Bayesian regression - Has anyone made an EA using this algorithm? - page 47
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
26000 hrn
Job description
PROFESSIONAL REQUIREMENTS (we value the desire to learn on the go the most):
Neural Networks: unsupervised learning: FNN,RNN (Recurrent Neural Networks, including LSTM), RBF, etc.
Programming experience (or ready to learn) in Python (and set of libraries for all things statistical)
Data analysis libraries in Python (theano, Torch, Pandas, NumPy, scikit-learn) or their equivalents in R
some experience with Machine Learning, collaborative filtering, cluster analysis, Graphs Theory
other blended approaches: ANFIS (adaptive network-based fuzzy inference system)
advanced knowledge of statistics and time series: Stochastic processesTools: SSA/SVD, RSSA, FIMA/ARFIMA, Nonlinear Autoregressive Exogenous Model (NARX), (N)GARCH and its derivatives, Hurst Exponent and its applications, Recurrence quantification analysis (RQA)
TASKS (prioritized order):
statistical analysis of financial data, econometric applications
building services and frameworks for interactive distributed query processing over large volumes of data
28000 UAH
Job description
PROFESSIONAL REQUIREMENTS (we value the desire to learn on the go the most):
Expertise in Natural Language Processing (NLP) technology
NLP implementations in Python (NLTK) or other developed implementations
Neural Networks: unsupervised learning: RNN (Recurrent Neural Networks), FNN, RBF, etc.
Programming experience (or ready to learn) in Python (and set of libraries for all things statistical)
Data analysis libraries in Python (theano, Torch, Pandas, NumPy, scikit-learn) or their equivalents in R
some experience with Machine Learning, collaborative filtering, cluster analysis, Graphs Theory
other blended approaches: ANFIS (adaptive network-based fuzzy inference system)
25000 UAH
Job Description
PROFESSIONAL REQUIREMENTS:
advanced knowledge of statistics and time series: representing time series with reduced dimensionality, such as Discrete Fourier Transformation, Single Value Decomposition (SVD), Discrete Cosine Transformation, Discrete Wavelet Transformation, Adaptive Piecewise Constant Approximation, Chebyshev polynomials, Symbolic Aggregate approximation, Indexable Piecewise Linear Approximation, etc.
experience in measuring similarity of time series: Euclidean distance (ED), Dynamic Time Warping (DTW), Longest Common Subsequence (LCSS), Edit Distance with Real Penalty (ERP), Edit Distance on Real sequence (EDR), DISSIM, Sequence Weighted Alignment model (Swale), Spatial Assembling Distance (SpADe) and similarity search based on Threshold Queries (TQuEST)
Serious knowledge of Stochastic processes Tools: RSSA, FIMA/ARFIMA, Nonlinear Autoregressive Exogenous Model (NARX), (N)GARCH and its derivatives, Hurst Exponent and its applications, Recurrence quantification analysis (RQA)
Programming experience (or ready to learn) in Python (and set of libraries for all things statistical) data analysis libraries in Python (theano, Torch, Pandas, NumPy, scikit-learn) or their equivalents in R, Octave for quick prototyping
some experience with Machine Learning, collaborative filtering, Graphs Theory
These methods are ACTUALLY used by market makers in their trading. And there are Bayesses somewhere out there too, of course. But not in a nakedly brazen way - for price forecasting, moreover, on spot forex !
Oops, forgot to add - I don't use any of that list above. That's probably why it works for me.....
I witness on a daily basis how someone using an obsolete MT4 with the corresponding primitive algo-TS generates a turnover of >$5 billion per day on the tailrace, making a daily profit of plus/minus a few hundred USD. The mathematical expectation is positive. Over a million USD a month for sure.
Where are your ***funds that can't replicate this kindergarten, lasting (judging by history) almost a year on real (even more in the tester)?
Stupid primitive strategy gives you more "quons" per month than your "quons" for years
Many times I've told you here that quons, algotraders, market makers - they're not idiots, that they have a GOOD understanding of mathematics, that they don't get paid 100K+ GEL per year + bonus, but you all don't seem to get it.
Where are they? Scoured all the forums and monitors. I have scoured all the forums and monitors, and no one trades this profitable pattern. And the history shows everything very well. No one cares. I try it myself - it comes out weak. But the fact is that triviality brings more profit than ***-funds.
