Each day, Wall Street churns out millions of words encouraging investors to buy or sell stocks, bonds and mutual funds.The article claimed that Artificial Intelligence, or AI, is taking over much of the task of analysis:
In the future, more of those words might not be written by humans.
As automation in financial services grows, computers and algorithms have taken on some of the traditional work of traders, clerks and financial advisers. Now, a host of startups that use artificial intelligence to write news stories and other reports have set their sights on writing work at banks and financial-service companies.
Narrative Science Inc., which launched in 2010 with computer-generated news articles, added products for financial-services businesses in 2013. Those firms now account for 60% of the company’s client base. Last year, Automated Insights added insurance company Allstate Corp. to a roster of clients that includes Yahoo! Inc.and the Associated Press. Other startups offering automated reports for financial-services firms include Yseop, Capital Cube and Goldman Sachs GS 0.37 % -backed Kensho Technologies Inc.
As artificial intelligence takes on ever more tasks, Wall Street is getting more comfortable putting it to use. Services launched in recent years are now gaining traction because the technology has become more sophisticated and banks are looking for ways to cut costs and increase efficiencies.
What do analysts do?
To evaluate the claim that AI could take over the job of company analysts, let`s consider the job of a sell-side equity analyst, from the tasks that are the easiest to replicate to the ones that are the hardest:
- Report on industry and corporate developments
- Issue buy and sell recommendations
- Communicate insights about a company and its industry
Most individual investors focus on (1) and (2), while institutional investors appreciate and pay for (3). Indeed, the job of reporting on industry and corporate developments could be automated today. A number of news organizations have experimented with using software to put out company earnings reports using a standardized template.
The "quant" investing robot
The task of issuing buy and sell recommendations have also been computerized for years. It's called quantitative investing, but differs from fundamental investing in a key way. Typical quantitative investing techniques rely on factors, such as the insight that "low P/E stocks tend to beat the market". So what you do is buy a whole bunch of low P/E stocks.
Where quantitative investing differs from the kind of fundamental investing done by company analysts is the degree of confidence in the forecast. Quants tend to take a statistical approach and buy a large and broad portfolio of low P/E names, whereas a company specific fundamental analyst will have a far higher level of confidence in his forecast.
Richard Grinold pioneered a principle of portfolio management called "The Fundamental Law of Active Management" which boiled down to the idea that you should bet in proportion to the degree of confidence you have in your forecast alpha (see my previous discussion Examining your assumptions: The Fundamental Law of Active Management). Based on those ideas, a typical quant portfolio will hold between 100-300 stocks, because it is betting on factors or models, whereas a typical fundamentally driven stock picking portfolio will have 20-60 names.
Today, you can buy quantitative system insights from outfits like Value Line, which has had an impressive multi-decade track record, and ranks stocks with Timeliness rankings of 1 (best) to 5 (worst). Does that count as a robot replacement? (Yes, I know that Value Line has human analysts, but not all ranking systems do and I use Value Line purely as an example of a quantitatively driven buy-hold-sell system.)
What is fundamental analysis?
It is the more "creative" part of fundamental analysis that will be difficult for AI software to replicate and model.
Let me give you an example. Early in my career, I had the privilege to work with a small cap analyst who went on to be a portfolio manager and now the co-head of equities of a major asset management firm. From watching what "Doug" did, I gained invaluable insights that made me a much better quant later in my own career. Small cap analysis is different from being an industry analyst because the drivers for each company and industry are different. When he picked up coverage of a company, Doug would spend several weeks chasing down what he believed the important drivers of value for a company by speaking to the company, its competitors, suppliers, customers and industry association. He would often waste days and weeks chasing down blind alleys only to discover that the issue that he was researching was irrelevant. Only once he had developed a framework for understanding how the business worked would he actually build the spreadsheet for valuing the company`s shares.
Can AI software do that? I doubt it, largely because we don't fully understand how human creativity works yet.
No doubt, we can build software to perform analysis according to certain standardized templates. Perhaps smart software might have been able to read financial statement footnotes to figure out the kinds of risks that Enron had been taking, as the Street seemed oblivious. Can software do channel checks to gain insights as to the public perception of a new product launch, maybe. Those levels of sophistication are probably one or two generations ahead - for now.
As well, investment insights are multi-dimensional and different fundamental analysts may have different insights that institutional investors find valuable and will pay for. As an example, I recall that as a portfolio manager, I derived different levels of understanding of tech darling Nokia during the late 1990's from different analysts.
The American analysts were much better at business strategy. Nokia was the dominant producer of cellphones and had enormous economies of scale advantages to their then rivals like Motorola, Ericsson, Samsung and others. Nokia's handset operating margins were in the high teens whereas the margins of their best competitors were in either high single digits or low teens. That, reasoned many US analysts, was a source of competitive advantage for Nokia and what made the stock a buy.
By contrast, the European analysts were much better at on-the-ground analysis of the success (or failure) of new handset models rolled out by the major European handset makers. That kind of channel check insight was also a source of tactical advantage. In effect, I read research from the American brokers for the big picture and spoke to the local European brokers because they knew where all the bodies were buried.
Risk is multi-dimensional. Alpha is multi-dimensional. We are not at that stage of software development where AI can give us insights for everything. True, many of the basic reporting functions can be replaced by software - that's not where the real analytical insight comes from.