Arden Rowell was the first Bigelow Fellow of the summer to present a work in progress with her article, "The Cost of Time.” In a host of risk-regulation contexts, regulators look at how much money people are willing to pay to reduce certain risks. Unfortunately, regulators haven’t been paying attention to the question of when those risks occur—or when participants in the studies think they will occur. As a result, the numbers regulators use are what Rowell calls “time-indeterminate”—they don’t have any particular timing attached to them.
Rowell argued that time-indeterminacy is a problem wherever we care about preferences, but it’s particularly a problem when we use money to measure things. That’s because money has a time-value—people would rather have a dollar today than a dollar a year from now. So asking how much people are willing to pay for a good (like a risk reduction), without pinning down when they think they’ll get the good, doesn’t actually tell regulators or anyone else very much that’s useful.
Rowell argues that this creates a particular problem for regulatory cost-benefit analysis. To see why, suppose you were asked how much you'd be willing to pay to ameliorate a 1 in 100,000 risk of dying of cancer. You might respond: $80. Regulators would take this to mean that they should pay $80 x 100,000 (or $8 million) to prevent the same risk from accruing to 100,000 people. If the risk won’t happen until 20 years from now, however, regulators wouldn’t spend $8 million today to prevent it—they would adjust for the fact that money is worth more now than later. Applying a standard discount rate of 7% would imply they should spend only $2,000,000 now to avert a cancer death 20 years down the line.
Herein lies the problem. Rowell points out that when the question asked to survey participants is time-indeterminate, we have no way of knowing if they themselves are discounting the risk already. That is, they think "cancer -- that likely won't kill me for another 20 years,” and give their answer accordingly (whereas if we asked them about a risk that seems more likely to happen immediately -- being hit by a bus, say -- their answer would not include a mental discount).
If this is right (and there is behavioral evidence indicating that it is), then regulators are undervaluing both immediate and future risk reductions. Immediate risk reductions are undervalued because the numbers regulators use actually incorporate a mental discount rate. And in valuing future risk reductions, regulators are actually discounting twice: once by incorporating the mental discount of the survey respondent, and then again by applying the standard discount rate. If we assume that they are discounting at the same rate (7%) outwards to 20 years in the future, then the amount worth spending to stop a cancer death now is $31 million , and to stop a death in 20 years, $8 million -- both drastically larger than the current method would advise. And as time horizons get longer, the time-determinate numbers diverge even more sharply from the indeterminate numbers that regulators use now.
This is a particularly big problem for cost-benefit analysis, Rowell argues, because there is no equivalent which would cause undervaluation of costs--there is no situation where one would retroactively incorporate past expenditures leading to "double compounding." That means that time-indeterminacy creates a systematic bias against the benefit side of the cost-benefit equation—towards underregulation. The solution Rowell offered is for regulators to start paying attention to the timing of preferences, and for regulators to use more time-determinate studies.
A lively discussion ensued, focused mostly on the broader implications of the paper. One professor suggested that there are other effects in cost-benefit analysis that push numbers the other way, so that there might still be too much regulation in the world; another countered that, if anything, Rowell was “underselling” the impact of the paper on both cost-benefit analysis and the valuation of life. Others wondered about the philosophical underpinnings of the willingness-to-pay paradigm, and whether they could be compatible with an ethical system that allots people certain fundamental entitlements. Finally, it was suggested that empirical studies of the magnitude of time-indeterminacy compared to other biases could provide useful contrast material.
This is great work, but why resort to handwavy empirical psychology? Just start measuring the frequency distribution of buying and selling. Use internal rates of return instead of market share to gauge who's cheating.
The blind spot for neoclassical theory was for rapid dynamics. But to be useful any alternative theory must also apply to large populations over long times. We need to see the frequency spectrum of activity to understand changes in time. I'd love to see such frequency spectra explained in terms of neural networks and their idiosyncrasies, but lots of those quirks wash out at scale. That's why neoclassical theory works at all.
Economics has lost its way. But it won't find it again in any laboratory.
Posted by: Michael F. Martin | July 23, 2009 at 02:47 PM