Home

Awesome

Reproducibility Checks:

While it is not required for the reproducibility checks:

FEWNet: Filtered Ensemble Wavelet Neural Network

This repository is the source code for the research paper "Forecasting CPI Inflation under Economic Policy and Geo-political Uncertainties". The paper introduces a new approach called the Filtered Ensemble Wavelet Neural Network (FEWNet), which can generate accurate long-term predictions for CPI inflation. The idea utilizes a maximum overlapping discrete wavelet transform on the CPI inflation data to extract high-frequency and low-frequency signals. The wavelet-transformed series and filtered exogenous variables are fed into autoregressive neural networks downstream to get the ultimate ensemble forecast. Our theoretical analysis demonstrates that FEWNet effectively minimizes the empirical risk compared to individual, fully connected neural networks. Furthermore, we provide evidence that the real-time forecasts generated by the suggested algorithm, using a rolling-window approach, are notably superior in accuracy compared to standard forecasting methods used for comparison. In addition, we utilize conformal prediction intervals to measure the level of uncertainty linked to the projections produced by the suggested method. The outstanding performance of FEWNet can be ascribed to its ability to efficiently capture non-linearities and long-range relationships in the data via its flexible architecture.

References