LLM Agents as VC investors: Predicting Startup Success via RolePlay-Based Collective Simulation

TL;DR

This work reframes startup success prediction as the outcome of a collective VC decision process, modeling a network of heterogeneous investors whose judgments evolve through role-playing with LLMs and graph-based interaction. The SimVC-CAS framework combines Startup Panoramic Portrait, Heterogeneous Investor Portraits, and a VGAT-enabled collective interaction module to capture enterprise fundamentals and investor-network dynamics. Empirical results on PitchBook data show substantial gains in top-k ranking metrics (e.g., $AP@10$) over strong baselines, accompanied by interpretable explanations of decision dynamics. The approach provides a generalizable paradigm for representing complex group decision-making in finance and other domains with rich relational structures.

Abstract

Due to the high value and high failure rate of startups, predicting their success has become a critical challenge across interdisciplinary research. Existing approaches typically model success prediction from the perspective of a single decision-maker, overlooking the collective dynamics of investor groups that dominate real-world venture capital (VC) decisions. In this paper, we propose SimVC-CAS, a novel collective agent system that simulates VC decision-making as a multi-agent interaction process. By designing role-playing agents and a GNN-based supervised interaction module, we reformulate startup financing prediction as a group decision-making task, capturing both enterprise fundamentals and the behavioral dynamics of potential investor networks. Each agent embodies an investor with unique traits and preferences, enabling heterogeneous evaluation and realistic information exchange through a graph-structured co-investment network. Using real-world data from PitchBook and under strict data leakage controls, we show that SimVC-CAS significantly improves predictive accuracy while providing interpretable, multiperspective reasoning, for example, approximately 25% relative improvement with respect to average precision@10. SimVC-CAS also sheds light on other complex group decision scenarios.

Figures (7)

• Figure 1: Unlike the previous single decision-making perspective methods, our method achieves joint modeling of the collective behavior of startups and investors by introducing the perspectives of multiple decision-makers.
• Figure 2: The overall framework of our proposed method.
• Figure 3: Performance comparison (F1 score) across varied interaction strategies with different $k$ values.
• Figure 4: The consistency experiment results.
• Figure 5: An example demonstrating how investors make investment decisions and ultimately revise their investment decisions through interaction with other investors.