Varojen allokointi Transformer-pohjaisten aikasarjamallien avulla

Financial asset managers have long recommended the well-known 60/40 asset class allocation, which is often regarded as a general one-size-fits-all solution for pension investors. This allocation, consisting of {60\%} equity and {40\%} bonds, provides a risk-return profile that suits most investors' appetites by simultaneously producing reliable returns and mitigating systemic risk from volatile equity markets. However, this asset class allocation isn't absolute, and many asset managers actively deviate from it in pursuit of a more efficient portfolio. While individual allocations within portfolios have long employed systematic methods to optimize returns, recent advances in machine learning, particularly deep learning, offer new opportunities to extend these approaches to asset classes as a whole. Deep Learning models have the ability to process large, complex datasets while maintaining a high prediction accuracy which is required in the attempt to predict macro scale shifts in global financial markets. This research investigates the extent to which deep learning models can achieve a more efficient portfolio compared to traditional asset class allocation strategies. This prediction can be increasingly valuable for asset managers trying to navigate in volatile and uncertain market conditions. The paper contributes to existing research by developing and evaluating a deep learning–based asset allocation framework that dynamically adjusts equity and bond exposures with the aim of improving portfolio efficiency and resilience during volatile market regimes.