“Risk Thinking” in an Unstable World
Businesses are being asked to measure their financial risk of climate-related change. But no one seems to know how
In October 2018, Ron Dembo was driving through an area of burned forest in Northern California.The CEO and founder of Toronto-based Riskthinking.ai, Dembo had been invited to give a keynote speech at the conference of the California Independent System Operator (CAISO), one of the world’s largest electricity distributors. Dembo had previously built Algorithmics Inc., which became the world’s largest enterprise risk management software provider. He was headed to the conference to discuss a new kind of “risk thinking” that he believes could supplant existing forecasting strategies.
“Every decision we make today affects our future,” Dembo says. “Yet, whether we are a corporation or an individual, our decision-making today is primarily guided by our attempts to forecast our future. Traditional forecasting doesn’t work well, because past data is of minor use in our fast-changing climate reality. Nearly every commercial and non-commercial sector—from finance, insurance, energy and transportation to local and federal municipalities—is affected.”
Conference organizers in California initially asked Dembo to avoid mentioning climate change, a request that was rescinded only after two months of haggling. As he drove through the burned forest on his way to the conference, he saw that work crews from Pacific Gas & Electric (PG&E), the state’s biggest utility, were installing new transformers in the same forests that had just burned down and in the same spots where they’d been previously installed—that is, the same forests that were prone to drought and fire, often caused by sparking transformers.
During his talk, Dembo showed a map of California with locations of transmission wires and transformers and how this is exactly where scientific analysis would predict fires to occur. The nature of a risk-thinking solution would be to hedge possible future events. “There is a need to replace forecasting with risk thinking,” Dembo says, “and to compare risk under alternative, forward-looking strategies and across institutions, stress-testing decisions under a well-defined and consistent set of scenarios.”
For example, if there were a likelihood of strong winds coupled with drought and extreme heat at some point in the future, PG&E would know that this could lead to one of its transformers igniting a significant fire. Many of the transformers pass through forests susceptible to fire, making such a scenario quite feasible given our changing climate. One possible hedge, although controversial and costly, is to shut down electricity in affected areas to prevent transformers from sparking.
Only a month after Dembo’s conference in Northern California, this scenario occurred. In November 2018, a spark from a transformer caused the deadly Camp Fire, which took the lives of 85 Californians and destroyed nearly 14,000 homes and other buildings. PG&E, one of the largest utilities in the world, lost 85% of its market value and went bankrupt just two months after this fire.
Now, a little over a year later, this “hedge” is being used to manage difficult scenarios. In an effort to reduce fire risk, PG&E has turned off electricity for days at a time, causing significant losses to the Californian economy.The cost of such a “hedge” is huge. Had the utilities been able to measure the risk and the cost of mitigating that risk back in 2018, they likely would have seen the need to move some transmission lines and/or bury some transformers to avoid the huge costs of shutting off electricity. This is a complex issue that involves multiple stakeholders and decisions, such as who should pay for mitigation. Still, risk thinking could have been used to put a value on mitigation versus simply betting that the event would never happen, as was the case.
Dembo argues that decision makers need to use prospective thinking skillfully to help navigate our more uncertain and unpredictable world. Risk thinking results in an action to move forward coupled with a hedge to manage extreme risks that may result because of uncertainty. In contrast, traditional planning involves following a forecast that changes periodically.
“Climate change will affect the balance sheets of banks and insurers through two new channels,” says Lucy Nottingham, a director at Marsh & McLennan Insights. “First, increasingly extreme weather- and climate-related damages will impact insurance claims and the creditworthiness of clients. Second, low-carbon regulations and clean technologies will disrupt many businesses, leading to potential investment losses and credit losses for financial institutions. These dynamics are not captured in financial models that rely on historical data. Quantifying climate risk requires a new breed of forward-looking, scenario-based models that embrace climate science and anticipate future regulatory trajectories.”
Scenario Analysis Challenges
One method for anticipating future risk or value is called scenario analysis, a process of considering a range of plausible scenarios against a particular mid- to long-term time horizon—say, 10, 25 or 50 years from now.
In December 2015, the G20’s Financial Stability Board, an international body that monitors and makes recommendations about the global financial system, established the Task Force for Climate-related Financial Disclosures (TCFD) to help companies, banks, sovereign funds, fund managers and others to measure the financial risk of climate-related change to their assets and operations. The task force recommended using scenario analysis for this work.
