Stochastic loss reserving models are crucial tools for general insurers to predict outstanding claims for meeting regulation requirements and risk management, and they have been widely studied in literature. However, previous studies often focus on identifying a single model that can generate superior predictive performance. This model selection approach may not fully utilize the strengths offered by different models and may generate volatile prediction outcomes. Although combining models is not new in practice, the model weights are usually selected subjectively based on previous business experience, which is not always suitable when updating models for new data. Therefore, this study aims to develop a rigorous way to combine notable loss reserving models such that the strengths offered by different models can be utilized effectively.