Authored by Morgan Stanley strategist, Vishwanath Tirupattur
A profound shift is under way among investors. The significance of quant in our clients’ investment process is clearly on the rise. Increasingly, they are applying sophisticated quantitative techniques in investment analysis and alpha generation, beyond pure quantitative or factor investing. Artificial Intelligence (AI) and Machine Learning (ML) stand out among these techniques: at a Morgan Stanley Quantitative Research and Investment Conference a few weeks ago, 66% of the investors surveyed leverage ML for investment research and alpha generation, with another 15% applying these techniques for portfolio construction and risk management.
[ZH: everyone just loves to overcomplicate their lives and come up with the most convoluted solutions, when all they should have done is buy the biggest shorts, and short the most popular names - the best performing strategy in the past five years - period, end of story]
At Morgan Stanley, our fundamental analysts and strategists are increasingly collaborating with quantitative analysts and data scientists, deploying AI and ML to generate insights that wouldn’t have been possible just a few years ago. Four recent reports – on diversity, sentiment analysis, munis, and mortgage prepayments – highlight these efforts.
- Gender diversity is a hot topic among corporates and investors alike. In Introducing HERS: Gender Diversity Pays Off, our analysts analyzed its impact on stock performance. To quantify diversity, they used four gender representation metrics across 1,875 companies in the MSCI World index to come up with the proprietary Holistic Equal Representation Score (HERS). They found that stocks with high HERS outperformed their less diverse regional peers even after controlling for size, yield, profitability and risk. It’s important to highlight that this analysis is not a simple stock screen. Given the scarcity of data for some representation metrics, our analysts used an ML technique called Hierarchical Clustering to create stock groups in each region that were both substantial enough for meaningful statistical analysis and similar enough to reduce industry-specific biases.
- In The Power of Words: Going Global, sophisticated quantitative techniques enabled our researchers to systematically extract alpha from Morgan Stanley Research. Our analysts harnessed Natural Language Processing (NLP) – one of the tools used by Siri – to measure the sentiment embedded in the text of our fundamental analysts' reports. They built a proprietary model with NLP and an AI technique called Convolutional Neural Networks (CNN). The model assigns each report a sentiment score from -100 (most negative) to 100 (most positive). Applying the model to 135,000 company-level reports published in the US, Europe, Japan and Asia ex Japan from January 2013 to May 2019, they demonstrated a positive relationship between the scores and subsequent stock excess returns across all regions, with more alpha outside the US.
- Another innovative application using NLP came from an unlikely corner – municipal securities. Every new muni bond comes with an 'official statement', a disclosure about the issuer and the issue that averages 120,000 words. That means an analyst covering the new issue calendar could easily have to read more than the equivalent of War and Peace several times a week. In What Official Statements Are Really Trying to Say, our muni strategists used NLP to parse the text of 150 official statements for new issues. Their analysis suggested rules of thumb for investors evaluating the calendar, identifying words and syntax with useful clues about future rating agency upgrades and downgrades as well as eventual defaults.
- What drives mortgage prepayments, especially in the new area of non-qualified mortgages? There’s an overabundance of loan-level data on multiple variables, including borrower and loan characteristics, geography, types of property, occupancy, documentation, and loan purpose. Complicating matters, the data are available both at loan origination and afterwards on an ongoing basis. The informational value of many of these variables overlaps, and their impact on prepayments is non-linear. In Good Variable Hunting, our mortgage strategists deployed two ML techniques – cluster analysis and a classification algorithm called Random Forest – to evaluate the importance of different variables in determining prepayments on non-qualified mortgages. Our strategists clustered mortgage loan pools into "fast" and "slow" prepay groups based on prepay curves and then used Random Forest to help investors think about prepayment speeds, shedding light on this nascent area of the US mortgage market.
These reports just scratch the surface of quant/fundamental collaboration at Morgan Stanley. Equally innovative application of quantitative methods is going on across a broad variety of sectors and markets – supply chains, CLOs and asset managers, to name a few others. With more powerful technology, better data and advances in quantitative methods, we expect the integration of quantitative and fundamental analysis to uncover new sources of alpha for our clients.