Chaos, Machine Learning and Deep Learning based Hybrid to forecast Consumer Price Index Inflation in India

In this study, we proposed a 2-stage hybrid approach for financial time series forecasting wherein chaos is modeled in stage-1 followed by forecasting is accomplished using machine learning and deep learning algorithms in stage-2. The effectiveness of the proposed hybrid is tested on forecasting Consumer Price Index Inflation of Food & Beverages, Fuel & Light, and Headline in India. This is a first-of-its-kind study where chaos is modeled and deep learning is employed in forecasting macroeconomic time series. From the results, it is inferred that Chaos + Machine learning hybrids yielded better forecasts than pure machine learning algorithms without Chaos in terms of Symmetric Mean Absolute Percentage Error (SMAPE), Theil's U statistic and Directional statistic across all the data sets. A deep learning model namely, Long Short Term Memory (LSTM) was also employed but without much success. The results of 2-stage hybrid models are compared with models without accounting for chaos. The results are encouraging and these hybrids can be applied to predict other financial time series.