pjrowe / economic-regime-analysis-and-factor-models Goto Github PK

This repository contains my work related to the EDHEC Business School specialization Investment Management with Python and Machine Learning Specialization on Coursera (https://www.coursera.org/account/accomplishments/verify/R5NTFNMF67P4). I have extended and adapted the base of code provided in the course for my own education and investing and trading activities at Crivano Capital.

The Specialization is comprised of four courses. https://www.coursera.org/specializations/investment-management-python-machine-learning#courses

• Course 1. Introduction to Portfolio Construction and Analysis with Python
• Course 2. Advanced Portfolio Construction and Analysis with Python
• Course 3. Python and Machine Learning for Asset Management
• Course 4. Python and Machine-Learning for Asset Management with Alternative Data Sets

Course 2 was too high level and general for a one person investor/trader. The theory is quite interesting, though, and I realized after this course that the practice of portfolio management has been largely automated (think ETFs) and things like portfolio rebalancing could be accomplished with the click of a button, after all the theory has been implemented. Thus, Course 2 did not have much for me to try to use, unless I intended to work for a large asset management firm as part of a team. That was my impression, at least.

Course 4 had some interesting applications of machine learning to investment process, but there were some applications that would be best described as 'smoke and mirrors', according to one headhunter. In the former category, analyzing newly available sources of data (cell phone geolocation data, transactional data from web sites, etc.) could yield useful insights of retail activity during a Christmas shopping season. In the less interesting category would be analyzing twitter feeds or 10-K reports for mentions of a certain company / competitor to analyze sentiment about a stock or company. This application seems very susceptible to change over time or manipulation. To create buzz around a stock, robots could be built to automatically comment about the stock on twitter, for instance. I can't imagine a portfolio manager could have much conviction repeatably over such a soft analysis.

That leaves Course 3, which was split into 5 modules / weeks. This Course did have some promise for application of a small trader / investor.

• Week / Module 1. Introducing the fundamentals of machine learning (background, nothing included in repo)
• Week / Module 2. Machine learning techniques for robust estimation of factor models (Module 2 folder in repo)
• Week / Module 3. Machine learning techniques for efficient portfolio diversification
• Week / Module 4. Machine learning techniques for regime analysis (Module 4 folder in repo)
• Week / Module 5. Identifying recessions, crash regimes and feature selection (Module 5 folder)

Week 2 and Week 3 were again, in my estimation, a bit too high level for me. Rebalancing a large amount of money across asset classes or individual stocks could be guided by such techniques, but this occurs at large funds / companies.

The topics of Week 4 were as follows: Portfolio Decisions with Time-Varying Market Conditions, trend filtering, a scenario-based portfolio model, a two-regime portfolio example, and a multi-regime model for a University Endowment. Considering the all-time highs in market multiples / valuations and hitorically low interest rates, I focused on the two-regime portfolio models. My intention was to use ML models to identify / predict a recession using publicly available economic data. Then, I could take a short position in the SP500 emini futures (very liquid, tax advantageous, low commissions) as my 'portfolio' in anticipation of a major stock correction / recession.

Well, Covid hit, and a sudden recession without precedent ensued. So much for model building. And then an unprecedented coordinated stimulus by Central Banks worldwide, leading to an all-time high in market multiples, a doubling in the Nasdaq futures from a low of 7000 in March to 14000 at its recent highs (May 2021). The doubling occurred even though the recession is still ongoing. Again, so much for building a model. The crash from February 2020 to March 2020 all happened within 4-6 weeks top to bottom, and the rally has hardly abated since the low. So, how useful can monthly economic indicators be?

Anyway, I've updated the course materials as a refresher, but as a practical method for a small fish to position in the current market environment, with bubbly prices and speculations on cryptos and big tech stocks, no, I don't think it is that useful. So many short sellers funds and value portfolio managers have been burned, closed down, or retired in this market mania, and they continue to warn about the inflation and crazy prices. The names include Buffett, Druckenmiller, Michael Burry, Chanos, Einhorn, and Fleckenstein, just to name a few.