Prompt Engineering: Optimization Comparison between ChatGPT and DeepSeek in Portfolio Selection

The present study analyzes the optimization efficiency of large language models (LLMs), including ChatGPT and DeepSeek in the portfolio selection process. Employing a novel prompt engineering method, we aim to identify their processing and training efficiency. The results show that the initial allocations of ChatGPT are more diversified than those of DeepSeek. However, as we move from basic to advanced prompts, the allocation efficiency of ChatGPT improves compared to DeepSeek in providing better risk-return allocations. Moreover, ChatGPT follows an aggressive stock allocation strategy compared to DeepSeek's balanced and stable allocation strategy. Further, the efficient frontier plots also support a better performance in allocation by ChatGPT compared to DeepSeek. The present study has important implications for prompt engineers, financial advisory firms, and retail investors.