Relevance-Based Prediction

The CSA Prediction Vault is powered by Relevance-Based Prediction, a new system for predicting player outcomes.

Overview

Relevance-based prediction (RBP) is a new model-free forecasting routine that overcomes the limitations of both classical prediction models and machine learning algorithms. RBP forms predictions as weighted averages of past observations, where the weights are determined using fundamental principles of information theory to gauge which observations and predictive variables are the most important for each individual prediction task. RBP’s unique approach inherently offers a level of transparency and intuition that is impossible to achieve with machine learning models.

Advantages of Relevance-Based Prediction

• RBP addresses complex data relationships that are beyond the reach of linear regression analysis.

• It is transparent and adaptive, unlike machine learning algorithms.

• It identifies the optimal combination of observations and predictive variables for each individual prediction task.

• It reveals in advance the reliability of each individual prediction.

• It guards against overfitting and offers protection from data errors.

• It includes a built-in measure of variable importance that is robust to multicollinearity and non-linearity.

• It is theoretically justified by information theory and other fundamental principles.

Awards and Additional Resources

To learn more about the founding principles of RBP, check out our 2022 book “Prediction Revisited: The Importance of Observation” by Megan Czasonis, Mark Kritzman, and David Turkington: www.predictionrevisited.com

RBP has been applied extensively in the field of financial markets and investing, and two of the seminal research papers have won prestigious industry awards.

The 2022 Harry M. Markowitz Award, First Prize

Best paper in the Journal of Investment Management for "Relevance" by Megan Czasonis, Mark Kritzman, and David Turkington. Final selection of the prize winners was conducted by a panel of Nobel Prize Laureates in Econmics, who are members of the Journal of Investment Management Advisory Board.

The 2023 Roger F. Murray Award, First Prize

Outstanding research presented at the seminars of The Institute for Quantitative Research in Finance (Q-group) for "Relevance-Based Prediction: A Transparent and Adaptive Alternative to Machine Learning" presented by David Turkington, co-authored by Megan Czasonis and Mark Kritzman.