Updated: Jan 15
We close out the summer scenario analysis blog series with a hard truth:
there are no easy shortcuts to scenario analysis design during a pandemic.
As noted last week, even when the most advanced artificial intelligence (AI) tools are deployed to address the challenge, scenario analysis still requires concrete, recent, relevant information in order to generate meaningful outputs.
The Bank of England today provides us with an assist. They just published over 50 pages of guidance on how to think about scenario analysis and risk management with respect to another situation at the tail of the distribution: climate change.
Concrete Data and Alternative Data
Our regular readers are chuckling at this stage. We have spent the last two weeks highlighting the significant and growing data gaps that arise due to the pandemic (See: Two Data Deficits and Why Data Deficits Matter). Different (but no less important) data gaps exist regarding climate change exposures. Those data gaps arise from the shifting nature of risk management and exposure to address a new area for which no data had previously been collected.
Last week, we pointed out why using bad or incomplete data within Machine Learning/AI systems exponentially increases the risk of problematic outcomes (See: Scenario Analysis Misconception 3: AI Is The Answer). Understanding the limitations of the current data environment is the first step towards devising a risk mitigation plan.
The good news is that pandemic-era data deficits and climate change risk management priorities arise at the beginning of the data revolution, which makes new and different data available for knowledge management and insight formation.
When access to solid traditional data is difficult to acquire, risks can become amplified because they are hidden from view. The large and growing universe of alternative data can provide proxies or substitutes for traditional data vectors.
For example, credit scores enable banks to assess the risk of loss from retail borrowers. However, that measure is meaningless for measuring risk relative to borrowers that have not used traditional bank-based borrowing in the past. For those borrowers, alternative data regarding financial reliability (e.g., payment history from utility bills, repayment history regarding alternative lenders, payroll deposits, average bank balances) can generate meaningful credit history information as many FinTech companies have discovered in recent years.
Alternative data regarding public policy risks increasingly are also becoming available. These data sets cover a broad spectrum from plain-vanilla traditional policy data presented digitally (e.g., voting records, campaign contributions, polling data) to our own newly created data from the language of the public policy process.
The point is that concrete, credible data can be found to plug the gaps so that scenario analysis can continue to generate meaningful insights.
Recent Data – The Importance of Nowcasting
Every risk manager and data scientist knows that all data is not created equal. It needs to be “fresh.” Scenario analysis extrapolates patterns and projections from data based on a key underlying assumption that somehow past behavior is indicative of future behavior.
As we noted for Interactive Brokers last year (before the pandemic erupted),
“Because nowcasting relies on the most current data in a series (e.g., the latest quarter data release), it focuses like a laser on the real-time inflection point implied by the data. In other words: it is not drawing conclusions or making outlooks based on past historical data. It instead assesses whether the current trajectory is consistent with, or departing from, the previous trajectory in order to assess whether an inflection point may be materializing.”
Adopting a nowcasting approach but using old data is fundamentally a wasted effort, particularly during periods of significant upheaval. Whether one focuses on the pandemic or climate change or the data revolution or political polarization or anti-globalization, or the Black Lives Matter movement, the fact is that behavior patterns at all levels of the economy and society currently are proceeding through a significant paradigm shift.
Decades or even years of data right now will not be nearly as meaningful a foundation for analysis as recent data. The current environment requires scenario analysis to shift focus towards a nowcasting approach, which implies a comparable shift in sourcing data inputs focused on the recent past.
Choosing appropriate data for crafting policy-related scenario analysis is as much art as science. It requires finding structured data that specifically addresses the issue at hand in a relevant manner.
Consider today’s climate change scenario analysis best practice guidance from the Bank of England regarding carbon tax scenario analysis:
“In developing a scenario, or assessing the plausibility of an existing scenario, it is important to consider climate policies across three dimensions:
• Timing: the economic consequences of emission reductions will be different depending on whether actions are taken sooner or later.
• Scale: refers to the speed and force with which climate policies are imposed, as well as their coverage; climate policy implementation could be smooth and gradual or could be accelerated, uncoordinated and abrupt.
• Fragmentation: addresses the degree of co-ordination across countries in tackling climate change.
Firms with exposures in one or a limited number of countries may of course choose to start by considering one country at a time.”
The point is to identify the potential financial impact on portfolio structure. Finding data to support robust scenario analysis concerning shifting policy priorities within one or more countries could be a challenge.
We believe that alternative data drawn from the public policy cycle can help risk managers immensely in this context. Converting the words used in the public policy process (using our patented system) provides structured data for scenario analysis. This more current and targeted data helps risk managers identify promptly shifts in language and policy trajectories.
For example, we can tell you that 2020 has been an active year for policymaking regarding climate risk disclosures. But some might discount that statement as biased. However, structured time series data demonstrate graphically not only that action is at an all-time high as of today but that policymakers are taking action directly....without leaks to the media (which would show up as red) and without major attention from major media (which would show up as blue):
This is the frontier of risk management. It has never been done before. But now that it is possible to convert unstructured verbal data into structured data using our patented process,