## Sort by: Recently added articles Most discussed Top rated Article Library

8/38
Ranking
Introduction:

It’s often suggested that forex and markets in general are a succession of random movements, a random walk. People trade based on logical reasoning but there are so many people affecting the same value that the movements of a price become impossible to tell apart from a random walk.
With a simple method of random walk visualization, we can easily show that the above theory is incorrect. Especially for short periods like ticks, prices tend to travel in easily identifyable trends, much more so than a random instrument would.
Random walk vs tick data

First let’s see what pure randomness, as far as computers can generate it, looks like:
The pictures above are generated by a random walk. On each step, the position has a 50/50 chance of going up or down and on the next step, a 50/50 chance of going right or left. The colors change accordingly purely for aesthetics.
Now we’dlike to do the same using actual tick movements from forex instruments. We generate the following images using forex tick data. We move up or right for rises in price and we move down or left for falls in price. We alternate horizontal and vertical movements. If forex is simply a random walk we expect t…
Translate to English Show original
olga 14 May

If I understand your comment correctly, I think you must have missunderstood salamandra. The position on the image moves horizontally and vertically one after the other, so if a price movement is random, the final drawing will have no global direction. The truly random images at the top could also be translated to a chart that goes only forwards in time. I just find this method of visualization much more telling.

ioannist 21 May

"There is a strong tendency towards diagonal movements" ...well, of course, since you have coupled the up movement with the right movement, and the down with the left. You did not do that with the random walk data.

ckdot 16 Dec.

"If you simply predict that the next tick movement will be the same as the previous, you’ll be right about 65% of the time". I've double checked that via a simple script for about 5 million ticks from January 2011 till November 2017 and the chance to guess the right movement is nearly exactly 50% - not even close to 65%. I just checked for all data if tick 3 is bigger than tick 2, if tick 2 is bigger than tick 1 - or tick 3 is small than tick 2, if tick 2 is smaller than tick 1. Maybe I'm mistunderstanding something right now, could you explain this a bit further please?

elfen 10 Feb.

yeah,i found the same with you ,you can filter the tick date with distance pipes.but the probability of trend will affected by the distance.

mc2Finex 24 June

As ckdot point out, the probabilities are very close to 50%. If they really were 70% you had found a "Golden Nugget".
Think about this. If any part in the market could predict it, they would not trade. Both must have same uncertainty to be willing to trade. What provides even uncertainty.........Randomness with 50% /50% probability. Any other distribution will collapse the market as one part would deny trading.

14/29
Ranking
Short Term Long Term memory, yet another machine-learning approach to trading

This approach is directed towards high frequency trading and high frequency trading is very hard because you need to beat the commission and the spread to make a profit, of course this is always true but even more so the smaller the possible profits
are.
The Long Short Term Memory network, or LSTM, for short, is the machine-learning answer to time-based systems like forex. It’s basically a recursive neural network, it keeps old information in memory, but it is very sophisticated and works much better than other types of networks on specific time based problems.
It’s beenproven to possess all these features which are all extremely interesting for trading:
1. Recognition of temporally extended patterns in noisy input sequences
2. Recognition of the temporal orderof widely separated events in noisy input streams
3. Extraction of information conveyed by the temporal distance between events
4. Stable generation of precisely timed rhythms, smooth and non-smooth periodic trajectories
5. Robust storage of high-precisionreal numbers across extended time intervals
It is bei
Translate to English Show original
Skif 11 Jan.

Algorithmic trading - always wanted to look into this. Your article might make me do it!

olga 11 Jan.

hey I got a small result, the LSTM, after a few days of data training, can predict with 60% accuracy whether the next tick will be higher or lower than the previous! I think that's pretty neat..

skynet 21 Jan.

