The new year always begins with optimism and expectations for a year better than the last. For the large banks piloting the Federal Reserve's first climate scenario, 2023 may be one of futility, and unnecessary burning of resources better used to manage another tumultuous year in the markets.
Modeling the physics of climate change has improved significantly in the past decade. However, the socioeconomic models underpinning bank climate scenarios at best provide regulators with a false sense of security that climate scenario results are credible and at worst may lead to suboptimal policy and investment decisions over time. We present a case for a robust simulation-based approach that regulators and large banks should adopt that would significantly improve results.
The
That flaw is the climate damage function, the dirty little secret of climate scenario modeling.
Imagine a situation where, with all the sophisticated engineering and physics required to send the SpaceX Dragon capsule to the International Space Station, it still relied on a handheld compass for its guidance system. That is comparable to the current state of climate scenario analysis.
For bank risk managers, the scientific models generating physical climate outcomes must translate those outputs into meaningful economic results. Integrated Assessment Models (IAMs) are those translators. They are core to any regulatory climate scenario analysis, including the
A climate damage function is an equation that uses physical variables such as temperature change and translates them into economic outputs such as the change in GDP. These damage functions are well established in economics.
However, unlike the scientific-based climate models, these damage functions have yet to progress much, and their simplicity significantly undercuts the scientific advances in modeling climate change. Current damage functions use naïve relationships depending only on temperature change and show smooth growth with no probability of sudden changes in the climate. The simplicity of these functions poses several issues that render any climate scenario of limited value.
First, these functions are coarse representations of more complex relationships affecting economic activity. Second, the validation of these models is nonexistent and would never pass muster under a regulatory-mandated bank
Finally, the empirical linkages between climate outcomes expressed in temperature changes, for instance, are weak at best. The combination of poorly specified IAMs with single path climate scenarios ultimately becomes a "feel good" exercise for the regulators rather than one that can be useful in guiding policy and hard money decisions.
We contend that current deterministic damage functions are not sufficiently robust and that regulators adopt a value-at-risk (VaR) or a climate VaR approach, based on the methodology summarized here.
For decades banks exploring the effect of changes in the market value of equity in their trading books have used VaR models extensively for those applications.
While such analysis can provide a great deal of insight into what the average or expected value of a portfolio might be over thousands of different economic outcomes, of more interest to risk managers is the tail of that distribution.
This type of analysis can answer questions such as what is the 95th percentile worst market value in my portfolio from climate change? For many asset types, we can compare portfolio values over different time horizons under a multitude of climate scenarios. We can then compare differences in baseline (current state) market value to any climate scenario at any percentile of the value distribution. Since these results are based on thousands of climate scenarios it makes the results more robust and, importantly, sharply reduces dependence on the accuracy of a single climate scenario.
One of the missing links from climate scenario analysis is the inclusion of "tipping point" events. These events include the possibility of the collapse of a significant part of the West Antarctic Ice Sheet or a sudden release of massive amounts of carbon dioxide and methane from thawing permafrost across Arctic regions.
Such events have a low probability of occurring in the next decade. However, the impacts could be enormous even on short and medium time scales (five to 10 years). We include possible effects of tipping points in our approach by modifying damage functions with probability distributions calibrated with parameters from the best current estimates of tipping-point probabilities and impacts.
Replacing a single scenario with a simulation-based approach for climate scenario analysis would improve the current state of climate risk management. This approach can leverage readily available risk management and capital requirements methodologies and consequently eliminate reliance on poorly constructed damage functions.
The Federal Reserve is effectively fighting the last war by applying the same kind of scenario analysis used for assessing the effects for a major financial collapse over nine quarters to climate change that unfolds over decades. These are entirely different extreme events. Managing the risks of climate change is a difficult enough sell without a solid analytical foundation.
