Forecasting

• Forecasting is the process of predicting future conditions of a time series.

• We will go through two related approaches which start from separate standpoints but overlap in some variations:

  • Machine Learning based methods handling time series as any other tabular data, possibly adding lagged variables to add a flavour of autoregressiveness.

  • Specific forecasting methods of included int he ARIMA family of methods, possibly adding exogenous variables bridging the gap to ML based methods. (We will use SARIMAX - Seasonal Auto-Regressive Integrated Moving Average with eXogenous factors and subsets of it).

• Neural network models for forecasting include: Recurrent Neural Networks, Long-Short Term Memory, Transformers, Retentive Networks, Facebook’s Prophet, Amazon’s GluonTS/DeepAR, etc.

  • Large Language Models, e.g., used in Chat GPT, Bard, etc., rely heavily on forecasting the next word(s) in a sequence.