Examining models for Capital Market Assumptions

Capital Market Assumptions (CMAs) are a cornerstone of strategic asset allocation, offering structured estimates of expected returns and risk premia across asset classes. This thesis evaluates the performance of various forecasting models for equities, government bonds, and corporate bonds, with the aim of enhancing SEB Investment Management’s approach to medium- and long-term return forecasting. The analysis uses a rolling out-of-sample framework and evaluates predictive accuracy using RMSE and MAE over 3- and 5-year horizons. The results show that time-varying models generally outperform static ones, particularly under changing market conditions. These models are better suited to capture valuation dynamics, macroeconomic trends, and shifts in credit conditions that static, backward-looking models often fail to reflect. For equities, a building block model that incorporates dividend yield, earnings growth, valuation changes, and buybacks delivered the most accurate forecasts across most indices. In fixed income markets, mean-reverting models were most effective for government bonds, while both mean-reverting and macro-sensitive models performed well in the corporate bond segment. Overall, the findings suggest that while no model performs best in all contexts, adaptive and economically grounded approaches provide a strong foundation for producing CMAs. The ability to adjust to structural changes and evolving market environments is essential for maintaining forecast reliability, particularly over medium- to long-term investment horizons.