This website uses cookies.

By clicking the "Accept" button or continuing to browse our site, you agree to first-party and session-only cookies being stored on your device to enhance site navigation and analyze site performance and traffic. For more information on our use of cookies, please see our Privacy Policy.

  1. Home
  2. ASSA Annual Meeting
  3. 2026
  4. Program
  5. Nonlinear Time Series
« Back to Results

Nonlinear Time Series

Paper Session

Monday, Jan. 5, 2026 1:00 PM - 3:00 PM (EST)

Philadelphia Marriott Downtown, Room 307
Hosted By: Econometric Society
  • Chair: Ulrich K. Müller, Princeton University

Deep Autoencoders for Nonlinear Factor Models: Theory and Applications

Zhouyu Shen
,
Peking University
Dacheng Xiu
,
University of Chicago

Abstract

Autoencoders are neural networks widely used in unsupervised learning tasks such as dimensionality reduction and feature extraction. This paper establishes nonasymptotic guarantees for deep autoencoders within a nonlinear factor model, demonstrating their ability to effectively extract latent components with errors that diminish as dimensionality and sample size increase. The extracted factors are shown to converge to the true latent factors, up to a functional transformation. Furthermore, we extend these results to supervised autoencoders, providing theoretical support for their application in factor-augmented prediction and structured matrix completion. Finally, we showcase the practical utility of supervised autoencoders in macroeconomic forecasting and asset return prediction, and of autoencoders in noise reduction for causal analysis.

Identification, Estimation and Inference in High-Frequency Event Study Regressions

Alessandro Casini
,
University of Rome Tor Vergata
Adam McCloskey
,
University of Colorado

Abstract

We consider identification, estimation and inference in high-frequency event study regressions, which have been used widely in the recent macroeconomics, financial economics and political economy literatures. The high-frequency event study method regresses changes in an outcome variable on a measure of unexpected changes in a policy variable in a (narrow) time window around an event or a policy announcement (e.g., a 1-day or 30-minute window around an FOMC announcement). We show that, contrary to popular belief, the narrow size of the window alone is not sufficient for identification. Rather, the population regression coefficient identifies a causal estimand when (i) the effect of the policy shock on the outcome does not depend on the other variables (separability) and (ii) the surprise component of the news or event dominates all other variables that are present in the event window (relative exogeneity). We establish the causal meaning of the event study estimand and the super- consistency of the event study estimator. We further show the estimator’s asymptotic normality, derive bounds on its worst-case bias, and develop bias-corrected inference procedures. Notably, this standard linear regression estimator is robust to general forms of nonlinearity. We provide a simple sensitivity analysis and apply our results to Nakamura and Steinsson’s (2018a) analysis of the real economic effects of monetary policy, which we use to revisit the recent debate on the “Fed information effect” and Blue Chip forecasts regressions.

Testing Whether Volatility is Local or Stochastic

Yacine Ait-Sahalia
,
Princeton University
Jean Jacod
,
Sorbonne Universite

Abstract

There is broad empirical agreement that the volatility of most economic and financial variables is time-varying. One possible way to capture this in continuous-time models is to make the volatility its own latent process, leading to a
stochastic volatility model. Another approach consists in
making the volatility of a given variable a function of the variable itself, leading to a local volatility model. Local models are more restrictive, but also simpler and lower dimensional, and are often used in applications. This paper tests whether the restriction imposed by a local volatility model is empirically justified. The proposed test is based on high frequency asymptotics, is robust to the potential
presence of jumps and/or noise in the data, and has power against a wide range of alternative non-local volatility models. The test is applied to stock price, interest rate, currency and volatility data and suggests that volatility is not local in all of them.

Efficient Estimation of Nonlinear DSGE Models

Eva Janssens
,
University of Michigan
Sean McCrary
,
Ohio State University

Abstract

This paper introduces a computationally efficient and accurate method for estimating nonlinear DSGE models by integrating model solution and filtering.
We combine local linear solution techniques with a time-varying Kalman filter, yielding the optimal linear filter for this solution method.
We benchmark the method using the canonical Diamond-Mortensen-Pissarides (DMP) model and find that it matches or exceeds the accuracy of global solution methods combined with nonlinear filters.
Moreover, it reduces computational time by several orders of magnitude relative to existing estimation approaches.
A key feature of the framework is that it produces state-dependent impulse responses, computed at each point in time from filtered states.
Applying the method to a New Keynesian model with labor market frictions, we find that endogenous variation in job creation profitability induces strong state dependence in monetary policy transmission.
In both the estimated model and reduced-form state-dependent local projections, the Phillips Multiplier is approximately 50 percent larger during expansions than during recessions.
JEL Classifications
  • C0 - General