Working Papers (Click to download)
We use machine learning techniques common in artificial intelligence literature to forecast the inflation rate with disaggregated data. We found that machine learning techniques have better forecasting performance than any traditional methods including autoregressive models, random walk models, and dynamic factor models. This work can help policymakers to form more accurate forecasts about some aggregate economic variable when they have access to the disaggregated data.
This paper investigates the implication of introducing multiple finite-state Markov extrinsic sunspot processes in a general univariate forward-looking model. In this model, each agent either does not observe any sunspots or observes only one of the sunspots. I show, for both the linear case and nonlinear case, that there exist adaptively stable restricted perception Markov stationary sunspot equilibria (RP-SSEs) near an indeterminate steady state.
A new behavioral concept, local rationality, is developed within the context of a simple heterogeneous-agent model with incomplete markets. To make savings decisions, agents must forecast the shadow price of asset holdings. In the absence of aggregate uncertainty, locally-rational agents forecast shadow prices rationally, and thereby make optimal state-contingent decisions; these agents then use estimated econometric models to extend their rational shadow-price forecasts to accommodate aggregate uncertainty.
I introduce local rationality to a New Keynesian economy with incomplete markets and sticky nominal prices. Agents are heterogeneous and face idiosyncratic wage risks. Both aggregate productivity shocks and monetary policy shocks are incorporated into the model. Agents are assumed to be locally rational in the sense that they make optimal state-contingent decisions in the absence of aggregate uncertainties. Agents use estimated econometric models to forecast their shadow prices to accommodate aggregate uncertainties. In a calibrated model that captures features of US income inequality, I implement multiple monetary and fiscal experiments and show that the aggregate responses to policies differ from their counterparts in a similar model with fully rational agents.
We examine global economic dynamics with a nonlinear interest-rate rule when private agents i) do not observe persistent exogenous shocks, and ii) adopt an autoregressive (AR) forecast rule. Provided these two conditions are both satisfied, the economy leaves the targeted steady state with only small shocks. Empirical evidence of the use of AR rule by private agents is provided. A carefully calibrated New Keynesian version of the model combined with AR rule is consistent with the recent downward trend in inflation and interest rate when real productivity shocks are measured to be persistent. (With this insight, we give suggestions on fiscal and monetary policies.)
Works in Progress
E-stability Principle with Social Learning (with George Evans and Bruce McGough)
We show with simulations that E-stability principle holds in a reduced-form New Keynesian model where agents hold different beliefs and update through genetic learning algorithm. In addition, we provide the details for how each fundamental parameter affects the speed of convergence.