On 23 May 2017, Émile Servan-Schreiber announced that economist Scott Sumner and the Mercatus Center at George Mason University were partnering with Hypermind to launch a new market forecasting US nominal GDP. The day before, Sumner had explained the project on EconLog, the economics blog where he writes. The market asked traders to forecast how much nominal GDP would grow between the first quarter of 2017 and the first quarter of 2018.
What nominal GDP is
Gross domestic product (GDP) measures the value of everything an economy produces in a given period. “Real” GDP strips out price changes to show how much more is actually being produced. “Nominal” GDP, or NGDP, does not: it is measured in current dollars, so it rises with both output and inflation. The Mercatus Center describes it as essentially the nation’s total income.
That matters for a long-running debate in monetary policy. The US Federal Reserve practises a form of inflation targeting: it aims to keep inflation close to 2% by moving short-term interest rates. Sumner and his Mercatus colleague David Beckworth argue for a different goal, called NGDP level targeting. Under it, the central bank would aim to keep total spending in the economy on a steady growth path. Because the target is a level rather than a rate, a shortfall in one year would be made up later.
Why a market
Sumner has long argued that policymakers could rely on a market to tell them what the economy expects. If traders can buy and sell contracts whose value depends on future NGDP growth, the market price gives a running estimate of where total spending is heading. A central bank could then see whether its policy was on track.
He was careful about what the 2017 market was and was not. For actual policy, he proposes “guardrails” at 3% and 5% growth. Under that idea, the central bank would promise to buy unlimited NGDP futures at 3% and sell unlimited futures at 5%, which would hold market expectations within that range. Separately, he wrote that he favours a permanent NGDP prediction market that produces point estimates of expected growth for policy research.
This is not a policy market.
Scott Sumner, EconLog, 22 May 2017
How the Hypermind market worked
Hypermind is a prediction market in which participants trade contracts on future events without staking their own money. The price of a contract can be read as the crowd’s estimate of the probability of an outcome. The best forecasters share prize money in proportion to their performance.
- An earlier round. Sumner wrote that a 2015 version had included an annual NGDP market and four quarterly markets, and that only the annual one had proved macroeconomically significant.
- The 2017 launch. The new market started with $5,000 in prize money for the annual contract, which Sumner hoped to increase through fundraising.
- September 2017. On 19 September 2017, the Mercatus Center announced it was sponsoring the market, which it described as run with Hypermind, a UK-based prediction market. It had two contracts: NGDP growth from the first quarter of 2017 to the first quarter of 2018, and from the first quarter of 2018 to the first quarter of 2019. The total prize was $70,000, which Servan-Schreiber summed up the next day as “seven hundred $100 bills” lying on the sidewalk.
- 2020. In December 2019, Servan-Schreiber promoted a $30,000 Hypermind contest, again sponsored by Mercatus, on quarterly and yearly US NGDP growth throughout 2020.
What the experiment was meant to show
Mercatus said it would use the data to study how NGDP expectations respond to economic and policy events. The longer-term question was whether such a market could track the path of NGDP growth well enough for the Federal Reserve to use, and help it move towards NGDP level targeting. In March 2018, Servan-Schreiber shared a Mercatus page that asked whether the market could set better monetary policy than the Fed.
Those are questions about the future, and the market produced forecasts, not certainties. A market price of, say, 4% expected growth is the crowd’s best estimate given what it knows, and it moves as new data arrive. We have not found a published evaluation of how accurate the 2017–2019 contracts were, so this article does not claim that they beat other forecasts.
