Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces a forecasting framework utilizing a time-varying autoregressive model (TVAR) that estimates parameters dynamically using deep learning techniques. By integrating both Gaussian and Laplace noise distributions, the approach effectively captures complex, nonstationary patterns in time series data while maintaining a transparent parametric structure. The findings demonstrate that this model can provide reliable predictions and uncertainty quantification, making it a versatile tool for forecasting in various contexts.
A simple deep learning model can accurately forecast complex dynamics by adapting its parameters over time, even under heavy-tailed noise conditions.
We investigate a forecasting framework based on a simple discrete-time dynamic model with coefficients varying in time. The parameters of the model are recovered within a deep learning framework, which makes it possible to retain a transparent parametric structure while simultaneously accounting for complex and nonstationary patterns in the observed phenomenon. Our analysis covers two specifications of the noise process. Besides the standard Gaussian setting, we also consider Laplace-distributed noise, which can offer a more adequate description in the presence of heavier tails and sharper local fluctuations. For both cases, we formulate the predictive scheme of the model and analyze the associated uncertainty quantification, including the construction of prediction intervals. The results illustrate that a relatively simple model, when combined with time-dependent parameter estimation, can serve as a mathematically tractable and practically flexible tool for forecasting complex dynamics under different noise assumptions. The general model is stated for TVAR($p$), while the prediction-interval formulas and the numerical experiments are developed for the TVAR(1) case.