#AltData, #Volatility and Language
Our latest analysis of why and how it is crucially important to view language as data, originally published in our LinkedIn Newsletter (AltData and Volatility Frontiers):
Two weeks ago, we made the case that alternative data vendors must speak Greek to the capital markets because Greek is the language of volatility. We close out the month of February with a very different kind of volatility and disruption that sadly makes our point for us with respect to language-derived data.
The unacceptable military invasion of a sovereign nation deliberately upends the post-WW2 infrastructure. The fabric of economic interdependence is unravelling through economic and financial sanctions as the world justifiably reacts to the disrespect of territorial sovereignty and the assault on the right to choose one’s leaders and one’s trade treaties. This newsletter is not the place to discuss the substantive issues. My colleagues at the Atlantic Council and across Wall Street have been doing a fabulous job of providing that analysis.
This is the place, however, to observe that the geopolitical risks on display right now
are expressed first, and foremost, through words.
Action may speak louder than words,
but words always precede action.
Our patented technology makes it possible to measure the language and related risks quantitatively, daily, globally, and.....objectively.
Technology makes it possible to adopt a data-driven approach to risk assessment regarding those (and other) words. Let’s dive in to a world where language is data. And why we must separate our normative judgements from those words in order to identify actionable trends.
Language as Data: The State of Play
The relationship between the policy process, the news flow and market volatility is well-established. Policymakers act. Journalists report. Markets price the implications. Markets accordingly acquire news content in machine-readable form in order to accelerate discovery of actionable information. Machine reading and machine learning intensify the reaction function by automating and accelerating pattern recognition regarding those words. Accelerated access to actionable information improves both the quality and the speed of financial decisions.
The operating assumption has been that the first, best, fastest route to acquiring machine-readable language for signal development involves the news cycle.
To date, markets have prioritized sentiment analysis as the preferred tool for acquiring language data at scale. Applied first to headlines and now to full-text search, the focus has been on discerning the implicit sentiment of the speaker in order to acquire better hints about potential future policy decisions.
Sentiment analysis has its role to play in policy analysis,
but it also has limits.
Official sector speakers are not always guided by sentiment.
They may prefer one outcome, but economic data or the broader political environment may constrain their capacity to act on those preferred choices.
More importantly, the process of converting one set of language (policy language) into another set of language (sentiment analysis) still relegates the analytical process to a verbal one. The number generated by sentiment analysis refers back to a verbal or emotional construct, not an actual financial risk that portfolio managers can use in their daily workflows. Integers deployed to convey emotion Also create material risk of importing bias into otherwise objective numbers.
Consider the current situation. Troop movements and missile launches are physical acts, but they occur only after verbal statements, declarations, and commands have been issued. Economic sanctions are concrete acts with real consequences on markets and economies, but they are expressed verbally first. The words trigger action. It follows then, that the words and the volume of words in any given 24 hour period needs to be measured.
The earliest signals are often missed by markets because they do not make it into the news coverage.
High quality fact-checked journalism is a
necessary but not sufficient condition
for identify looming risks from language data because small
technical issues tend not to make it into the news stream.
The fact that the newsflow only shows the tip of the iceberg for strategically significant developments expressed verbally is NOT the fault of the media. Some developments may not be deemed newsworthy at the time, even if they are strategically significant to a specific portfolio or investment thesis. Journalists and your external experts are human; they can only read so much at any given moment in time.
The “hot issue” of the day generates the media version of a crowded trade: journalists seek a unique and insightful angle on the hot topic. When media coverage focuses on a specific issue, it is an excellent opportunity for policymakers to use distraction to their advantage by releasing information publicly that others might not notice amid a noisy news cycle.
Again, consider the current situation. Capital market automated news feeds are understandably dominated by the geopolitical situation in Ukraine and the fallout from the economic sanctions. But where in the last 24 hours did you see anything in the media cycle related to the upcoming major monetary policy meetings at the reserve currency central banks starting this week?
War and broad-based economic sanctions against a top 10 global economy have material implications for inflation and interest rates. Wouldn’t you like increased visibility into what policymakers say as they approach these meetings?
Language as Data: The Frontier
Measuring the volume of activity regarding technical keywords (as our patented process provides at BCMstrategy, Inc.) delivers increased visibility into the policy process using publicly available, actionable information. Measuring the activity automatically delivers an additional layer of objectivity. A machine never needs to “get your head around” a situation. It is never upset about what it sees. It does not need to adjust its framework to account for the possibility that a broader range of outcomes has just become possible. The machine just reports the data; it is never distracted by the implications.
Consider today's PolicyScope data (28 February 2022) regarding monetary policy:
Underlying inflation trends have not disappeared because Ukraine was invaded. In fact, some inflationary trends (particularly in relation to energy prices) will intensify in the near-term. But there was more action in the last 24 hours on other important issues, like financial stability.
The work of government and monetary policy formation continues...and has become more complicated. Only by measuring momentum in relation to the words can one see the activity. And only BCMstrategy, Inc. holds the patent for measuring the words.
Treating language as data and using volume-based, objective metrics for that data makes it to spot the signal despite the news cycle. It also becomes possible to spot when policy language and the news cycle are amplifying each other. Reducing language to an integer eliminates the human emotional reaction to the words, facilitating the kind of analysis that supports investment decisions. You may not like a trend, but as capital markets know, “the trend is your friend.” Now you can measure that trend in the language, in real time. Welcome to language-derived alternative data.