I witness on a daily basis how someone using an obsolete MT4 with the corresponding primitive algo-TS generates a turnover of >$5 billion per day on the tailrace, making a daily profit of plus/minus a few hundred USD. The mathematical expectation is positive. Over a million USD a month for sure.
Where are your ***funds that can't replicate this kindergarten, lasting (judging by history) almost a year on real (even more in the tester)?
Stupid primitive strategy gives a month more than your "quants" for years
Where are they? Scoured all the forums and monitors. I don't see this profitable pattern. And the history shows everything perfectly. No one cares. I try it myself - it comes out weak. But the fact is that triviality brings more than ***-funds.
Check out the book "quanta as wizards". In it, all the mathematicians use quantitative methods, not some wizened formulas. There are doctors and professors in there.
So why don't they use complicated mathematical constructs? Because they probably don't work. The market is not linear.
In my opinion using higher algebra in trading is like putting a square in a round. Mathematics is certainly needed. For example using linear regression is obviously a good example of using maths in trading.
Claiming to make money from trading... No, it isn't. No mathematics is required to make money from trading, not at all.
Everything I've earned has been thanks to maths so dumb it's hard to believe. As soon as I do something complicated I immediately lose.
But it's not even my own example. It's about what I was able to observe from the outside. It is the primitive that makes a huge profit. On the tailrace, at least.
Bitcoin could once be traded using a simple EMA-crossover strategy. It worked until the bankruptcy of one of the major exchanges mtgox, sometime before spring 2014, then it broke down. And I'm sure forex could be traded using the same EMA-crossover strategy decades ago. So yes, primitive algorithms work very well sometimes. My personal view is that forex pairs charts are almost random, very much noisy, but at the same time they obey some simple laws. The problem is that forex charts as if trying to get rid of the patterns found over time, the more people trade the same strategy, the faster it will stop working. If you find a simple pattern that nobody has found yet, you can make money.
Bitcoin could once be traded using a simple EMA-crossover strategy. This worked until the bankruptcy of one of the major exchanges mtgox, sometime before spring 2014, then it broke down. And I'm sure forex could be traded using the same EMA-crossover strategy decades ago. So yes, primitive algorithms work very well sometimes. My personal view is that forex pairs charts are almost random, very much noisy, but at the same time they obey some simple laws. The problem is that forex charts as if trying to get rid of the patterns found over time, the more people trade the same strategy, the faster it will stop working. If you find a simple pattern that nobody has found yet, you can make money.
This thread has become littered with haters. This does not apply to you.
I hope it will be possible to discuss something constructively.
Responding to your post, I can agree and have been digging the topic of long term and recurring patterns myself for a while now.
Take a look at this (it's PR of course, but not the product but the raw idea): https://c.mql5.com/1/37/teaser2.JPG
What is it? It's the result of machine learning validation on predicting the sign of forex price increments (regular spot forex from the terminal). Inside the whisker box lie 49 samples, each with a few thousand observations. Specifically, the dots show the level of accuracy of the binary classification for each of the validation samples.
Why are there so many samples? They come from the same time interval of about 5 years for 5 currency pairs taken together, that is, 25 years in total. But in each sample the observations are taken in randomly large increments to make them mutually independent. And each of the samples independently of the others covers time periods within the validation period.
The short conclusion from this graph is that I get a stable 55% accuracy of sign recognition for 30+- minute forecast horizon. And it can be statistically shown that this result is not random. So the machine catches long-term dependencies in 5-major data at 10 * 5 = 50 years of training and passes the validation test. All this, again, for 5 currency pairs taken at the same time and the machine does not distinguish between them.
Another pruf: https://c.mql5.com/1/36/charts-5-9.JPG
This is already an incremental regression for the same data. Same 49 validation samples and specified metric = 1 - (MAE for predictions / MAE for sample mean). That is, by how much we reduce the absolute error compared to the sample mean (it is very close to zero). The statistics is powerful enough to say that the result is not random.
Again the machine learns universal and stable patterns.
Blog on this experiment: https://www.mql5.com/ru/blogs/post/661499
But I haven't managed to get a positive mathematical expectation in points. And I already doubt that I will be able to. Perhaps I'll try to make models for each pair separately. If such predictions may somehow be applied to trading - that's a big question for me.
Your opinion will be useful.
This thread has become littered with haters. This does not apply to you.