Investment Rating - The report does not explicitly provide an investment rating for the industry or specific companies [2]. Core Insights - The report focuses on the challenges faced by traditional portfolio optimization methods due to unknown asset return distributions and proposes a Bayesian Predictive Synthesis (BPS) framework to address market uncertainties. This framework integrates multiple expert predictions using a Dynamic Linear Model (DLM) to create a posterior predictive distribution of asset returns, offering a new approach for robust asset allocation in uncertain environments [2][3]. Summary by Sections Introduction - Portfolio optimization is a key challenge in investment, aiming to appropriately allocate various financial assets to achieve ideal asset management. Traditional methods like mean-variance optimization require knowledge of asset return distributions, which are often unknown and can significantly impact portfolio performance [14][15]. BPS Framework - BPS is a Bayesian framework that integrates multiple expert predictions into a unified posterior predictive distribution. The use of a Dynamic Linear Model allows for capturing non-stationarity and time-varying characteristics in financial time series data, providing robust inputs for subsequent portfolio optimization [3][21]. Portfolio Construction Methods - The report discusses how to utilize the posterior predictive distribution generated by BPS to drive three mainstream portfolio construction strategies: - Mean-Variance Portfolio: Explores constrained optimization forms based on posterior mean and variance [32]. - Quantile-Based Portfolio: Introduces Bayesian versions of VaR/CVaR and VoR/CVoR as optimization objectives or constraints [34]. - Risk Parity Portfolio: Defines marginal risk contributions and seeks weights to equalize contributions from each asset [37]. Empirical Analysis - Empirical tests in the US and Japanese markets demonstrate that the BPS-based portfolio optimization method (BPPS) performs well without significant performance degradation, showing robustness against poorly performing predictive models [5][38][50]. Conclusion - The study introduces a method for optimizing portfolios based on posterior predictive distributions obtained through BPS, effectively addressing uncertainties in asset return distributions. The integration of expert predictions through a Dynamic Linear Model captures the uncertainties in time series data, confirming the effectiveness of the proposed methods through empirical testing [51][52].