Dennis Umlandt

Research


Working Paper

This paper proposes a novel approach for estimating linear factor pricing models with dynamic risk premia based on a generalized method of moments (GMM) framework. Time-varying risk prices and exposures follow an updating scheme that aims for the steepest descent of the conditional moment-criterion function. The most informative moment for inferring risk premium dynamics comes from the cross-sectional pricing equation estimated in the second stage of the widely used Fama-MacBeth regression approach. Monte Carlo results show that the new approach is able to adequately filter various types of risk premium dynamics. An application to the Fama-French 5-factor model shows that the GMM-based procedure can largely reduce pricing errors compared to other dynamic and static approaches. The results show that premium dynamics vary across factors, and while they are generally countercyclical, they exhibit significant declines at the beginning of crisis periods.

Presentations: Statistical Week 2023 (Dortmund), German Finance Association DGF 2023 (Hohenheim), University of Liechtenstein, University of Graz, CFE 2023 (Berlin), Midwest Finance Association 2024 (Chicago), SNDE Symposium 2024 (Padova), QFFE 2024 (Marseille), SoFiE (Pre-)Conference 2024 (Rio de Janeiro), IAAE 2024 (Thessaloniki), EEA-ESEM 2024 (Rotterdam, scheduled), VfS 2024 (Berlin, scheduled).


We propose a novel dynamic mixture vector autoregressive (VAR) model in which time-varying mixture weights are driven by the predictive likelihood score. Intuitively, the state weight of the k-th component VAR model in the subsequent period is increased if the current observation is more likely to be drawn from this particular state. The model is not limited to a specific distributional assumption and allows for straightforward likelihood-based estimation and inference. We conduct a Monte Carlo study and find that the score-driven mixture VAR model is able to adequately filter the mixture dynamics from a variety of different data generating processes which most other observation-driven dynamic mixture VAR models cannot appropriately cope with. Finally, we illustrate our approach by an application where we model the conditional joint distribution of economic and financial conditions and derive generalized impulse responses. 

Presentations: CFE 2021 (Online), SNDE Symposium 2022 (Online), CEF 2022 (Dallas), IAAE 2022 (London), VfS 2022 (Basel), Statistical Week 2022 (Muenster).

Revise and Resubmit at Journal of Applied Econometrics


Recursively identified vector autoregressive (VAR) models often lead to a counterintuitive response of prices (and output) shortly after a monetary policy shock. To overcome this problem, we propose to estimate the VAR parameters under the restriction that economic theory is not violated, while the shocks are still recursively identified. We solve this optimization problem under non-linear constraints using an augmented Lagrange solution approach, which adjusts the VAR coefficients to meet the theoretical requirements. In a generalization, we allow for a (minimal) rotation of the Cholesky matrix in addition to the parameter restrictions. Based on a Monte Carlo study and an empirical application, we show that particularly the “almost recursively identified approach with parameter restrictions” leads to a solution that avoids an estimation bias, generates theory-consistent impulse responses, and is as close as possible to the recursive scheme. 

Presentations:  VfS 2022 (Basel).


Publications

Journal of Econometrics, 2023, 237 2C: 105470. https://doi.org/10.1016/j.jeconom.2023.05.007.

    Best Doctoral Paper Award DGF 2019

Journal of International Economics, 2021, 133: 103541. https://doi.org/10.1016/j.jinteco.2021.103541.