Learning in economic systems with expectations feedback (Q2503442)

These notes study the interplay between realizations of economic variables and agents' beliefs concerning their future development. A widely accepted concept to describe the way in which agents form and adjust their forecast rules as new data becomes available over time is that of adaptive learning schemes. While many economists use the learning schemes based on the ordinary least squared method when modelling learning agents, there are some open issues which have not been resolved yet. The notes are motivated by them. In particular, they deal with the following topics: the criteria that assure global convergence for nonlinear models, selecting between multiple rational-expectations solutions criteria during the estimation process using economic criteria, a systematic treatment of models which involve forecasts for more than two periods ahead of the current state, and learning schemes that generate forecasts converging to rational expectations equilibria in the case of economies whose evolution is driven by nonlinear maps. The notes start with a brief Introduction. After that, Chapter 2 presents the general setup for nonlinear economic models with expectational leads. The basic methodological concepts are contained in Chapter 3. They include a learning scheme based on the approximate-maximum-likelihood (AML) algorithm of \textit{T. L. Lai} and \textit{C. Z. Wei} [Stochastics 16, 227--254 (1986; Zbl 0595.60048)] which converges globally for all initial conditions under standard assumptions. Methodologically, four separate issues are distinguished: existence of a desired unbiased forecasting rule, the dynamic stability of the system under an unbiased forecasting rule, the dynamic stability of the system under the applied learning scheme, and the success of the learning scheme in terms of providing strongly consistent estimates of the parameters. In Chapter 4, necessary conditions for the existence and uniqueness of unbiased forecasting rules in the case of stationary noise are established. The nonparametric estimation techniques developed by \textit{X. Chen} and \textit{H. White} [J. Econ. Theory 82, No. 1, 190--222 (1998; Zbl 0910.90002) and Stud. Nonlinear Dyn. Econom. 6, No. 1, Article 1, 53 p., electronic only (2002; Zbl 1178.62089)] are combined in Chapter 5 with the notion of an economic law to establish conditions under which strongly consistent estimates of an unbiased no-updating rule exist. As a first application of the results obtained in Chapters 2 and 5, a standard version of the overlapping generations model with pure exchange is discussed in Chapter 6. The analysis of adaptive learning in a financial market with heterogeneous traders, presented in Chapter 7, concludes the notes. The notes are written carefully and with considerable rigor which should make them particularly relevant for those interested in perfect forecasting rules taking into account the correct feedback structure of an economy.