AI Powered ETFs – A Look at iShares Evolved
When the world’s largest investment manager decides to build AI powered ETFs, you know that artificial intelligence is now mainstream. With offices spread around the globe housing 13,000 employees, Blackrock (NYSE:BLK) has more than $6.2 trillion in assets under management. To put that number in perspective, all the of the world’s 2,397 billionaires have around $7.4 trillion. It’s safe to say that when Blackrock builds an investment product, they’ve done their homework.
iShares Evolved – AI Powered ETFs
About one-sixth of their Assets Under Management (AUM) can be found on Blackrock’s Exchange Traded Fund (ETF) platform, “iShares”, which offers over 800 products to institutions and individuals. In March of this year, iShares launched seven US sector ETFs using a new approach to sector classifications powered by artificial intelligence.
Industry Classifications – The Old Way
In our earlier article on “Artificial Intelligence and Industry Classifications”, we talked about “the old way” of doing industry classifications such as the Global Industry Classification Standard or (GICS). While this method provides some kind of hierarchical sector and industry structure, it relies on backward-looking product and service inputs, and requires companies to sit in only one sector/industry basket. Classifications are done manually by John-in-Mumbai type analysts, then reviewed by developed market teams who point out all the mistakes that were made. Classification reviews are typically performed infrequently, usually about once a year.
This kind of approach doesn’t accurately represent companies that either have business activities in more than one basket or change business activities frequently. Take Amazon for example. They started out as an online book retailer but have since expanded into other business areas like Amazon Web Services (technology) or Whole Foods (discretionary spending). In order to dynamically capture these changes, we need a more robust method of classifying companies by industry.
Industry Classifications Using AI
Blackrock is now proposing an overhaul of these old classifications using Natural Language Processing (NLP) and machine learning to browse through companies’ public documents like regulatory filings or earnings reports, and look for keywords and phrases used by these companies to describe themselves. Companies then are grouped into sectors and industries based on similarities in the language used.
This method provides a forward-looking and automated result which is also more flexible than legacy structures. Here is a look at how these sectors currently differ from the market leading GICS classification according to Blackrock:
There are some significant differences between the two, but this looks far from a complete overhaul. This is only the first benefit of the new classification though. Blackrock’s AI recognizes companies present in multiple business areas using the same NLP process and assigns weights to each business. Each company is represented in a sector index proportionately to its weight in that particular sector, multiplied by its market cap.
The Difference Between AI and Humans
These new strategies sound quite advanced, but what we want to look at is just how different the output is. Let’s take a quick look at one of the iShares Evolved ETFs to see the differences when compared to a more traditional equivalent U.S. sector ETF. We’ll use the financials sector as an example. To keep things simple, we will henceforth refer to these two strategies simply as “AI” and “Human”.
|iShares Evolved U.S. Financials ETF||iShares U.S. Financials ETF|
|Number of stocks||261||292|
|Top holdings||JPMORGAN CHASE & CO, 8.79%||JPMORGAN CHASE & CO, 7.24%|
|WELLS FARGO, 8.44%||BERKSHIRE HATHAWAY INC CLASS B, 7.16%|
|BANK OF AMERICA CORP, 8.01%||BANK OF AMERICA CORP, 5.49%|
|US BANCORP, 2.7%||WELLS FARGO, 4.47%|
|GOLDMAN SACHS GROUP INC, 2.55%||VISA INC CLASS A, 4.44%|
|Assets Under Management||$5 million||$2.26 billion|
First thing that stands out are the low fees for the AI ETF, something that has a huge impact on performance over time. (It makes one wonder if Blackrock is concerned at all about cannibalizing the $9.7 million in fees coming from their legacy sector ETF. Note that the legacy ETF uses a Dow Jones index so that’s probably costing them something.)
Next, let’s take a look under the hood to see how the two ETFs differ in respect to the stocks each contains:
Here we can see that both ETFs differ quite significantly. Only 142 stocks are found in both ETFs. Let’s take a look at the top-10 stocks by size which are included in the AI ETF but not in the human ETF:
|UNITEDHEALTH GROUP INC||0.76%|
|AUTOMATIC DATA PROCESSING INC||0.55%|
|FIDELITY NATIONAL INFORMATION SERV||0.51%|
|WORLDPAY INC CLASS A||0.28%|
|TOTAL SYSTEM SERVICES INC||0.24%|
Then, let’s look at the top-10 stocks by weight that the humans included but that AI completely avoided:
|PNC FINANCIAL SERVICES GROUP INC||1.33%|
|AMERICAN TOWER REIT CORP||1.13%|
|SIMON PROPERTY GROUP REIT INC||0.92%|
|CROWN CASTLE INTERNATIONAL REIT CO||0.79%|
|AON PLC CLASS A||0.66%|
|PROLOGIS REIT INC||0.64%|
|PUBLIC STORAGE REIT||0.57%|
Kind of ironic that the AI decided not to include Blackrock. And what about Citigroup, one of the “big four” banks that hold 39% of all U.S. customer deposits? Surely the AI must know something we don’t.
What’s even more interesting is that 74% of the stocks that are only found in the human ETF are actually Real Estate Investment Trusts (REITs) which the AI algorithms seem to have largely avoided. (Searching the AI ETF for mentions of the acronym “REIT” only returned six results.) This makes sense because GICS actually added an 11th sector for REITS back in 2016 implying that they all behave differently enough to merit their own industry classification.
Broader comparisons of performance, liquidity, or assets under management are not going to be meaningful at this time because of the recent launch date of the Evolved offering. It may also make sense to take a look at another pairing of “AI” and “human” ETFs to see what we can learn from one more comparison. The avoidance of REITs by the AI algorithms obfuscated this comparison a bit.
Other AI Powered ETFs
Blackrock is not the only company creating AI powered ETFs. Recently we looked at a company called Innovation Shares which uses AI to select stocks that have exposure to blockchain technology. This is quite similar to what Kensho Technologies does, creating non-traditional “21stcentury” sectors and sub-sectors like space, robotics, or clean tech using NLP. Of course the methodology for weighting these various indices differ as well.
Kensho partnered with State Street Global Advisors to launch three passive ETFs on the Kensho indices, an event which took place earlier this year. These SPDR ETFs use representative sampling in their investment process which means not all constituents that appear in the index are held by the fund, but the performance is tracked using a very tight margin of error. Then Kensho was acquired by S&P Global leaving everyone to speculate as to what that might mean for future ETF offerings that run on Kensho indices.
The iShares Evolved offering is managed by Blackrock’s Systematic Active Equities division which means there is more leeway for managers to deviate from the original benchmark index. Blackrock calls their method “systematic, quantitative technique” which suggests fund managers aren’t picking single stocks, but instead applying some kind of proprietary rules-based algorithm to make their stock picking decisions. This kind of approach is blurring the lines between traditional iShares passive investments and Blackrock active strategies, creating a new space that the company has been piloting over the past years.
Slicing and dicing the investment opportunity by putting companies in multiple brackets is nothing new. Financial services firm MSCI has done this with their Economic Exposure indices which look at a company’s revenue sources by country instead of its incorporation or listing. Similarly, their Value and Growth methodology also allows companies to be partially present in both value and growth. The real novelty in the iShares Evolved methodology is the combination of AI and flexibility that allows the indices to become more representative and reactive to market and corporate events. That’s important in a world where global corporates are evolving more quickly as evidenced by examples like Amazon. If you’d like to read about the iShares Evolved approach in greater detail, check out their white paper on the topic.