on Facebook
on Twitter
on YouTube
Spotlight Interview with Robert Lempert
 | Robert Lempert is an expert in science and technology policy, with a special focus in climate change, energy, and the environment. An internationally known scholar in the field of decisionmaking under conditions of deep uncertainty, he is a Fellow of the American Physical Society, a member of the National Academy of Science's Climate Research Committee, and a member of the Council on Foreign Relations. He is now completing a study on the Federal role in terrorism insurance and is leading a large, multiyear effort funded by the National Science Foundation on improving decisionmaking under uncertainty. |
How Do You Make Policies and Decisions When the Future Is Deeply Uncertain?
When we can capture future uncertainties fairly well, we have some effective tools to help in making decisions or policies.
That's true. For years, policy analysts at places like RAND have relied on decision analytic techniques, ranging from decision trees to Monte Carlo analysis, to help policymakers make better decisions. As commonly used, these approaches are extraordinarily effective when uncertainty is well-characterized. When you can assign probabilities to each uncertainty, those probabilities can be used to identify a best (or optimal) decision or policy.
But what happens when the future is deeply uncertain?
When uncertainty is particularly large—large enough that we cannot characterize it well—such traditional decision analytic approaches are error-prone or just plain difficult to use. For example, uncertainties for which there are either no data or poor data are often left out of the analysis entirely, which means decisionmakers can severely underestimate problems. And if policymakers can't agree on probabilities for some uncertainties, you can end up with gridlock within a business or organization. Finally, some vital information about potential opportunities or risks simply does not fit well within the traditional approach; as a result, it gets left out, which can leave decision- or policymakers vulnerable to unpleasant surprises down the road.
You rely on a different approach in dealing with decisionmaking under deep uncertainty.
The traditional approach relies on a "predict-then-act" paradigm. Our basic idea is to free ourselves from that paradigm—from the need to precisely predict the future before acting—by using computer simulation to help identify and assess strategies that work over a wide range of plausible futures. Instead of eliminating uncertainty, we highlight it and then find ways to manage it.
You call the approach Robust Decision Making, or RDM. How does it work?
RDM exploits the recent trends in computer simulation to create hundreds to millions of plausible futures and then relies on search algorithms, interactive visualization, and statistical analyses to help users identify "robust" strategies that perform well—regardless of whether the most uncertain assumptions turn out as expected—and then characterize the few deep uncertainties most important in choosing among strategies. Often, a robust strategy will sacrifice only a small amount of performance in the best-estimate case while significantly improving a business's or organization's performance if the future takes a turn toward the unexpected.
You've used RDM in a number of situations so far. Can you talk a little about some of them?
One of the more recent uses was in trying to help policymakers assess policy options for renewing the Terrorism Risk Insurance Act, or TRIA, which was established after the 9/11 attacks. TRIA is set to expire at the end of 2007, unless Congress takes further action. But deciding what to do is very hard, precisely because this type of decision is fraught with deep uncertainties.
Such as?
For example, there are uncertainties about the frequency and types of terrorist attack—conventional, or nuclear, biological, chemical, or radiological (NBCR). And there are uncertainties about the rate at which businesses would "take up" insurance coverage for policy losses under difference government interventions in the terrorism market. And there is also uncertainty about whether and how much the government will compensate businesses that fail to purchase terrorism insurance.
How did you apply RDM in this situation?
We used the RDM approach to examine the performance of three government interventions: leaving TRIA as is, letting TRIA expire (the equivalent of no government intervention), and extending TRIA to require insurers to offer coverage for both conventional and NBCR attacks. We looked at how these alternative interventions perform over a wide range of conventional and NBCR terrorist attacks. The RDM approach allowed us to consider important but deeply uncertain factors, such as the amount of compensation the government might decide to provide to uninsured victims in the aftermath of a terrorist attack, that have been ignored in most other analyses.
How did you measure performance?
We considered a series of outcome measures that cut across stakeholders, like the cost to taxpayers, the fraction of losses for which victims receive no compensation from insurers or the government, the extent of the insurance industry's role in paying losses from any terrorist attack, and the fraction of the insurance industry's net worth used.
So is TRIA robust?
It is, at least for conventional terrorist attacks. We find that for such attacks, extending TRIA is a more robust policy than letting the program expire. It reduces the fraction of uncompensated losses and over a wide range of futures, increases the insurance industry role, and decreases costs to taxpayers. TRIA achieves these benefits by reducing the industry role for the very largest attacks, those greater than $40 billion in losses, which would increase taxpayer costs but only in those very large attacks. To put the size of attack we're talking about in perspective, the 9/11 attacks on the World Trade Center caused about $23 billion in losses. Experts believe that such large attacks are much less likely than smaller ones and that, as a result, the taxpayer savings in the small attacks outweigh the taxpayer costs in the large attacks. Furthermore, TRIA is sufficiently well structured so that even if those expert estimates are wrong by a wide margin (and large attacks are more likely than the experts believe), we still find that on average that TRIA actually saves the taxpayers money.
What about NBCR attacks?
Our analysis shows that the current TRIA program provides far more limited benefits in the case of NBCR attacks. Legislation recently passed by the U.S. House of Representatives would extend TRIA to require that insurers offer coverage for such attacks. We analyzed a policy similar to what the House proposed and found it more robust than the current TRIA. Compared to TRIA, it reduces the fraction of uncompensated losses and, over many futures, increases the insurance industry role. We find that when properly designed, a TRIA with NBCR program can on average save taxpayers even more money than the current TRIA. It works this way because for small NBCR attacks, there would be more private insurance available, which would reduce the taxpayer costs for compensation of uninsured losses. As was the case in the conventional example, the experts believe that the small attacks are more likely than the large ones, where a government role in TRIA would be required.
In what other areas has RDM proven useful?
RDM can prove useful in any areas where uncertainties run deep, such as environmental issues. For instance, we are currently using RDM to help water agencies in Southern California adjust their long-range investment plans to better address the potential impacts of climate change. While the best scientific projections remain deeply uncertain, these agencies have to confront the possibility that over the next decades the Southland may see more and more lengthy droughts. In other recent work, we have used RDM to help the state of Louisiana and the U.S. Army Corps of Engineers develop and assess long-range plans for protecting New Orleans and the Gulf Coast against hurricanes, whose frequency and magnitude may also change in coming decades.
Is RDM mostly useful for the type of decisions faced by government agencies?
No, RDM has also been used in a variety of private-sector applications. For instance, we've conducted a series for projects for automobile manufacturers. We helped one firm develop technology investment plans in the face of substantial technological, market, and regulatory uncertainty. We helped another firm identify a plan for launching a major new product line that would do the best job of capturing upside potential while minimizing downside risks. We've also explored applications in other areas, such finance and insurance.
So RDM can help us predict a better future?
RDM isn't about better predictions. It's about helping us better manage an uncertain future. The approach is applicable to a wide range of other challenges, from bringing new products to market, managing the nation's entitlement programs, even defeating terrorism. Science and technology cannot change the future's intrinsic unpredictability, but it can help answer a fundamentally different question: Which actions today can best usher in desirable outcomes regardless of what future we face? When the future is most ill-defined and unpredictable, new computer tools may still help policymakers take actions today that can positively shape our longer-term future.
Top