Advancing Long-Term Consistency in Multihorizon Equity Return Forecasting: A Conditional Clustering-Based Linear Regression Approach

Forecasting equity returns is a cornerstone of modern portfolio theory and pivotal to devising various portfolio construction strategies. Long-term forecasts have traditionally relied on methods grounded in valuation mean reversion, such as Shiller’s cyclically adjusted price-to-earnings (CAPE) ratio, alongside <italic>sum of parts</italic> approaches. The development of a consistent multihorizon forecasting model, capable of predicting one- to two-year equity returns in alignment with long-term methodologies, however, remains a significant challenge. Addressing this gap, our article introduces a novel framework that synergizes a 10-year long-term fair value model with a conditional clustering-based linear regression approach. This hybrid model ensures alignment with the long-term model while enhancing multihorizon forecast accuracy. Our empirical analysis, conducted across global markets, confirms the model’s robustness, offering a viable alternative for global multihorizon portfolio construction. The results not only reinforce the model’s applicability across different economies but also highlight its potential to revolutionize strategic asset allocation by offering more reliable multihorizon equity return predictions.