A Trading System

From biology we know the fundamental principle that only the simplest systems work (biological systems are wastly complex but they are /composed of layers upon layers of simple sub-systems, organized hierarchically).

In the context of trading complex systems will perform not much better than a random entries and exits, while consistent survival could be exibited only by simple systems with a few well-choosen consistent rules.

A system which just consistently sets stop-loses to zero and tighten stop-loses as the market moves in a trend will outperform most of amateur human traders. Well, it also should not exit positions prematurely.

Most of well-known trading systems, such as ones described by Elder, are explicitly designed for humans. They are not suitable for machines.

These are just simplest but consistent mathematical models which use most down-to-earth universal notions, such as distance (range), speed and momentum (as generalizations, of course).

Basic frequency-based statistics can be used to describe the current state (and never for inferring future!) or to “see” (literally) what is going on. Ranges and averages only.

May be some Kalman-filters, which are, basically, constantly adjusted (by “experience”) activation thresholds.

Memes aside, supervised learning cannot be used in the Markets. Training of a neural net requires, in principle, a stable environment.

Unsupervised learning on an unstable environment is modern day’s astrology.