Chapter 7 Learning dynamics

Comment on “Response of Asset Prices to Monetary Policy under Abenomics”

Rational expectations in the overlapping generations model

Indeterminate equilibrium rational expectations (RE) models are ubiquitous in both theoretical and applied work in dynamic macroeconomics. The issue of characterizing the exact dimension of indeterminacy—i.e. of deriving the full set of causal and stable solutions to linear RE models—has only recently been addressed in the context of general and multivariate settings. This paper complements existing results by identifying bounds on the observable dimension of indeterminacy of linear RE models in the presence of arbitrary initial conditions. Implications for the estimation of indeterminate equilibrium RE models are discussed.

Dynamic Monetary and Fiscal Policy Games under Adaptive Learning

The idea that markets might aggregate and disseminate information and also resolve conflicts is central to the literature on decentralization (Hurwicz, 1972) and rational expectations (Lucas, 1972). We report on three series of experiments all of which were predicted to have performed identically by the theory of rational expectations. In two of the three series (one in which participants trade a complete set of Arrow-Debreu securities and a second in which all participants have identical preferences), double auction trading leads to efficient aggregation of diverse information and rational expectations equilibrium. Failure of the third series to exhibit such convergence demonstrates the importance of market institutions and trading instruments in achievement of equilibrium.

Characterizing Markov-Switching Rational Expectations Models

Panoptic segmentation that unifies instance segmentation and semantic segmentation has recently attracted increasing attention. While most existing methods focus on designing novel architectures, we steer toward a different perspective: performing automated multi-loss adaptation (named Ada-Segment) on the fly to flexibly adjust multiple training losses over the course of training using a controller trained to capture the learning dynamics. This offers a few advantages: it bypasses manual tuning of the sensitive loss combination, a decisive factor for panoptic segmentation; allows to explicitly model the learning dynamics, and reconcile the learning of multiple objectives (up to ten in our experiments); with an end-to-end architecture, it generalizes to different datasets without the need of re-tuning hyperparameters or re-adjusting the training process laboriously. Our Ada-Segment brings 2.7% panoptic quality (PQ) improvement on COCO val split from the vanilla baseline, achieving the state-of-the-art 48.5% PQ on COCO test-dev split and 32.9% PQ on ADE20K dataset. The extensive ablation studies reveal the ever-changing dynamics throughout the training process, necessitating the incorporation of an automated and adaptive learning strategy as presented in this paper.

For the last quarter of a century, the Real Business Cycle model has dominated the interpretation of business cycles in mainstream economics; yet a number of significant empirical objections to it justify exploring an alternative approach. This paper proposes to base such an approach on a predator-prey mechanism, along the lines of the classical Lotka-Volterra model for ecosystem dynamics, where agency costs play the role of the predatory activity, in a process very similar to the one proposed by the classical Austrian School interpretation of the cycle (Hayek/Schumpeter). The model is consistent with both Rational Expectations and the Efficient Markets Hypothesis, and predicts that stock market valuations will regularly present bubbles and crashes synchronised with the business cycle without this implying any irrational behaviour on the part of the investors.

Learning Process and Rational Expectations: An Analysis Using a Small Macroeconomic Model for New Zealand

Neftci (1984) proposed a nonparametric test procedure for determining whether business cycles are asymmetric in the sense that contractions are steeper than expansions. He found evidence of this type of asymmetry in postwar quarterly U.S. unemployment variables.' This comment identifies a probable error in Neftci's empirical work that reverses the significance of his results for the unemployment rate, and it suggests that this test may have low power and be sensitive to measurement error.

Rational expectations are frequently justified as the point of convergence of agents' learning process. When agents' learning feeds back on the actual law of motion of the economy convergence of their learning rule to a rational expectations equilibrium (REE) is not guaranteed however. Applying new methods to analyse the convergence of learning in a model of UK stock prices we find evidence that agents could not have learned to form rational expectations if they had attempted to estimate the long-run dynamics of the model. If, however, agents have strong priors and impose a unit root on the model, thus confining their learning to the short run dynamics, there is evidence that recursive learning may eventually lead them to a REE. The learning process on the path to this equilibrium is highly volatile, suggesting that learning may help to explain excess volatility in UK stock prices.

