Christopher Sims, Information Theory in Economics

Christopher Sims, Harold H. Helm '20 Professor of Economics and Banking at Princeton University, discusses the application of information theory to economics.

[@noisepuppet]

But seriously, if our model of economic behavior assumes that each person is an actor or agent or decision node, controlling some finite amount of capital with which it seeks to perform optimally in markets, based on available information, i.e., we assume that in markets, we're observing aggregate behavior of discrete rational self-interested economic agents; and furthermore we take it as given that each node has finite signal processing capacity; and yet further we're forced as always to account for Shannon "entropy" or uncertainty; and we recognize that nodes or agents act in competition with other nodes, which is to say they seek to optimize their performance specifically with respect to other agents, in a relativistic way; and lastly, we acknowledge that each node is able to produce information and broadcast its own signals and tends to do so in the most optimally self-interested way; and these assumptions taken together comprise (roughly) the basis of our model; then we find that misinformation must be accounted for. I don't know the theories very well, but if any of them formalize this variable, then I've missed it, and I think most of them omit it, and the omission is fatal to any model's ability to predict and explain market behavior in more or less the terms given here. Misinformation means not only PR, propaganda, advertising, self-promotion, and so on, on which activities our rational self-interested economic agents spend stupendous resources in the real world, but also actions that are taken as economic signals by other nodes in the system, when such actions are not optimally self-interested except in the sense that they signal false information that the agent wants competing nodes to receive and process. We can think of many examples of this sort of behavior, a sort of bluffing, buying when one should sell or vice-versa, upsizing when downsizing is the rational move, and so on, in order to keep up appearances or signal confidence or a strong market position or whatever. And given the low cost of producing signals, and the high cost to other nodes of processing it, and their limited capacity to do so, and the potential advantage of the resulting uncertainty about oneself, each agent will be motivated to generate a lot of worthless signal or noise, which is not necessarily false but nonetheless interferes with competing agents' ability to make rational self-interested decisions vis-a-vis itself. We see this in everything from resume padding to feel-good official statements to, well, the vast majority of advertising campaigns, which are broadcast everywhere possible at all times. Much of this signal is empty, perhaps corresponding roughly to noise in a Shannon type of channel. Worse, much of the remainder is false or misleading, such that accurate transmission of the signal, with its successful reception, results in a reduction of the receiver's capacity to act in a rational, self-interested way, and furthermore, this is precisely the intent of the sender. We could guess what will happen. This misinformation immediately comes to dominate the information network to the maximum extent that each node can make it so, in its own self-perceived interest. And I propose that this is what we see in fact, in the real world, with the commensurate volatility and unpredictability of markets and seemingly irrational or disengaged behavior of individual economic actors. This manipulation and jamming of signal in a network is well understood and accounted for in counterintelligence models. So why is the concept of misinformation missing from economic models that purport to place information at the crux of market behavior? I suspect that the conflict aspect of such theories is too informative to be much loved by most economists, and they find it in their rational economic self-interest to promulgate theories that omit it. That's my guess, but I'm no expert.

[@Ntifragility]

!