In 1968, Harvard Business School professor Howard Raiffa published a book called Decision Analysis: Introductory Lectures on Choices Under Uncertainty. It was an attempt to translate Bayesian statistical thinking into terms understandable to laypeople, and it introduced the use of decision trees and numerical probability assessments to business and government decision-making.
One of the book’s early readers was a precocious high school student named Larry Summers, who was so taken with Raiffa’s probabilistic reasoning that he’s been using it ever since. According to The New Yorker’s Ryan Lizza, Summers’ penchant for assigning numerical probabilities during his time as director of President Obama’s National Economic Council led the President to tease him by saying things like “I think there’s a 72.7% chance that I’m going to wear black shoes in the morning.”
I asked Summers about his approach to one of the highest-stakes decisions made during his time in the White House, the choice of what to do about ailing auto giants General Motors and Chrysler. What follows is an edited version of our conversation.
When Obama came into office, what were the alternatives?
There were three broad categories of alternatives. Let nature take its course, bailout without bankruptcy, and managed bankruptcy. They each had strong arguments in their favor:
The consideration for nature taking its course was that the companies had been very badly managed for many years, and people needed to take responsibility for failure. Bankruptcy was a mechanism for cleansing the economy and providing for creative renewal. Government involvement would inevitably complicate and retard the process of reform.
The automobile companies, meanwhile, believed deeply that bankruptcy would be fatal because no one would want to buy a car from a company that was going bankrupt. They pointed to a very sharp decline in the used-car price of models that had been discontinued as evidence of that. They stressed that purchasing a car was very different from flying an airline — it was the difference between a date and getting married. They also stressed that not long ago that they had been quite profitable, and their major problem was the magnitude of the recession and the associated drop in demand. One was looking at the risk of automobile sales falling below 10 million vehicles in a year, and it was an industry that was normed for 15 million. It was an outside force like a hurricane that had affected them.
The third view was that you would not be able to bring about enough change without bankruptcy, and if bankruptcy was substantially managed, it could be credible that the companies would come through. Strategic options could be found that would recognize that the normal debtor-in-possession (DIP) financing in bankruptcy was not likely to be available in the context of the worst financial crisis in 50 years. That was the ultimate course that the president chose, and ultimately it was the course that both General Motors and Chrysler went through.
How systematic was that whole process? Were you drawing up decision trees and assessing the probabilities of different outcomes?
For me in advising the president, there was not the kind of decision tree that would be included in a classroom exercise. The problem was too complex and multidimensional to be reduced to that. But the main elements of the process that one learns in a class in decision analysis very much shaped the approach that was taken: You articulate each of the alternatives, then you attempt to figure out how you would optimize within each of the alternatives.
So I constantly put the question to my colleagues: Suppose government finance was not an option, what would be the best thing to do? Suppose we couldn’t put them through bankruptcy but we wanted as much rigor as possible, what would be the best thing to do?
Articulate alternatives, optimize each one of them, evaluate the results with the alternatives, assign probabilities in situations of uncertainty, make a judgment about expected values, calculate the response of other actors (assuming they’re going to do what’s best for them not what is best for you), and recognize your risk aversion — and that an important objective, particularly in a moment like 2009, was to protect against the worst possible alternatives.
Another element that comes from the systematic application of decision analysis was that on a number of occasions I would ask each person for their own opinion, or their own probability — and would always be careful to ask the most junior people in the room first for their opinions, so they were not simply following the lead of the more senior people in the room. On occasion I would ask people to write down their judgments on certain questions so that they were not overly influenced by the judgments of other people. It was not a literally structured decision analysis, but the approach I took and the way in which I presented the problem to the president and the way in which I asked the staff who were working full-time on this was very much shaped by my early-life exposures to decision analysis.
There was another very important aspect that I think also comes out of economic reasoning, but less out of classic decision analysis. That was engaging in calculations around calibration and tuning where there are tradeoffs. So, for example, how much money was the government prepared to provide as a DIP financier? If you provided more money, it was more costly to taxpayers and you were exposing the government to greater risk.
Fairly constantly in actual decision-making, it’s not so much choosing alternative A or choosing alternative B, but how you operate within the alternative you have chosen. And at least for me, it was useful to apply the economist’s principle of optimizing at the margin. That is, find the point where the marginal benefit of spending more money is just equal to the cost of spending more money, and choose that point.
These kinds of calibrations at the margin apply to money but also to many other dimensions. Enormous energy is spent in the White House — it was done with respect to the automotive bailout, but with respect to many other things as well — when the president is going to announce a decision. It is desirable to avoid leaks, so there’s a clear announcement. It’s also desirable to give advance notice to the many stakeholders. How do you trade those things off?
Overall do you think the GM and Chrysler managed bankruptcies were the right set of decisions? How do you tell?
That question is actually a part of my decision-making process. I always ask, when I am advocating a course of action or someone else is advocating a course of action to me: How will we know five, ten years from now whether we were right? What would convince us that we had screwed up?
It’s much more complicated than simply evaluating whether the outcome was good. When you play poker, if you draw to an inside straight, you’ve always done the wrong thing, but sometimes you’ll get a good outcome. If you do not draw to an inside straight, sometimes you will regret that, but it was always the right decision. So I evaluate decisions on the basis of whether aspects of the problem manifest themselves that had played no role in the deliberative process. If so, the decision making process was a clear failure. If events to which we assigned a very low probability happened, that makes the decision process somewhat suspect.
In the event, I’m proud of the role I was able to play supporting Team Auto and advising the president — because very little surfaced as an issue that we had not anticipated; because the events we judged very unlikely in fact did not happen; and because ultimately, our judgment that the companies could weather bankruptcy and that the situation was dire enough that it was important that the automobile industry be maintained, seemed to have been vindicated by both the subsequent seriousness of the broad economic situation and the way in which the companies have been able to return substantially reformed and trimmed to the private sector.