19 Jun 2019
Not sure how to decipher interest rate speak? Don’t have time to listen to a CEO conference call? No problem – there’s a bot for that.
Get ready for the disruptive power of natural language processing (NLP) tracking every word a bank governor, company CEO or CFO has ever uttered. Algorithms are not just for targeting Facebook ads at your latest whim. The world’s largest fund managers, brokers and analytics firms are in a real-time digital arms race using algorithms to divine corporate speak from spin.
"In the world of words, nothing is more arcane, or valuable, than cracking the enigma code of the central bank’s interest rate speak.”
Why read 20 pages or labour through a conference call when an algorithm can alert you when the company’s investment thesis is pivoting? Or when voice forensics suggests the sector’s top analyst has changed their view before the broker note hits the street?
The signal in the noise at the recent U.S. National Investor Relations Institute (NIRI) conference was the ‘bots’ have arrived and corporate communications may never be the same again.
For those in the words business who are diligently focused on qualitative tone, the rise of the machines may seem all a bit geeky. But take a decade of stock exchange announcements, underpin them with terabytes of audio and video recordings, overlay with 24/7 tracking of consumer and public commentary and you have the building blocks of fact checkers, reputation analysers and lie detectors.
As Bloomberg’s chief equity strategist Gina Martin-Adams puts it, the future is likely to be “less about the narrative and more about the data”.
In stock picking, where a 1 per cent advantage can be the difference between average and out-performance, securing a ‘quantamental’ edge is gold.
“Quantamental” is the intersection of data and intuition. The quantitative data overlays traditional bottom-up valuation approaches, acting as a brake, boundary or evidence check against inbuilt subjective bias.
BlackRock, the world’s largest fund manager, is setting up an artificial intelligence lab in San Francisco to crunch alternative data sets - not just the content posted on the stock exchange.
In the world of words, nothing is more arcane, or valuable, than cracking the enigma code of the central bank’s interest rate speak. Remember Alan Greenspan’s famous aphorism – perhaps apocryphal – about bank-speak: “If I seem unduly clear to you, you must have misunderstood what I said.”
Enter Prattle - a tech company dedicated to deciphering central bank communiques.
Prattle’s founder, Evan Schnidman, says in central bank speak “modest” is at the “dovish” 30th percentile spectrum of interest rate speak while “moderate” is at the “hawkish” 70th percentile end, based on the US’s Federal Open Market Committee’s (FOMC) published notes.
Recently acquired by global ‘dark pool’ trading platform Liquidnet, Prattle also analyses thousands of stock exchange statements and conference calls to compare every semantic sentence of every CEO and CFO.
UBS’s Evidence Lab offers alerts aptly named the forward-looking score, stability score, surprise score and concern score.
Amenity Analytics has a conference call ’deception monitor’ to pinpoint which topics company management are most uncomfortable, uncertain, hostile or evasive about.
But algorithms are not just for traders, they’re headed for the C-suite.
“If you trust a map app more than your spouse to get you across town, who is to say in 10 years’ time your most trusted colleague or director won’t be your co-bot?” says Malcolm Frank of IT services and analytics giant Cognizant.
Think of Amazon’s Alexa in the boardroom or a Google Assistant on your desktop reviewing a draft company announcement for legal, reputational or share price impact to ensure the “best way to deliver bad news” and prevent “the worst way to deliver good news” says Frank.
Case in point: the Reputation Institute recently bought media analytics group Mettle to correlate the impact of traditional and social media on corporate and brand reputation.
Measure it in real time and you can forecast sales, earnings and ultimately company valuations.
But text on a page is just one part of the artificial intelligence equation. Voice diagnostics are almost here.
In healthcare, as Bill Gates noted, a voice app might soon predict Parkinson’s or Alzheimer’s disease. In law enforcement, the US Coast Guard is using Carnegie Mellon University’s Language Technologies Institute to create voice profiles and find hoaxers making bogus emergency calls. And when it comes to telling the truth, research tends to reinforce human intuition - a few extra milliseconds are needed to tell a lie given the additional mental processing required.
So what might the conference call of the future sound like?
“Good morning everyone, before we begin could I ask you to turn your lie detector app to off and your sentiment analysers to neutral.
It has come to our attention that due to an overnight software upgrade your device may be using old code.
So let me foreshadow in my script today I will use the words sound, strong, and robust 86 times and say weak, slow and volatile 44 times.
This produces a sentiment score outside the standard distribution of our past 16 calls, but it reflects a series of one-off semantic events.
Overnight we ran the presentation script past our PR advisors, legal counsel and the stock exchange’s own price-semantics app and they all showed the tone and ratio of key words to be consistent with previous years.
For those analysing my voice, the hesitation is because I have a cold, not early onset dementia.”
The views and opinions expressed in this communication are those of the author and may not necessarily state or reflect those of ANZ.
19 Jun 2019
05 Jul 2018