Forex Neural Networks 1

Arrgghhh long time no write my blog log... today i want to write somethin i have to learn.
Yes.. its all about my forex still.. on my boree of my life, but i want to did something usefull for me and you, readers of course.. :D

Now i have think and i want to learn about Neural Networks Spot, Forex Forecasting... for now on, I dont know WTF this Syatem. I grab this articles from MQ4 dot com. it will separated on 4 Post by meh. latter.. edit this... blablabla


Introduction

During the last several years we observe the explosion of interest towards neural networks, successfully used in different spheres - business, medicine, technology, geology, physics. Neural networks are widely used in spheres that require forecasting, classification and management. Such an impressive success is determined by several reasons:

* Wide possibilities. Neural networks are a very powerful modelling tool, allowing the reproduction of immensely complicated relations. In particular, neural networks are nonlinear by nature. Over a period of many years linear modelling was the main method of modelling in the majority of spheres, because optimization procedures for it are well developed. In tasks, where linear approximation is not enough, linear models operate poorly. Besides, neural networks overcome "the curse of dimensionality", which does not allow modelling linear relations in case of a large number of variables.
* Easy usage. Neural networks learn by examples. The user of a neural network collates representative data and then starts the algorithm of training, which automatically accepts the data structure. Of course the user should have a set of heuristic knowledge about the way of selecting and preparing data, of choosing the appropriate network architecture and interpreting results. However, the knowledge level, needed for a successful use of neural networks, is much lower, than that needed in traditional statistics methods.

Neural networks are attractive from the point of view intuition, because they are based on the primitive biological model of nervous systems. In future, the development of such neuro-biological models can lead to the creation of really intelligent computers. [1]

Forecasting financial time series is a primary member of any investment activity. The whole idea of investment - investing money now with the purpose of getting profit in future - is based on the idea of forecasting future. Accordingly, forecasting financial time series lies at the root of the investment industry - all stock exchanges and over-the-counter (OTC) markets.

It is known, that 99% of all trades are speculative, i.e. are aimed not at a real trade turnover, but at taking profit using the scheme "buy cheap - sell dear". They are all based on the forecasts of price movements by a trade participants. What is important - forecasts of the trade participants are contrary to each other. So the amount of speculative operations characterizes the difference in the forecasts of the market participants, i. e. actually - the unpredictability of financial time series.

This most important feature of market time series underlie the theory of the "efficient" market, described in the thesis of L. Bachelier in 1900. According to this thesis, an investor can rely only on the average market profitability, assessed by indexes, like Dow Jones or S&P500 for New-York Exchange. Any speculative profit is of random nature and is like gambling (there is something attractive in it, isn't there?). The reason for the unpredictable nature of market curves is the same, as why money rarely lies on the ground in public places: too many people wishing to take it.

Naturally, the theory of an efficient market is not supported by the market participants (which are looking for the money lying about). Many of them think that, despite the seeming stochasticity, all time series are full of hidden regularities, i.e. are predictable, at least partially. The founder of a wave analysis R. Elliot tried to find such hidden empirical regularities in his articles in the 30s.

In the 80s this point of view was unexpectedly supported in the newly appeared theory of dynamic chaos. This theory is based on the contraposition of the chaotic state and stochasticity (randomness). Chaotic series only seem random, but as a deterministic dynamic process they allow short-term forecasting. The sphere of probable forecasting is restricted in time by the horizon of forecasting, but this can be enough for getting real profit from forecasting (Chorafas, 1994). And those, who use the best mathematics methods of extracting regularities from noisy chaotic series, can expect large profits - at the expense of less equipped fellows.

The last decade was characterised by a persistent growth of the popularity of technical analysis - a set of empirical rules, based on different indicators of the market behaviour. The technical analysis focuses on the individual behaviour of this financial instrument, apart from other securities. But technical analysis is very subjective and works inefficiently on the right edge of a chart - exactly where we need the forecast of a price direction. That is why more popularity is gained by the neuro-network analysis, because, as opposed to the technical one, it does not set any restrictions on the type of the entry information. This may be indicators of the given indicator series, as well as the information about the behaviour of other market instruments. Not in vain neural networks are widely used by institutional investors (for example large pension capital funds), working with large portfolios, placing great importance on the correlation between different markets.

Pure neuro-network modelling is based only on data, not using any antecedent arguments. This is its strong and week point at the same time. The available data may be insufficient for training, dimensionality of potential entries may be too large.

That is why for a good forecast one should use neuropackages with large functionality.


... read More !..