Researchers propose autonomous trading signal discovery using AI evolution

Researchers from institutions including Shanghai Jiao Tong University have published a comprehensive framework for autonomous trading signal discovery, introducing a novel perspective that treats this as a symbolic evolutionary optimization problem. The work, titled 'Towards Autonomous Formulaic Alpha Discovery: An Evolutionary Computation Perspective' and published on arXiv (arXiv:2608.01789), presents a six-component analytical framework that characterizes existing approaches through representation, variation, fitness evaluation, selection, memory, and adaptation.

The team also proposes an eight-dimensional evaluation framework covering search efficiency, fitness reliability, residual alpha quality, and economic diversity among other metrics. This systematic approach aims to address key challenges in automated trading signal generation including market nonstationarity, noisy fitness estimates, and semantic redundancy. The work unifies diverse AI techniques including genetic programming, reinforcement learning, and large language models under a common evolutionary computation paradigm, providing researchers with tools to diagnose component-level limitations in existing systems.