Negotiating the Wilderness of Bounded Rationality through Robust Policy

We show how the “wilderness of non-rationality" posed for the policymaker may be negotiated by designing a robust Taylor-type monetary rule across a RE NK model and competing behavioural alternatives. The latter consist of a model with “Euler learning" and a bounded rational one with myopia due to Gabaix (2020). For the former expectations of endogenous variables take the form of a general heuristic rule, encompassing simple adaptive expectations, that is supported by an experiment study. This gives four competing NK models, the benchmark one with rational expectations (model RE), Euler learning with a simple adaptive expectations heuristic rule (model EL-SAE), Euler learning with the general rule (model EL-GAE) and the Gabaix bounded rational model (model BR). In our novel forward-looking approach, policymakers weight models based on relative forecasting performance rather than Bayesian model averaging. Our main results are: first, three models completely dominate model EL-SAE with weights wRE = 0.4, wEL−GAE = 0.32 and wEL−BR = 0.28. Second, whereas Bayesian model averaging would design a welfareoptimized rule that hits the ZLB with a probability solely based on the Gabaix model, we find that our prediction pool using these weights choice has a significant impact on the robust optimized rule. Third, there are significant differences between the optimized rules for each model separately highlighting the need for seeking a robust rule. Fourth, we find that robust optimized rule found using optimal pooling weights is very close to the price level rule. This confirms good robustness properties of such a rule found in other studies. Finally to achieve a probability of hitting the ZLB constraint on the nominal interest rate of 5% per quarter, the robust optimal rule requires a target (steady-state) net inflation annual rate of between 3% and 4%.