When the stock market plunged 1300 points in two days on Oct. 10th and 11th, many analysts scrambled to reconcile the market turmoil with U.S. companies reporting strong corporate earnings. While U.S. stocks roared back in subsequent sessions on healthy earnings
from banks and tech stocks, the surge in volatility adds to uncertainty about the U.S. economy.
Was this market rout a one-time event, or did it portend the start of volatile price swings caused by rising interest rates and worries about U.S. trade disputes with China?
In the future, analysts will consult artificial intelligence software to crunch big data, find historical precedents, and run statistics to predict what lies ahead, all within a few minutes.
Increasingly analysts and portfolio managers are turning to A.I.-based technologies such as machine learning to figure out the impact of events on stocks and other financial assets.
One has to look no further than Kensho, an artificial intelligence and machine learning company that has been dubbed the Siri of Wall Street by Forbes.
The A.I. software uses big data to analyze new events, answer questions, produce reports, and predict where markets may be heading.
Kensho has focused on solutions for the buy-and sell-sides that help traders and other people make sense of market-moving events, or news happening in the world. “It answers the question: what statistically happened when that thing happened before,” said Adam
Broun, president and COO in an interview.
To enable this analysis, Kensho built a comprehensive organized history of events and timelines including corporate actions, earnings, mergers, macro-events, weather, politics and geopolitical events, trade events, disputes, and military events.
“If it’s an earnings event or a conference or new product launch or a hurricane, you want a timeline of previous similar events that you can do statistics against,” said Broun.
Based on its models, the system can identify when there is a statistical or causal effect.
For example, say a trader owns stock in Southwest Airlines, but Delta Airlines and Alaska Airlines are reporting their earnings this week. Kensho’s models might identify that Delta’s earnings are having a strong statistical effect on Southwest’s stock price
and alert the trader to pay attention to Delta’s earnings, but at the same time find no effect from Alaska Airline’s earnings.
Last March, Kensho was acquired by S&P Global for $550 million, in an eye-popping deal
that underscored the future of A.I. in finance.
Founded in 2013, Kensho has already made inroads at Goldman Sachs, one of its investors, where it helped sales traders answer questions from institutional clients.
For instance, a client might call up the trading desk at a big broker and say, what events are coming up this week and which ones should I care about, said Broun. “Kensho has enabled those banks to be very responsive and very quantitative with their answers.”
In the past it may take two or three days to get back with an answer, assuming they could get time from the research team. Now, the trader can get answers live while the client is on the phone or within a few minutes, said Broun.
Earlier this year, Kensho’s pre-trade analytics was integrated into the FlexTrade execution management system (EMS) to allow traders to surface connections between their watch list of stocks and corporate events, such as earnings calls. “It says for these
assets, here are the events that are coming up this week that are scheduled and you should pay attention to given past performance,” explained Broun.
While Kensho has been on the forefront of expanding investment analysis with big data, it is not alone.
Many startups are applying A.I. techniques, such as natural language processing and machine learning algorithms to huge data sets to identify alpha-generating indicators.
Experts think that artificial intelligence could help asset managers add alpha to their portfolios and operate more efficiently with their own internal data.
“There is an enormous amount of progress made in this space in last 20 years. And it seems to be on an exponential curve in the last two to three years,” said Dushyant Shahrawat, managing director at FinTech Associates, LLC who spoke at AIR Summit in September.
While A.I. techniques were advanced five years ago, the economics of computing was not available. “Today, you’ve got enormous amounts of data, low cost computing and major platforms from outside the industry that can process this stuff more efficiently,”
“Using the combination of data and AI techniques to generate alpha could create a different shape of returns for the investment industry in the next 5-7 years, said Shahrawat.
Several startups that presented at AIR Summit are utilizing A.I. techniques that can automatically scan massive data sets including earnings calls and regulatory filings.
Finding Signals in Earnings Calls
Listening to earnings calls can be a very time-consuming activity, but what if A.I. technology could analyze every call? Prattle, a research automation firm, deciphers market-moving language from earnings calls by tracking every U.S. listed company. It utilizes
natural language processing or NLP to convert unstructured content into structured data.
“What we do at Prattle is automate the analysis of complex communications,” said Evan Schnidman, CEO and co-founder. In addition to earnings calls, Prattle analyzes central bank communications, said Schnidman, a former academic who is co-author of “How
the Fed Moves .”
During an earnings call, it’s identifying every linguistic pattern and linking it to the historical response to predict how the market is likely to respond to communications. It also identifies who spoke, what percentage of the time, and the sentiment of
every speaker including corporate officers and analysts. In addition, it pulls out key words and phrases that may move the stock price, and are likely to be used in news stories, said the CEO.
In the end, Prattle generates a single quantitative score representing how positive or negative the market is likely to react to a particular communication. “Not only does it streamline your process, it gives you the ability to quantify that language and
plug directly into a model,” said Schnidman. “If someone covers 30, 50, or 100 different stocks it can be alerted which 10 or 12 warrant attention without bothering to dial into an earnings call, “said the CEO.
Data scientists have also turned their tools on analyzing regulatory filings. Skopos Labs scans government and legal events which can have a dramatic impact on companies and on markets, said John Nay, founder and CEO, at the AIR Summit. There are 5,000
federal bills introduced each year across 200 federal agencies, and over 20,000 substantive regulatory documents produced each year.
The firm assembled a team of machine learning PHDs., data scientists, former law firm partners, and Capitol Hill staffers, and built a platform to specifically assess the impact of legislative events. Over the past 10 years it has linked 90,000 events around
bills to specific public companies. Every day it looks at the securities universe and calculates the likelihood that each event will impact those securities. For example, there could be 200 bills that it links as relevant to Exxon Mobile. In each case, Skopos
is calculating the probability that the event will be enacted into law, and what’s the probability that it will have a positive or negative impact on the company. “If it is very positive, it doesn’t matter that it has a 1% chance. It need not score high to
be something that is fundamentally important,” said Nay.
Discretionary fundamental investors are using these predictions for long-term analysis, for screening theories, or for going deep on a new angle in their portfolio, said Nay.
“In the past it was not quantifiable; now they can quantify the risk or opportunity that they face from any company they are holding from law and policy,” said Nay.
Systematic investors can use this to predict volatility or integrate this as another signal in a quantitative investing strategy. “You can create long-short portfolios based on higher level aggregate scores that we aggregate from the executive branch as
a whole,” said Nay.
Kensho is also working with natural language processing to let people conduct smart searches of documents, whether it’s filings, transactions, or third-party research, said Broun. Someone could say, “Find me documents, news stories or social media posts
that relate to Mexico and trade agreements,” he illustrated. The system will not only find that content, but it will learn over time what topics —and topics within these topics—that the user is interested in, and present quantitative information that tends
to back up the things someone is reading, said Broun.
Despite fears that artificial intelligence could eliminate jobs on Wall Street, humans are not going away, though their jobs may change. “Think of it more as a as smart research assistant”, said Kensho’s president. “Smart people with degrees from good places
are spending 80% of their time swimming in stacks of print outs, highlighting things and very little time is spent on thinking or the analytical part of the job,” said Broun.
In his view, the machines will allow people to do much more productive work and much less unproductive work. “The machine is not making a judgment, nor is it telling someone what to do,” he said. “Rather, it is providing all the information that someone
needs to make a decision and do so in a very streamlined and fluid way.”