In January 1974, I was asked to create a model of the risks of nuclear power plant accidents for the American Electric Utilities. It seemed such a far-fetched idea that any such accident would occur that no insurer had looked at it in painstaking detail. It was a dauntingly complex challenge …that any such accident would occur that no insurer had looked at it carefully.
A month before, I had been building Monte Carlo simulation models for “widgets” at Wharton. That helped me become the first Wharton MBA hired by the Marsh New York headquarters consulting group, called Unit X. My first assignment was this project, modeling the risks at nuclear power plants.
The first commercial nuclear power plant in the United States, Shippingport Atomic Power Station, had opened in 1958. The industry had grown safely without any publicly disruptive incidents. There seemed to be no downside.
Nevertheless, insurance was needed – no matter how remote the possibility of failure was. I was given authority to work on behalf of the electric utility industry for the placement of their risk in support of our Marsh brokers at Lloyds, the ultimate insurer.
So how do you build something that has never been built? You start with what data can be obtained.
I contacted the Atomic Energy Commission for data on nuclear power plant accidents. They referred me to the Machine-Readable National Archives (MRNA). When I spoke to the director in charge of maintaining this data, he responded enthusiastically: “You know, we’ve been keeping these records on thousands of nuclear power plant accidents for 15 years, and you’re the first person who has every requested the data!” The beauty of government regulators is that they keep track of things.
My effort may have been the first analysis in the insurance industry to work with BIG DATA for an entire industry. To model the risk, we started with data on relatively minor accidents, with thousands of incidents up to a few million dollars in value.
There were two manufacturers of nuclear power plants, GE and Westinghouse. Each had two models, Pressurized Water (PW) and Boiling Water (BW) reactors. So there were four sets of plans, 4 chief engineers, 2 each at GE and Westinghouse.
I was assigned a large conference room at Marsh’s New York offices where I was able to spread out each of the 4 sets of plans. With each of the chief engineers, I would go through the plans to understand the process and construction. Each key system component had a number: for each item I would ask, What’s the chance of it failing (frequency) and how bad would it be (severity). Both of which would then be recorded on the corresponding decision trees being built for each of the four reactor models.
After working through the plans for single system failure, we then worked the possibilities of two, three and finally four concurrent system failures – which ironically reflected the Three Mile Island event which would occur over 5 ½ years after our work. Cataclysmic failures are often events that trigger other events and a domino effect of failures.
In this process I had four sets of probability models, one for each reactor. These were factored into the model on a weighted average basis according to the respective numbers of plants in operation.
The MRNA data was invaluable in providing empirically based frequency and relatively low severity to the model.
The model was completed in about 4-5 weeks. Models try to simulate reality, but they need to be run and rerun to find any flaws in the logic. I then ran it and re-ran it. It was not as quick and easy as it is today, we did not have ready access to personal computers. I was lucky, though. Marsh had its own CDC 6600 Mainframe Computer. While it was kind of busy during the day, waiting for a keypunch manchine (to write lines of code that were punched into punch cards) or wait to have your job run, I was able to have the machine virtually to myself overnight. So I would be sure to get to the computer room by 10:30 or 11 pm, and I could happily have the machine virtually all to myself until around 7 the next morning. It made running my model and doing the project more feasible in the limited time we had.
Once the model was tuned, we were able to supply our brokers in London with estimates of the probabilities and ranges of expected losses within different successive reinsurance layers. They worked with Lloyds to fill the slips for the placement of the risk.
The good news is that the mathematics worked! We were able to understand the realistic insurance cost for potential economic impact of failures.
Five years later when the catastrophic accident at Three Mile Island occurred in 1979, there was no news story about a financial crisis in the insurance market. The electricity generation industry had been adequately prepared. This was a point in time when the industry could have collapsed if they had not been prepared.
Now nuclear power provides 20% of the US national power needs and we are the leading nation in the world using nuclear power technology. Nuclear energy and renewables are far, far safer than fossil fuels with regards to human health, safety, and our carbon footprint. It would have been a sad day for the country if the nuclear power industry had become economically unsustainable due to the Three Mile Island incident. That collapse was prevented by adequate insurance coverage.
Since then, I have developed many similar models to assess risk in other industries. The first step on any journey is the hardest. The nuclear industry was a start, but the thinking process gave us confidence to look at large data sets in many other fields. We are still doing it today.
I am proud of my part in helping the country maintain this energy option, particularly these days when fossil fuel options are becoming less and less attractive due to their impact on the environment. I also feel fortunate to have been at the right place at the right time, with the right skill set of skills – thanks to my professors at Penn and Wharton.
0 comments on “Modeling the Risks of Nuclear Power before Three Mile Island”