However, financial institutions that have committed to voluntary disclosure say they face specific challenges in adopting the recommendations, including the application of scenario analysis to calculations of financial risk embedded in varying asset classes, investment portfolios and industries. According to a 2018 report from Marsh & McLennan Companies Global Risk Center titled “Reporting Climate Resilience: The Challenges Ahead,” applying scenario analysis to climate change is one of the top three challenges for companies in adopting the task force’s recommendations.
The report states: “Determining the best approach to modeling climate scenarios and mapping a pathway…for incorporating climate risk into future financial planning is a daunting process.
“Organizations face a broad array of scenarios…Predicted outcomes vary widely across even the most authoritative models. As one company noted, ‘In some sense, the process involves a lot of very educated guesswork, but not everyone guesses in the same way.’”
This is where Dembo’s risk thinking gets interesting. He has derived a science-based algorithm for generating scenarios, thereby taking the guesswork out of risk thinking. What’s more, he can, in many circumstances guarantee that his algorithm will produce the very best and very worst scenarios—the “tails.”
Sanjay Khanna, a futurist at Baker McKenzie and a board advisor to Riskthinking.ai, emphasizes to Leader’s Edge the crucial importance of scenario analysis, along with its challenges. “Scenario development is notoriously complex and prone to selection bias. Instead of carefully weighing challenging plausible scenarios, organizations may choose, based on groupthink or other cognitive biases, to favor convenient scenarios aligned with the current business plan and strategic time horizon. The gap between more convenient scenarios and more challenging ones reflects a critical risk gap that must be made tangible for decision makers.”
Dembo agrees. “Riskthinking.ai’s proprietary algorithms use artificial intelligence to generate scenarios which create consistency, repeatability and the ability to compare risk across institutions,” he says. “In particular, we solve the problem of generating consistent scenarios for transition and physical risk modelling as required by TCFD.”
Nick Martin, U.S. sustainability practice lead for engineering and environmental consulting firm Antea Group, describes the TCFD recommendations as doable. “But in practice,” Martin says, “it is very difficult to not only define scenarios but more importantly to translate into financial risk and actionable decisions. Dembo clearly sees this frustration brewing and as a niche for a new method/tool.”
AI in Risk Thinking
So how does Dembo’s method work? “One source that we have not tended to use can provide useful information for scenario generation,” he says. “In particular, using experts, coupled with machine learning, we aggregate the best data. That is, the forward views of many experts with the latest scientific analysis. We aggregate data from all available sources and use machine learning models to capture the sentiment in trusted voice recordings, video and scholarly articles from respected institutions in order to build a picture of the risk factors and their future uncertainty. From their collective scientific-based analyses/opinions, and because of the uncertainty in their models, we obtain a distribution for each and every risk factor. These distributions embody the future uncertainty based on the best science today. And as scientific evidence is gathered and changes, they change.”
Using this uncertainty estimate of the future, Dembo generates what he defines as a “spanning set” of scenarios, one that almost always contains the best opportunities and “black swans.” Naturally, this is a dynamic process because our knowledge about the world changes as climate models improve and as science discovers more each day.
“I believe Dembo’s method could advance the confidence in the likelihood judgment through crowdsourcing (essentially) and consolidating the best available inputs from multiple formats for a particular risk (for example, pace and intensity of sea level rise in Long Island),” Martin says. “There are quite a lot of sophisticated models on the physical risk side, so maybe the method can bring in more of the qualitative context or validation (for example, data from a politician’s interview) that might not be currently used. I could definitely see his method being valuable with transition risks such as changing customer behavior where it is unlikely to be able to source insights from a single data source to get a clear picture.”
However, Martin notes, “While some companies might find more confidence in such a model, others will recognize that so many data points can require a lot more assumptions, weighting factors, less distinct results, etc. …Sounds great on paper, but to a potential customer it would raise a lot of flags in terms of the value of outputs specific to their investment decisions versus the potentially enormous investments to use the method.”
With the growing internal and external pressure on organizations to undertake climate-related scenario analysis, Dembo should have opportunities to prove his system’s worth. He believes the sooner we start using these techniques, the better will be our ability to manage climate uncertainty. And he’s not alone.
David Dodge, an economist and former governor of the Bank of Canada, says climate change poses a risk to individual institutions and to the financial system as a whole. “It’s incredibly important that there be some way to assess that risk,” Dodge says, “and to make it comparable across different institutions, such that one can make useful comparisons.”
Dodge would like to see regulatory authorities and central banks assess the systemic risk caused by climate change, “just as they have tools to assess risk to the system as a whole from inadequate holdings of capital by banks,” he says. “The problem is corralling the varieties of information and assessing the risk out there. At the moment, we think that AI and machine learning should give us a way to do this. That, in essence, is what Riskthinking.ai is figuring out a way to do.”
