We amend adaptive expectations (AE) to make it robust in the face of nonstationarities such as those emerging in self-referential forecasting systems. We borrow insights from robust control in engineering and propose that the learning rate α in adaptive expectations is to be modulated in a way to minimize surprise relative to a reference model. As reference, we suggest the Kalman filter model recently used in a study examining how professional forecasters predict economic outcomes. We show how this prescribes changing α in the direction of autocovariance of prediction errors. We refer to the resulting forecasting model as Robust Expectations Adaptation (REA). Ours contrasts with the traditional prescription in reinforcement learning, which is to change α in the direction of the change in the size of the prediction error, the Pearce-Hall model, recently imported into the economics literature. Using 40,000+ forecasts from experiments on self-referential economic markets, we discover that participants change α as in REA, but generally only if surprise is above the median experienced by the individual. The Pearce-Hall model almost never fits the data.

Optimisation uncertainty --- the difficulty of mapping a representation of the decision problem into an action --- is particularly consequential in strategic games. We show this using confidence elicited in a repeated, randomly rematched public goods game, closing each source of encoding uncertainty studied in the literature one at a time. Encoding uncertainty here has two inputs: noisy beliefs about others' contributions, and noisy access to one's own preference. Confidence does not respond to a treatment removing the first, nor does it rise with repetition as the second would predict, leaving optimisation uncertainty as what remains. We estimate ideal (full-confidence) and default (zero-confidence) decisions from choices and confidence reports. We find no evidence of compression in low-confidence decisions, unlike other settings, because defaults are no less responsive than ideals. The ideal, net of confusion by construction, evaluated at a group contributing nothing, is close to what the kindness-versus-confusion debate has sought to estimate; it is positive for a majority of participants, so contributions are not motivated by confusion alone.

Suppose that all asset market traders are proficient at reading the market. Would markets become more stable, resulting in lower volatility and fewer price bubbles? To answer this question, we test whether Theory of Mind (ToM) capabilities enhance expectation coordination and reduce expectation heterogeneity and price bubbles in learning-to-forecast experiments. We compare the price and expectation dynamics between markets composed of participants with either high or low ToM capabilities as measured by the eye gaze test. Despite an economically substantial difference between the two groups, we find no statistically significant differences in the measures of expectation coordination, price bubbles, market stability, and expectation heterogeneity.

A large body of literature concludes a negative association between ethnic diversity and pro-social behavior. Inspired by the works suggesting that the costly punishment would sustain the contribution level in public goods experiment, we compare the economic behavior of Mongolian- and Han-Chinese and investigate how ethnic diversity would affect contribution, punishment, and the marginal effect of punishment on contribution. We find that the association between ethnic diversity and pro-social behavior is not a simple negative relationship but rather depends on both cultural traits and ethnic fusion when we take punishment opportunity into consideration. Ethnic diversity may help promote contribution, alleviate the punishment level, and increase the efficiency of introducing a punishment mechanism in some circumstances.

This paper reviews the recent development and new findings of the literature on learning-to-forecast experiments (LtFEs). In general, the stylized finding in the typical LtFEs, namely the rapid convergence to the rational expectations equilibrium in negative feedback markets and persistent bubbles and crashes in positive feedback markets, is a robust result against several deviations from the baseline design (e.g., number of subjects in each market, price prediction versus quantity decision, short term versus long term predictions, predicting price or returns). Recent studies also find a high level of consistency between findings from forecasting data from the laboratory and the field, and forecasting accuracy crucially depends on the complexity of the task.