Prediction Market Miscalibration Reveals Aggregate Investor Behavior in Financial Networks

In a prediction market (PM) investors trade contracts with payouts tied to real-world event outcomes, such as political races or sporting competitions. After many trades, contract prices are expected to converge to a value that accurately reflects the probability of an event outcome [1,2]. There are exceptions however; during the 2016 “Brexit’ referendum, PMs suggested a fait accompli for “Stay in the EU” despite polls suggesting a close contest [3,4]. Most research has focused on whether PMs correctly forecast the final (binary) outcome. But comparatively little attention has been paid to how well calibrated these markets are: do the implied probabilities of particular events represent the true odds over many realizations?

Here, we use ~3200 unique political PM contracts to see how investor behavior leads to price convergence. We use a novel metric to retrieve contract prices on days before all information is incorporated, asking if they accurately reflect the ground-truth likelihood of similar past events. We find that investors systematically overvalue contracts, an effect that is counterintuitively most pronounced in high-profile markets such as US presidential races. Our results highlight investor behavior, and should be taken in context with modeling price dynamics over financial networks [5].

References:

[1] J. Wolfers and E. Zitzewitz, Interpreting Prediction Market Prices as Probabilities, (2006). Tech. rep., National Bureau of Economic Research.

[2] J. E. Berg, F. D. Nelson, and T. A. Rietz, Prediction Market Accuracy in the Long Run, Int J Forecasting 24, 285 (2008).

[3] T. Auld and O. Linton, The Behaviour of Betting and Currency Markets on the Night of the EU Referendum, Int J Forecasting 35, 371 (2019).

[4] J. Fry and A. Brint, Bubbles, Blind-Spots and Brexit, Risks 5, 37 (2017).

[5] A. Mancini, A. Desiderio, R. D. Clemente, and G. Cimini, Self-Induced Consensus of Reddit Users to Characterise the GameStop Short Squeeze, Sci Rep-Uk 12, 13780 (2022).