SpiralSeer: A Stage-Wise Risk Prediction Framework for Algorithmic Stablecoins in DeFi

Abstract

Algorithmic stablecoins play a critical role in the Decentralized Finance (DeFi) ecosystem by aiming to maintain price stability without relying on traditional collateral reserves. However, these systems are prone to catastrophic failures known as death spirals, leading to irreversible price collapse and systemic instability. Despite increasing attention from both academia and regulators, existing approaches fall short in providing proactive prediction of such destabilizing events-an issue that poses serious risks to the sustainable development of the blockchain financial ecosystem. In this paper, we present SpiralSeer, a novel framework for fine-grained prediction of death spiral risks in algorithmic stablecoins. At its core, SpiralSeer introduces a stage-wise risk model that integrates on-chain user behaviors and off-chain market data, formalizing distinct states in risk evolution. We then develop a LightGBM-based risk detection model capable of detecting vulnerabilities in emerging stablecoins, and incorporates an additive feature attribution mechanism to reveal the most influential factors contributing to risk across different stages. Extensive evaluation demonstrates that SpiralSeer achieves a $27.8 \%$ improvement in precision over state-of-the-art baselines, and can flag early risk stages well before catastrophic decoupling events. By enabling early risk identification, SpiralSeer offers a practical foundation for building more resilient algorithmic stablecoin systems and enhancing risk transparency in DeFi.