Technical Architecture

 

Process streams

User Interface to pick stock portfolio

  • Drupal - web component fronted by Drupal that allows the picking of stocks, display of charts, and status updates from the model building, Daily allocation and King of the hill testing

  • The interface pulls data from the R/C++ backed

  • This is what the users interact with, and the rest of the streams are not bound by it. All of the other streams and Dependant upon each other, but can work completely without the web interface

  • Right now it’s terrible and not completed

http://unicorninvesting.us/

Model Building

  • Automated machine learning model building

  • Utilizing the GA Package for feature selection and search from all of the data sets

    • GA process has generated 5000 Neural net models for one portfolio in less than 3 days with the overal population performance improving by 20% over that time

    • performance gains are expected to slow, but it will provide the system market reaction flexability so that the portfolios model isn't locked into one strategy.

    • Utilizing the FCNN4R package for building and pruning the Neural networks

      • Very Fast.. Builds a 200/400/15 Neural Net, combines all the data and evaluates in less than 5 minutes

Back testing and Model/strategy evaluation

  • Currently exploring the Psychoanalytical and PortfolioOptimization packages for cross validation, back testing comparrison and to gain ideas for additional levers to add the model building sections.

Data gathering and cleaning

  • quantmod and quandl

  • Custom cleaning and normalization

Forecasting

  • forecast and PSF packages provide the forecast as additional data sets for inclusion in the data sets

  • Structuring output to be useful for the NN is proprietary

Daily allocation creation

  • The current Primary model for each portfolio is utilized to determine the portfolios percentage allocation for tomorrow

  • Current Primary model is back tested with current data and it’s current score is calculated to account for changing market conditions (It was the best model when it was created, but is it the best model now?)

Dependency Tree

Each of these items is Dependant on everything above it, everything below it is unnecessary for the process to run. Most of it runs in parrallel and so is very scalable

  • Data Gathering, Cleaning, and Normalization

    • Forecasting

    • Daily allocation generation Utilizing the Primary Model (once it’s in place)

    • Neural Net Model Building

      • Neural net testing, tuning

      • Genetic Algorithm Model Building.   The structure allows for the expansion of the genome as new datasets are available without the loss of evolutionary gains.

        • Feature selection

        • Primary Model selection

  • User Interface ---It’s a web system so all parallel

    • Portfolio asset selection

    • Display of model state and progress

Add new comment

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.
CAPTCHA
This question is for testing whether or not you are a human visitor and to prevent automated spam submissions.