Fast Artificial Neural Networks

The so-called Fast Artificial Neural Networks are an extension of the famous original implementation created by Stefan Nissen in 2003 where his autonomous agent had the ability to learn and store information from experience.

The best way we can describe how our trading FANNs work is by using the analogy of playing the board game called *Monopoly* by Waddingtons the toymaker.

Open the box and you will see a set of rules written in black & white, a dice and board and some bits & pieces that make up the *game*. Let’s call this the *Input Layer*. At the end, somebody wins and gets a reward. Let’s call this the *Output Layer*. In between are the *Hidden Layers*. If you played the game over and over you begin to notice things and store that information in your memory bank to give you an *edge*. For example, a set of two dice has the highest probability of throwing a total of seven (6+1, 1+6, 3+4, 4+3, 2+5, 5+2). Or, owning the Purple & Orange properties have a better landing rate than say *The Greens*. And so on.

If your memory bank or *Library* is Bayesian in its thinking and has the ability to learn and you will find that your FANN can consistently win (if your strategy is valid).

In the trading environment we can start building our *Input* layers quite easily. For example, volume of trades or volatility can vary according to the time of day or trading session and indeed the day of the week and month of the year. We also know that trading is flat when awaiting a news announcement. We also know that markets tend to move violently up or down then consolidate. These movement structures & patterns are quite distinctive, recognisable and often repeat themselves.

We have access to a collection of first-rate programmers from all the countries of the world —- they can appear out of *nowhere* —- in the Columbian heartlands, the veld of the Ukraine, Dhaka, London or Ramsberg, Sweden. They all have one thing in common —- the ability to recognise and code a *pattern* —- as simple as that.