Nonlinear inflation forecasting with recurrent neural networks

Motivated by the recent literature that finds artificial neural networks (NN) can efficiently predict economic time-series in general and inflation particular, we investigate if forecasting performance be improved even further using a particular kind of NN—a recurrent network. We use long short-term memory network (LSTM) was proven to highly efficient for sequential data computed univariate forecasts monthly US CPI inflation. show though LSTM slightly outperforms autoregressive model (AR), NN, Markov-switching models, its is on par with seasonal SARIMA. Additionally, conduct sensitivity analysis respect hyperparameters provide qualitative interpretation what learn applying novel layer-wise relevance propagation technique.