Like many of my fellow traders I have a dream of creating a mechanical strategy which makes profitable trades for me.
I have attempted a number of semi-automated systems in my time, but my efforts have been characterized by sporadic bursts of trial & error and a lot of half-finished attempts that never made it past the first testing rounds.
My hope with keeping this journal is to keep momentum in an attempt to explore the areas I believe are most likely to present a winning solution - there will be failed attempts along the way, but having a journal to document the process will hopefully help in keeping the focus.
The areas I am currently exploring are:
Data Mining
By analysing financial data-series I hope to extract repeatable patterns or find statistical characteristics of the data-series that can be used in a profitable strategy
Clustering
Clustering is finding a partition of the data such that data in each cluster are more similar to each-other than they are to any data in a different cluster - this is probably tge most important problem of unsupervised learning.
Neural Networks based on Adaptive Resonance Theory
Artificial Neural Networks are good at non-linear approximations and with their great adaptation to imprecision and noisy data lends themselves well to be employed in a mechanical trading system. Using ANNs are more than simply define a few layers of neurons and run training. Careful selection of inputs for the network is one of the major challenges.
Minority Games
Is a model of the collective behavior of competing agents - so the idea is to make a model of the market with a set of different agents emulating the participants and follow who comes out on top, hopefully leading to an adaptive mechanical strategy.
All four areas are supporting each other in the attempts to find a mix that suits a winning strategy.
The tools i primarily will be using is:
Sierra Charts
R
C++ (I am using QTCreator and Visual Studio as IDEs)