In this project, we examined the correspondence between individual participant behaviour and financial markets in three laboratory experiments. In our first experiment, we explored whether market features would occur in a domain quite different from that of markets: a temporal estimation task in which we asked participants to reproduce the amount of time a circle appeared on a screen by holding down the space bar. Participants performed hundreds of trials of this task, and while not informed of this, the temporal target varied from trial to trial as a continuous random walk. We analysed the series of temporal estimates participants produced, and very surprisingly found that the temporal estimates themselves display some of the signature behaviours of financial systems: heavy tails and volatility clustering. This occurred despite any interaction between agents or any hint that this task was related to financial markets.

​

Experiment 2 mirrored the design of Experiment 1 but asked participants to estimate the next value in a series of numerical prices. This maintained much of the design of the previous experiment, including the underlying process of the target series and the trial-by-trial estimation procedure, but had a more plausible connection to financial markets. The results mirrored those of Experiment 1, showing heavy tails and volatility clustering, providing assurance that this was not simply a superficial correspondence.

​

Experiment 3 extended this design to examine price estimates in small experimental markets: groups of participants each made separate predictions of the price of a risky fictional asset based on a common history, with the actual price of that asset then being the average of all participants’ estimates in that period. This was based on the ‘learning-to-forecast’ experiments of Hommes and colleagues. Again, we found that heavy tails and volatility clustering occurred, showing that aggregating across participants did not remove these effects.

To explain these results, we developed a computational model based on the idea of sampling: that instead of reporting the expected value of their posterior beliefs, participants approximate this by sampling from their posterior beliefs. Crucially, instead of generating samples independently on each trial, sophisticated sampling algorithms generate autocorrelated samples. We showed that this model produced the same pattern of results we observed with human participants.

​

Overall, we examined the correspondence between the fluctuations of cognitive beliefs and financial systems across several tasks. We find that both these empirical data and a model of mental sampling able to reproduces many of the signature properties of actual financial markets, potentially indicating that such behaviours originate in part in the methods by which beliefs are produced by individual agents. While there is clearly need for more study, these results suggest the importance of individual expectations even within much larger systems: markets are ultimately built from individuals, and so are shaped by the beliefs of their members. We therefore hope that this study can act as a foundation for further examinations of macroeconomic behaviour through the lens of individual psychology, as well as the contrast of this behaviour with other complex systems.