Inflation Forecasting for Pakistan using Artificial Neural Networks

Abstract

Inflation forecasting is an important activity at central banks to formulate forward looking monetary policy. So the interest rate can be adjusted in order to curb inflation in the country. For this various techniques and models are proposed, machine learning is one of those. In machine learning, artificial neural network (ANN) is a popular tool. Further, RNN-LSTMs are special type of artificial neural networks that learn better from sequential data. In the past, researchers used both regular and RNN based artificial neural networks, but either they used only inflation data for training or their model was implemented for any other country. While there are other factors too that influence the inflation rate. Therefore, we attempted to solve the same problem in context of Pakistan. Not only we implemented RNN-LSTM but used other relevant features such as oil prices and exchange rates to train the model. We found best network architecture for our RNN-LSTM based neural network and the baseline model by exhaustively trying different number of nodes and layers. Then we trained both models first using only year-on-year monthly inflation and after that using all available features. Thus we got univariate and multivariate versions of both models i.e 4 models in total. Further, we ran all 4 models on Pakistan and the other four countries’ datasets. At the end, our RNN-LSTM based model clearly outperformed the baseline model not only in case of Pakistan but for other countries as well.