Title: Pearl Judea, Causality: Models, Reasoning, and Inference, Second Edition (2009)

Authors: Feng Li

Journal: International Journal of Forecasting

Summary:

With the big popularity and success of Judea Pearl’s original causality book, this review covers the main topics updated in the second edition in 2009 and illustrates an easy-to-follow causal inference strategy in a forecast scenario. It further discusses some potential benefits and challenges for causal inference with time series forecasting when modeling the counterfactuals, estimating the uncertainty and incorporating prior knowledge to estimate causal effects in different forecasting scenarios.

DOI | Eprint

Causality inference research has advanced significantly in the past two decades. Due to increases in the volume of available data, causal inference with graphic models has been applied to handle large-scale and high-dimensional data sets. Causality inference provides flexible approaches for handling complex data structures and the incorporation of prior knowledge. Various causal discovery and inference algorithms have been developed. These algorithms provide automated methods for inferring causal relationships from observational and experimental data. New methods have also been developed to address challenges when inferring causality from observational data, such as propensity score matching and regression discontinuity designs. These techniques aim to overcome confounding biases and establish causal relationships.

The integration of causality and machine learning has received significant attention. Methods have been proposed that combine causal inference with machine learning techniques, such as causal forest, causal boosting, and causal generative models, in order to leverage the strengths of both fields and enable causal reasoning in predictive models. Interestingly, although many machine learning and deep learning methods have been successfully adopted in the forecast community, there is still relatively little focus on causality for time series forecasting. This is demonstrated by Pearl’s collection of journal reviews for his book from various disciplines, ¹ which is a useful addition to the book. It is not surprising that I found no related reviews from a forecasting perspective.

To review the concepts of causality from a forecasting perspective, we can consider a typical forecasting scenario where we want to forecast the sales performance of different retail stores over time such as the M5 competition (Makridakis et al., 2022). The following questions are of immediate interest to us.

• •Why should forecasters know about causality?
• •How do I know if I have enough data for causal inference?
• •What tools can be directly picked up by forecasters?
• •Are we still restricted by computational challenges?