]

Why ask us.
Last time i looked this was an Engineering Tips Forum. Maybe we might know something about engineering, but realistically what would we know about Forex.

If we knew enough about foreign exchange deals to be making a fortune
1) Would share our FOREX tips on a site like this.
2) Would we be bothered mucking around on an engineering tips site at all.
3) We would tell you to go and learn it the hard way the same as everyone else does.

Regards
Ashtree
"Any water can be made potable if you filter it through enough money"