Description:
Given the unobserved nature of expectations, this paper employs latent variable analysis to examine three financial instability models and assess their out-of-sample forecasting accuracy. We compare a benchmark linear random walk model, which implies exogenous instability phenomena, with a linear state-space model and a nonlinear Markov regime-switching model, both of which postulate endogenous fluctuations phenomena due to heterogeneous behavioral heuristics. Using the S&P 500 dataset from 1990 to 2019, results confirm complex endogenous dynamics and suggest that the inclusion of behavioral nonlinearities improves the model’s predictability both in the short, medium, and long run.