Rapid technological changes have blurred the lines between industries as companies bring together seemingly unrelated lines of business in unconventional ways. Awada’s new research, along with that of Harvard Business School Professor Suraj Srinivasan and doctoral candidate Paul J. Hamilton, uses machine learning to identify such nuances in companies.
Since 1999, analysts and portfolio managers have used the Global Industry Classification Standard (GICS), a taxonomy that includes 11 industries, 25 groups of industries, and other subsets, to compare stocks. Standard & Poor’s, Morgan Stanley Capital International, and other companies who maintain the system review the categories on a regular basis.
“BUSINESS MODELS DRIVEN BY DIGITAL TECHNOLOGY HAVE MADE THE BOUNDARIES MORE DIFFUSE IN THE LAST 20 Years.”
The researchers, on the other hand, used machine learning to analyze 10-K filings dynamically, weighting each company across 15 “TOPICS” and three levels. In a study published in February, researchers found that portfolios of long-short equity designed with TOPICS classification outperformed portfolios created with GICS categories by up to 2.5 percentage points in terms of risk-adjusted return for sectors such as health care, utilities, and energy.
The TOPICS method employs sophisticated financial models that reveal hidden similarities between risk and return profiles of stocks, which appear to be different. This shows new investment opportunities. The models can also be used to highlight the opposite profiles of sectors.
Recently, we spoke to Awada & Srinivasan about the study. They are the Philip J. Stomberg Professor of Business Administration.
Danielle Kost: Why should investors reconsider the way they categorize companies?
Srinivasan: Conglomerates were the norm 30 or 40 years ago. They have changed with time. There were a lot more pure-play firms. For a while, the industry classifications had some meaning. The system was not perfect, but it was the only option.
Digital technology-driven business models have shifted the boundaries of businesses in the past 20 years.
If you look at Walmart and Amazon today, their business models have evolved over the years. Many companies are turning into tech companies. Apple Pay makes the difference between Visa and Apple disappear. Amazon is now competing with pharmacies.
What defines the industry membership of an organization? Where does an industry begin and end? Parallel to these are our abilities to use alternative data techniques and assess a business.
“WE WANT MONEY AND ALWAYS LOOKING FOR ALPHA IN OUR STRATEGIES OF INVESTMENT, SO MACHINE LEARNING CAN HELP US BE PROFITABLE AND IMPROVE OUR RISK MANAGEMENT AND TRADING EFFICIENCY. SO BE IT.”
Kost: Machine learning is a key factor in this case. What is the institutional investing community’s attitude towards artificial intelligence? Are investors embracing it? Are there doubts?
Awada: From my past work [as a hedge fund manager], I know that we are always at the forefront of machine learning. The bottom line is that we are always trying to improve our trading and risk management and want to be profitable.
Srinivasan: It is definitely being adopted. Machine learning is the newest quantitative technique. But it’s also being used in other areas of investing.
Human-machine ideas are changing the way things are done. Machines can do things that humans do not have to. In many cases, we may still need a human to help us.
If you are a quant trader, then you will always be improving your models to drive your trading strategy. Machine learning is a current way to achieve this. In other places where it was more human-based decision-making, it’s now human-plus-machine decision-making.
Kost, Do you see a potential shift away from GICS?
Srinivasan: We create certain standards to help us understand and communicate our work. It provides a set of common techniques that everyone can use. It isn’t easy to give up something unless you have a better alternative. It’s embedded in models and contracts.
If you have a measurement system, a fund manager is paid according to how much their return exceeds that of the GICS portfolio. This is fixed, so if I am a fund manager, I will get paid according to how much better I do than my GICS segment. Then, I am stuck using the GICS framework. You have to tell us what we should use instead of GICS if you say that GICS is no longer available.
Simple solutions become more difficult when I declare that the world has become more complex. How can I determine a person’s pay based on their performance? It isn’t very easy to keep changing the benchmark. There’s an advantage to using standard frameworks.
Awada: I have a more practical approach. Portfolio returns have been negatively affected in the last five to six years, particularly during times of market distress.
Take COVID 2020. You had stocks in certain industries behaving a certain manner and not being aligned.
Look at Amazon, for instance. Walmart’s stock price was falling, while Amazon’s was rising. If you’re a portfolio manager and you look at the retail industry, you might be puzzled as to why Amazon is performing so well while Walmart is not. Amazon’s success is due to the fact that it is a technology firm in many ways and it deals with information technology. This is more than just selling records, books, and retail online.
It’s possible that many portfolio managers began to see a misalignment of performance between stocks supposedly belonging to the same GICS sectors. This led to big losses in their books. Fund managers sought to mitigate the risk by adjusting their risk profiles. How can I protect myself against the risk of a portfolio with retail stocks and tech stocks? You’re not allocating the right amount of notional risk to that sector if you put Amazon in the retail category. The sector classification becomes critical to the way in which you manage conceptual risks.
Kost – How can finance reconsider other benchmarking approaches?
Awada: Is the S&P 500 representative of the market or not? Why 500 stocks? Why don’t we have stocks with a clear sense of the sector? According to our methodology, 300 stocks could represent the real market.
I FEEL THAT THE S&P CAN BE A DIFFICULT INDEX DOWN THE ROAD.
Euro Stock 50 is the index or stock that represents Europe. The number of stocks is not as important as how representative they are. The S&P 500 will face competition in the future as an even more representative index that gives investors a better idea about where to allocate their funds.