Accurate affective forecasting (i.e., predicting future emotion) is important for daily and

major life decisions. While people are generally good at predicting the valence of their feelings,

they fail to make accurate estimations of their intensity. A recent finding suggests that statistical

models could be better than humans at predicting emotional states. The study by Takano and

Ehring (2024) found that the Kalman filter outperformed participants in predicting their own

emotions in hour-long forecasts. We will attempt to replicate this finding with newly collected

experience sampling (ESM) data. In addition, we will compare the prediction accuracy of

participants and the Kalman filter with a second statistical model, the multilevel Bayesian

autoregressive model. Lastly, we will examine whether and how the prediction accuracy changes

when we use interval-based forecasting instead of traditional point estimates.