How do people predict a random walk? Lessons for models of human cognition.

Area of Science:

• Cognitive science
• Behavioral economics
• Computational neuroscience

Background:

• Forecasting fluctuating values, like in random walks, is common.
• Humans deviate from rational forecasting (repeating the last value), showing excessive volatility.
• These deviations may reveal statistical signatures of cognition.

Purpose of the Study:

• To explore cognitive models of forecasting using human deviations from random walks as criteria.
• To identify which models best account for observed human forecasting behaviors.

Main Methods:

• Compared human data from two experiments against Bayesian, error-based learning, autoregressive, and sampling models.
• Analyzed systematic deviations from random walk properties (e.g., volatility).

Main Results:

• Sampling models provided the best fit for both aggregate and individual human forecasting data.
• Human forecasting variability is linked to stochastic prediction systems.

Conclusions:

• Cognitive sampling mechanisms, characterized by computational noise, explain human forecasting deviations.
• Variability in predictions stems primarily from internal decision-making noise, not output stage noise.