Using Machine Learning to Invest – Podcast Review
This is a review of an episode of the Invest Like the Best podcast. In this episode, the guest, Michael Reece, Chief Data Scientist at Neuberger Berman talks about using Machine Learning to invest in companies profitably to generate alpha (excess returns). The conversation touches on a lot of fascinating topics such as building an automated investing model based on Machine Learning, constructing kick ass passive strategies that track slightly above the index as well as tips for budding data scientists in the investing field.
Podcast Name: Invest Like the Best
Episode Number: 91
Episode Name: Tim Cook’s Dashboard, with Michael Reece
The crux of the discussion is to merge Fundamental Analysis techniques with quant tools like Machine Learning algorithms to plug in key parameters and enrich the model. He argues there’s still a lot of scope to exploit widely available data such as credit card transactions to gain an informational advantage and alpha. But for this to come to fruition, analysts have to evolve from Excel models to using tools like Python to work on large datasets and extract insights hidden in the data.
The key techniques discussed are detailed below.
The Basics on Using Data and Machine Learning to Invest:
– He calls data generated from transactions Digital Residue. He uses this type of data, generated mainly from digital marketing and transactions to infer who’s winning in the marketplace.
– An example includes using weather, location and credit card data together to map consumer behavior and value companies accordingly.
– The ultimate result of collecting and analyzing this data may be to put it into quant models to build a automatic valuation model for the stock market – kind of similar to what Zillow is doing in real estate.
Striking the Right Balance between Fundamental Analysis and Quant Strategies:
– According to Reece, quants are an inch deep and mile wide, while fundamental analysts are a mile deep and inch wide. This means while quants invest in many stocks with shallow fundamental knowledge, fundamental analysts invest in a few stocks that they’ve deeply studied.
– The Goldilocks Zone of current analysts is to merge fundamental and quant investment strategies. One way to do this is to build fundamental models and plug in key parameters using Machine Learning algorithms.
Key Characteristics of Quant Analysis:
– Interestingly, even though data like credit card transactions have been available for a long time, they’ve not been used fully yet. So informational advantage can still be gained. There are a lot of such widely available datasets available that analysts are still not using. Reece advises analysts to look closer into these.
– In terms of predictions using Machine Learning (ML), Its much easier to predict downsides (failures) than the upside (successes). For instance, its easy to map out low scoring essays, while its hard to do the same for high scoring ones because there’s such a wide variety of factors.
– According to Reece, Machine Learning works much better when data is relatively stationary (like company performance) vs non stationary (in the case of market prices).
The Basics of Machine Learning:
– Machine Learning is basically forming a hypothesis, gathering, analyzing and training data, using backpropagation to adjust factor weights, and constantly iterating to improve the model.
– He stresses on the importance of relying more on vectors (numbers with direction), instead of scalars (just numbers) for better financial analysis.
– BONUS: Here’s an awesome explainer on Machine Learning from Google.
His Advice to Budding Data Scientists in Finance:
– LEARN CODING.
– If you can build those complex models in Excel, you can also learn to code.
– He stresses the importance of Python using Jupyter Notebooks for data scientists. Because once you get to millions of rows of data, there’s no other way.
– In Excel you see the data but not the formulas. In Python, you see the formulas but not the data. So there’s always a trade off, but coding is essential for large datasets – which is the key to using Machine Learning techniques.
His One Awesome Idea:
– Using Machine Learning to build an Index Fund that uses AI to track slightly above the market by underweighting potential losers and vice versa.