Federal regulators released a report about what caused the San Onofre nuclear power plant in California to leak radioactive steam in January. It said excessive wear caused steam tubes at the plant to rupture – an event that “had never been seen before,” according to a federal official, and hadn’t been anticipated by the manufacturer’s computer model.
Faulty computer models aren’t confined to the world of nuclear power, of course. Just ask Jamie Dimon. But it’s the managers – not the models – that must shoulder the blame when things go wrong.
The report by the Nuclear Regulatory Commission found that the steam tubes at the San Onofre nuclear power plant were weakened by higher-than-expected vibration in the generators. The tubes did not perform according to computer models developed by their manufacturer, Mistubishi Heavy Industries. According to media coverage on the NRC report:
“Greg Werner, who headed the federal team, said a Mitsubishi computer analysis vastly misjudged how water and steam would flow in the reactors. Also, changes intended to improve manufacturing were never thoroughly reviewed in the context of the generator design, resulting in weaker support around bundles of tubes that contributed to vibration, he said.” (Emphasis added.)
This isn’t the first time that realty turned out to be different from what the computer models predicted. In announcing J.P.Morgan’s recent trading loss, CEO Jamie Dimon cited a change in the bank’s model used to calculate the value-at-risk (or VAR) for the Chief Investment Office (CIO), where the losses occurred:
“We are also amending a disclosure in the first quarter press release about CIO’s VAR, value at risk. We’d shown average VAR at 67. It will now be 129. In the first quarter we implemented a new VAR model, which we now deemed inadequate, and went back to the old one, which had been used for the prior several years, which we deemed to be more adequate. “
When a problem arises in a complex system like a nuclear power plant or a portfolio of credit-default swaps, it’s tempting to point to faulty computer models.
But it’s not really the models that fail, it’s the managers. They often put too much faith in the models and abandon their own judgment.
Successfully managing a big bank or a nuclear plant comes down to people and their responsibility to make informed decisions about risk. Good managers draw on data like the kind produced by computer models but rely on other inputs as well, like first-hand investigation, past experience and instinct. Models are not a substitute for thinking.
To his credit, Jamie Dimon didn’t limit blame to the faulty model. He was quick to cite management failures that produced a strategy that was “flawed, complex, poorly reviewed, poorly executed and poorly monitored.” He moved quickly to replace the executives responsible for the trading loss.
The right message for shareholders isn’t a promise to improve the models, but a commitment to sharpen management oversight.