I’m working on demand forecasts for five different market segments, each with its own product type. Each product type has 20 years of monthly data, which gives me 240 data points to rely on. My goal is to forecast demand one and three months ahead, for tactical and strategical purposes.
Currently I’m using the traditional ARIMA approach, however the latest predicitons where disappointing. Now I am looking for methods that will work better in a highly volatile and uncertain market environment. I have many economic and industrial features that I can include, however the problem lays in the low number of data points for my demand. This makes it difficult to get accurate predictions using for example machine learning approaches.
I am thinking about hybrid methods such as combining ARIMA with ANN or LSTM to improve my forecasts, but I’d love to hear any other suggestions for making better predictions with these limitations. Are there any other approaches that allow for many features and give accurate predictions even for short data sets?