Using Prediction Markets in Macroeconomics

or: On the Inconsistency of Optimal Plans

An unanticipated tax on consumption is the same thing as an unanticipated tax on wealth. After all, the value of wealth is what you can buy with it. It hardly matters whether we reduce wealth directly, or simply make it so that it can buy less than before. Such a tax has the great advantage that, if it is unanticipated, it is perfectly efficient. People embarked on a plan to maximize their utility over their lifetime, taking into account that they could only work during the beginning of their life, and would need to spend their savings at the end. Since people don’t work after they retire, a tax on either wealth or consumption can’t affect the amount of labor they supply, which is zero.

If the tax is anticipated, however, then it will distort consumer behavior. People will save less, work less, and spend more now. With less saved, the real interest rate will go up, and real output in the long run will be lower. The degree of distortion will be proportionate to how likely people anticipate the tax to be. If people are absolutely certain that the tax will be imposed, then not imposing it would cause all the distortions of the tax without any of the tax revenue. At lower levels, then there will be a smaller distortion; but so long as people do not believe that the chance of the tax being imposed is zero, then the mere possibility is distortionary.

If the government does not intend to impose a consumption tax, then it could increase economic output simply by credibly promising never to impose the tax. By leaving itself with more options, it actually makes everyone worse off. The intellectual precursor to this is Kydland and Prescott (1977) who are concerned mainly with monetary policy, though their analysis generalizes to any instance in which the decision maker makes the locally optimal decision. People might, for example, build in a flood zone, since they know that they will be bailed out if a flood comes. The best policy would be if the government credibly committed to a rule – that we will not bail anyone out – and no one built in the flood zone to begin with.

How large is this loss? We are constrained by the fact that we don’t know how much people anticipate a tax on wealth. People’s savings are determined by multiple unobserved factors. To answer this question, I have proposed the creation of a market on this to both Kalshi and Polymarket. We shall see if they accept it, but this segues into a potential use of prediction markets which I have not seen discussed – they allow us to observe otherwise unobservable parameters in a macroeconomic model.

We do this already through conventional securities markets. For example, the US government issues bonds which are inflation adjusted, and bonds that are not. If you want to know the expected rate of inflation, you can refer to the TIPS spread, or the difference between the two. This is not observable from just Treasury bonds alone, because we would be unable to separate out changes in the real rate of interest from changes in preferences, from changes in the expected course of inflation. Note that this information is unusable by the government – if everyone knew that policy were dependent upon prices, then prices would become indeterminate – but it is useful for researchers and the public.

Prediction markets can be the equivalent of randomized controlled trials for macroeconomic puzzles. The ideal use of RCTs is to precisely identify the parameters you want to find. This allows you to go beyond just the effect of a policy, and evaluate similar policies in other places which will occur through the same mechanisms. Macroeconomists have not reacted enough to the possibility of just asking the questions you want answered! I don’t know any papers which just straightforwardly take parameters from relevant prediction markets, although my ignorance of macroeconomics should never be taken, strictly speaking, as evidence of absence.

The lack of uptake is likely due to the fact that wide-ranging prediction markets are a very recent development. Early work, like Wolfers and Zitzewitz (2006), is concerned largely with presidential election prediction markets largely because those were the only ones to exist at the time. The idea that we could get a prediction market on any subject under the sun, and have thousands of people trading on it, would be considered wildly optimistic.

I would like to hear your best criticisms of using prediction markets, before I stake a research agenda on it. It could be the case, for example, that prediction markets will be used by people who have views which differ from that of the public at large, and so long as the events we are testing for are idiosyncratic, the market cannot converge to rationality. The people who use prediction markets could be hedging in correlated ways, which will distort the probability away from the true probability. The probability of an event happening could be mixed up with people’s uncertainty about the market as a whole, or by changes in the marginal utility of income.

If you think that it is still a useful source of parameters, please do reach out. I would love to work with you! If you have a contact within Kalshi or Polymarket, please also reach out. Many of these markets are fairly specific, and are not guaranteed to be created. (With apologies to Manifold, if it isn’t for real money, I am not going to really believe it – or at the very least, the referees won’t!) I think that this project can improve our understanding of the world substantially.