Bayesian predictive synthesis: tutorial

Bayesian predictive synthesis combines predictive distributions from models or agents in a Bayesian analysis. The synthesis model represents the relationship between the outcome and the agents’ forecasts, allowing for biases, calibration, and dependencies. Dynamic models allow these relationships to change over time.

The collection includes papers by Ken McAlinn and other authors, with links to papers and code. Related work includes Bayesian causal synthesis, predictive decision synthesis, and the foundations of model combination.

Reading

The foundations develop Bayesian analysis of expert opinion and predictive distributions. Dynamic Bayesian predictive synthesis in time series develops the framework for forecasting; the multivariate paper extends the analysis to several outcomes.

The causal inference and decision synthesis papers extend the framework to treatment effects and decisions. The extensions and applications develop predictive synthesis for different data and decision problems. The code provides examples of the models and computations.

Foundations

Dynamic models

Theory and computation

Causal inference

From Trial Averages to Target Effects: Bayesian CATE Synthesis with Randomized and Observational Evidence

Ogawa, H.; Sugasawa, S.; Takanashi, K.; McAlinn, K. 2026. Working paper.

Decision synthesis

Extensions and applications

Economics

Finance

Health

Real estate and ecology

Others

Review and related work

Ensembling under Statistical Pluralism: A Taxonomy of Formulations, Diagnostics, and Reporting

McAlinn, K. 2026. Working paper.

Code and data

Code links are given where available. A paper without a code link may have additional materials available from its authors.

Updated: October 2026.