Poverty traps, the money growth rule, and the stage of financial development

IntroductionRecent Developments in Monetary Macroeconomics David Altig The contents of this volume hardly require explanation beyond its title: "Recent Developments in Monetary Macroeconomics." Our intent for the conference was to collect a set of papers reflecting the cutting edge of applied monetary macroeconomics. As befitting such an ambitious-sounding goal, the contributions are wide ranging. For purposes of this brief introduction, however, we might organize the papers as answers to three questions. (1) What is the "optimal" Taylor rule? (2) Are "New Keynesian" or "New Neoclassical Synthesis" models the final word on the monetary transmission mechanism? (3) Is there a role for money in the conduct of monetary policy? What is the "Optimal" Taylor Rule The inclusion of the Taylor rule issue is almost a prerequisite for any collection of papers purporting to survey recent developments in monetary macroeconomics. In policy discussions, the Taylor rule is ubiquitous, and it is currently the choice among alternative specifications of central bank behavior. More precisely, perhaps, the general form of the Taylor rule is the choice, as it has come to represent the general class of equations that relate the federal funds rate to some measure of an output gap and inflation rate. There is substantially less unanimity about whether output gaps and inflation rates should be past, present, or (expected) future values, whether past values of the funds rate need to be included, and what are the appropriate magnitudes of the responses to each of these measures. Marc Giannoni and Michael Woodford offer the natural approach to addressing the dispute: find the representation that is optimal within the framework being employed for policy analysis. The model in question here is essentially a derivative of the "New Neoclassical Synthesis" class of models that are currently the workhorses of most monetary policy analyses among academic and central bank staffs alike. Their approach to discovering the optimal policy within this structure has the flavor of [End Page 1039] the "Ramsey problem" familiar from optimal tax policy, although with the strong requirement that the derived policy rule be "robustly optimal": it must support the optimal equilibrium no matter what the distribution of disturbances the model policymakers face. Giannoni and Woodford offer two essential lessons. First, whether optimal policy incorporates forecasts of future price-level growth or output gaps depend critically on the dynamics of the inflation rate. If the adjustment of the price-level to shocks is inertial, then optimal policy necessarily depends on forecasts of future inflation. Second, the response of the funds rate to its own past is inertial. In fact, it is super-inertial, meaning that (all else equal), the contemporaneous funds rate responds more than one-for-one with the lagged value of the funds rate (and lagged changes in the rate). The proposition that monetary authorities ought to aggressively respond to lagged values of the funds rate also arises in the papers by Jess Benhabib, Stephanie Schmitt-Grohé, and Martín Uribe, and George Evans and Seppo Honkapohja. In the latter case, the authors consolidate and expand on their well-known work on learning dynamics. A central contribution of the work presented in this article is the notion that convergence to a unique rational expectations equilibrium under learning, as well as the stability of that equilibrium, serves as basis for the choice (or rejection) of an optimal policy formulation. An apparent lesson from Evans and Honkapohja's analysis is that the learnability criterion prescribes a policy rule that differs in some important ways from the conventional wisdom coming from analyses that invoke the Taylor rule in a pure rational expectations environment. In particular, they conclude that the optimal policy rule in the environment they consider requires the monetary authority to respond directly to private-sector expectations. The environment they consider is essentially the same as in Giannoni and Woodford, with the exception of the central bank's assumed loss function: Giannoni and Woodford assume a preference for interest rate smoothing, Evans and Honkapohja do not. In his comments, John Duffy points out that the introduction of interest-smoothing motive yields an optimal policy rule under learning that is much closer to the conventional view. In particular, it does not require...

has strongly questioned the basic economic logic of current mainstream monetary policy analysis, arguing that the standard notion --that "determinacy" of a rational expectations (RE) equilibrium suffices to imply that stable inflation behavior will be generated --is incorrect. This is because New Keynesian (NK) models are typically consistent with the existence of RE paths with explosive inflation rates (in addition to one or more stable paths) that normally do not imply explosions in real variables relevant for transversality conditions. Consequently, the usual logic does not imply the absence of explosive inflation. That result does not, however, justify negative conclusions about NK analysis. For there is a different criterion that is logically satisfactory for the purpose at hand. This is the requirement that, to be plausible, a RE solution must satisfy the property of least-squares learnability. Adoption of this criterion, which should be attractive to analysts concerned with actual monetary policy, serves to justify in principle the bulk of current mainstream analysis.

This article presents a laboratory experiment about capital structure irrelevance. The equity and debt issued by a firm are traded in two double-auction markets. In the Public treatment a signal carrying information about the firm’s value is known to all traders, while in the Private treatment, it is only disclosed to some randomly chosen participants. In treatment Public the conditions for capital structure irrelevance hold. The experimental results support the irrelevance of the capital structure but reject the rational expectations (risk-neutral) valuation. Treatment effects are significant and information is more efficiently embedded in prices in the Public treatment. The results from treatment Private on prices support the prior information over the rational expectations equilibrium. The analysis of the final portfolios of informed and uninformed participants suggests incomplete adjustment towards the prior information equilibrium predicted holdings.