Neural Network is a family of models that is used to estimate random variables that depend on a large number of inputs that are mostly unknown. What has attracted the most interest in neural networks is the possibility of learning. Given a specific task to solve, and a class of functions F, learning means using a set of observations to find F which solves the task in some optimal sense. This video explains in simple terms what are neural network.
These series of videos try to demystify neural nets for a layman.
This is the second part of the above video in which you learn how to implement neural net using Python language.
In the last video neural net was used to make predictions but those predictions were not good. In this video you learn how you can train the network to learn from its past predictions and improve them further.
So this is how neural nets work. Sounds complicated? Ofcourse the theory of neural nets is complicated. We are interested in seeing if we can develop neural nets for pairs like EURUSD, GBPUSD, USDJPY, NZDUSD, AUDUSD etc. If we can develop a model that can predict the range of the next daily candle or the next four hourly candle with 80% accuracy, we can use that prediction and combine it with our technical analysis skills to improve the accuracy even further to above 90%. The most important thing is to somehow predict the daily range of the candle. If you know this you can time your entry and exit with more precision.
You can read this interview in which this champion trader explains how he used the neural nets in his trading systems. Of course there is no holy grail. Neural nets can make wrong predictions as well. What we are interested is in a model that can make predictions consistently with an accuracy of above 80%. As said above then we combine that prediction with our skill in reading candlesticks and make the right entry and exit decisions.
This is another simple example that explains how to model neural networks with R with the perspective of a Quant. Quants don’t believe in technical analysis like us. They believe in their statistical models that they can implement using powerful computers to search and identify hidden patterns in the market that they can use in their automated trading systems.
This paper reports empirical evidence that a neural networks model is applicable to the statistically reliable prediction of foreign exchange rates. This paper claims that it’s neural network model was able to achieve a predictive accuracy of above 65% on pairs like EURUSD, GBPUSD, USDCHF and USDJPY. You can download this 23 page PDF and go through it. Now there are some forex neural network software available in the market that are claiming a predictive accuracy of above 80%. However we haven’t test them.
In the next series of posts, we will try to develop our own neural network models for EURUSD, GBPUSD, NZDUSD, AUDUSD and XAUUSD using excel and R. so stay tuned.