Oakville Chapter of The Canadian Society of Technical Analysts
Oakville Chapter of
The Canadian Society of Technical Analysts
Durham Chapter of The Canadian Society of Technical Analysts
AmiBroker Technical Platform Meeting
October 2020.
Presenter: Dr. Howard Bandy of Blue Owl Press
Topic: “Transition to Machine Learning for Trading”
From Dr. Bandy...
"It [the presentation] begins with a trading system written in AmiBroker that I published in "Quantitative Technical Analysis". I rerun the backtest through current data, showing that it continues to be profitable. I use that as the vehicle for my discussion of risk and profit potential. I spend some time going over fundamentals of systems analysis, workflow, data transformations, model development, model validation and how these tasks are different for machine learning than AmiBroker.
I discuss the differences in model development between AmiBroker and Machine Learning. And demonstrate a roughly equivalent system written in Python. Since AmiBroker uses a single decision tree as its model, I begin with a demonstration of that in Python. Then expand to show several alternative models and the results.
The later part demonstrates ensembling (the technique I suggested at the last meeting), including random forest, bagging, boosting, and stacking. I show results using scikit-learn's random forest, then go on to XGBoost -- one of the overall best models available today. I also very briefly demonstrate deep learning using Keras and TensorFlow.
Here is a list, in order of presentation, of some of the key points I will be discussing:
... Transition from traditional development platform to machine learning
...... What you have
...... What you get
...... What it costs
...... How to get started
... A change in perspective
... Indicators largely disappear
... Backtester largely disappears
... Use state signals and mark-to-market daily
...... Some diagrams
... The target is completely up to you
... Get used to transformations
... Data preparation
... The metric by which systems are compared
... The four faces of risk
... AmiBroker example from page 274 of QTA
... Switch from AmiBroker to Python
... Python references, installation, books, tutorials, etc
... A side trip to describe safe-f and CAR25
... The relative importance of winning trades and losing trades
...... Rule One! Avoid losing trades
...... Demonstration of Rule One
... Backtester for Python Models (It is about 4 lines of code)
... Mantra for success
... Metrics
... Workflow
... Several variations of the same trading system using different models -- all in Python
... Demonstrate optimize, walk forward, and validate all in a single run.
... Ensembles
...... Random forest
...... XGBoost
... Deep learning
...... Keras / TensorFlow
... Trading management (very briefly -- worth another presentation, perhaps)
... Summary and Questions"
About :
Dr. Bandy's education includes degrees in mathematics, physics, engineering, and computer science. His graduate studies and research were in the areas of modeling and simulation, statistics, and artificial intelligence. His graduate advisors were among the pioneers in the computer industry. In artificial intelligence, he did some of the early work in nearest neighbor analysis.
Dr. Bandy is retired from a position as a tenured full professor of computer science and mathematics, and a university dean.
His professional career included:
- University teaching and research in computer science and mathematics.
- Management of commercial and industrial computer divisions (what would today be called chief information officer).
- Applications of modeling and simulation to financial trading systems.
- Design and programming for a system to select individual stocks to trade, and to generate buy and sell signals for them.
- Senior research analyst for a Commodity Trading Advisor (CTA), where he held a Series 3 license.
- Invited speaker at local, national, and international conferences.
- Author of six books related to trading system development and management.
* You can find out more about Dr. Bandy's work by visiting his website at http://blueowlpress.com/
** Questions and feedback about this presentation may be directed to our forum post here.
Hope to "see" you all online on Oct 6th!
To Your Continued AmiBroker Learning!,
Dave Di Marcantonio
AmiBroker Canada User Group
www.amibroker-canada.org
&
www.AmiBroker-Courses.com
Not a member? Join here.