Hi olga, thanks for the article. Do you know any libraries that already implemented LSTM ?

olga 21 Jan.

sure, https://github.com/evolvingstuff/LongShortTermMemory. This one's very straight forwards to use, simple but limited in terms of the architecture it implements.
the one here: https://bitbucket.org/dmonner/xlbp is more complex to use but it has more possibility for complex architecture and there's more support

Victor 29 Jan.

very good article

2/29
Ranking
Machine learning can help us optimize automatic trading strategies. By studying the huge amount of past information, we can identify patterns that help us predict the evolution of the market to a sufficient extent. This is of course what some traders have been doing for a long time but the automatization of the process allows us to find much better strategies and much faster than it would take a human. Here we propose a speculative strategy that has been successfully tested and demonstrates the possibilities brought by machine-learning in forex.
Automatically finding a winning speculative strategy on eurusd

EUR/USD is a very lucrative pair for a speculative strategy built from machine-Learning algorithms, although our method is able to find winning strategies on other instruments and some that work across several instruments, the strategies developed for EUR/USD give the best returns. This is how the strategies are built.
We cannot feed the actual price to the algorithm because we want it to recognize patterns independently of their height on a chart. We therefore feed it price movements, from high to high and low to low (better than open to close). This is a simple kind of indica…
Translate to English Show original
Likerty 8 Jan.

Interesting article, really. Just a bit of a problem lies in the price feed of the past - the exact levels that are in play - rarely get reaction "to the pip" - more often they get overshoot as the real accumulation is happening just beyond the level (10/40 pips), not exactly at it. So, by studying past price is difficult to find things.. I would rather shoose to look for levels in advance and by watching price reaction to the spot in real time, I can say - is it a correct level or not..

Skif 11 Jan.

automatic trading - its benefits are well documented!!!

olga 11 Jan.

cheers, I should have said something about the benefits yeah. for me, the major advantage is that these machines don't have emotions like us, but humans manage and test them and they still have emotions.

Airmike 12 Jan.

nice article olga. looking forward to next article about machine learning

nick21 31 Jan.

Well done Olga

23/40
Ranking
IntroductionUpon request, I will explain in a little more detail how an Artificial Neural Network (ANN) works. I then give an example of a simple network and continue to the actual technique I am implementing called a convolutional net. This article is rather technical and thus may not appeal to many people. The perceptron The perceptron is the basic unit of the neural network. It receives a signal (a number), passes it through an activation function and outputs the result. There are different activation functions depending on the type of ANN but the one most commonly used is the sigmoid function or hyperbolic tangent.An input on the x-axis returns the corresponding output on the y-axis, always between 0 and 1 or alternatively between -1 and 1.Network and weights The individual perceptrons are connected with weights that multiply the incoming signals.The first two perceptrons receive an input and have a corresponding output, these outputs are multiplied by the relevant weights and then added to each other, this is the input of the third neuron. At least three layers of neurons are necessary and sufficient to model any system given enough individual perceptrons in the middle layer, …
Translate to English Show original

I completely disagree with citikot. I think trading is all about patterns. Big banks are really big on pattern recognition its really important aspect of trading and neural networks can be a vital component in recognising new patterns. Like this article.

drishti 12 Mar.

Once again, you explained very good about ANN network, how it works? Actually I was also thinking to writing about ANN, but unfortunately did not get time. Good work. +1

olga 12 Mar.

I understand what citikot is saying and I'm afraid HFT will be too expensive for me. On the other hand I seem to have found profitable rules on the weekly time-scale too, strangely, ANNs are not able to model those easily but decision trees are.
What I hope to do with this is join the two time-scales into a strategy that is addapted to my trader status.

citikot 13 Mar.

@DannyTrader You are completely right about the fact that big banks and market makers are really big in pattern recognition. Because they perfectly know that herd of retail traders make their trading decisions based on patterns. And they create patterns to catch retail traders in trap and unload own positions to stupid retailers. That's why I stated that You can apply statistics to HFT to catch couple of pips in market noise. But when You are trying to make more You have to consider other matters - from skill to read what market makers do now to macroeconomics factors.

LSD 13 Mar.

@citikot ... and they are big not only in pattern reco... ;)