This award funds research that develops a new modeling framework for incorporating the discovery of new ideas (by firms, consumers, and other economic decision makers) into a class of models that are widely used in modern economics.
These models assume that individual actors use Bayesian inference to adjust their expectations over time. In this model of learning, over time the agents observe the world, compare that to their list of possible explanations, and eliminate possibilities that do not explain what they see while also adjusting their beliefs about the remaining possibilities. There is no idea here of discovering entirely new possibilities that were completely unknown at the start. The PI will generalize the Bayesian method to allow for this kind of discovery in a logically consistant way. He will determine the class of dynamic models that can be analyzed in the proposed new framework.
The project has transformative potential for macroeconomic theory, because the results may yield a useful new method for thinking about how new scientific and technological discoveries affect macroeconomic behavior.
Broader impacts include the potential to improve our ability to consider the role of scientific discovery on economic outcomes.
The work has generated several models of learning. The most closely related to the theme of the proposal was the paper published in the 2013 JPE and coauthored with Szentes. The paper deals with venture-capital outcomes most of which are failures. Like scientific discovery, VC activity deals very much with the unforeseen in a sense that gambling at the casino does not. The high-tech projects that much of venture deals with are not the type that merely implements a given blueprint such as would, say, the construction of a bridge or the paving of a highway. With high-tech projects one cannot enumerate the possible outcomes, simply because one is not aware of what they can generate. Because of this, the outcome is highly uncertain and as the investment proceeds (always in stages) VC terminates most of the projects in his portfolio because it becomes clear that the probability that the project will pay off is too low. One could not anticipate, before investment begins, what the result would be and what the precise nature of the innovations will be -- it is unforeseen. What does matter, however, is that the investor -- in this case the VC -- can have an idea about the monetary payoff that the discovery will generate. This is enough to provide him with the right incentives. The paper shows that even though the VC and the firm's founder cannot forecast the outcome, having the correct distribution of the monetary payoffs is enough to produce a socially efficient level of investment in venture capital. I extended this line to the macroeconomy in the forthcoming 2014 JPE paper with Rousseau which deals with private equity as an asset. Private equity includes buyout funds as well as venture funds. Because it focused on aggregate issues, the paper did not delve into information theory. My work on misallocation and growth (published in the AER in April 2014) deals with information about the quality of workers and firms, and with the effects on growth and development that would result if the quality of the information were better. The effects are not large on a yearly basis, but they cumulate over time precisely because they are growth effects, not level effects. Currently I am working on how learning can generate a so-called Kuznets curve in a group of decision makers each of whom wants to solve the same problem. A Kiznets curve is a pattern of inequality within a group -- initially inequality rises, and subsequently it falls. A good example of how learning can produce such a phenomenon is the ability to read. At a young age, no one can read and children are, in this sense, equal. And at a distant future date, every adult can read, and so people are again equal. But in the intervening periods, some are luckier than others in managing to learn, and a temporary inequality develops. Of course societies are trying to solve more complex problems, but some have pulled ahead of others, yet we expect that the laggards will catch up eventually.
