DeXposure-FM: A Time-series, Graph Foundation Model for Credit Exposures and Stability on Decentralized Financial Networks

Credit exposure in Decentralized Finance (DeFi) is often implicit and token-mediated, creating a dense web of inter-protocol dependencies. Thus, a shock to one token may result in significant and uncontrolled contagion effects. As the DeFi ecosystem becomes increasingly linked with traditional financial infrastructure through instruments, such as stablecoins, the risk posed by this dynamic demands more powerful quantification tools. We introduce DeXposure-FM, the first time-series, graph foundation model for measuring and forecasting inter-protocol credit exposure on DeFi networks, to the best of our knowledge. Employing a graph-tabular encoder, with pre-trained weight initialization, and multiple task-specific heads, DeXposure-FM is trained on the DeXposure dataset that has 43.7 million data entries, across 4,300+ protocols on 602 blockchains, covering 24,300+ unique tokens. The training is operationalized for credit-exposure forecasting, predicting the joint dynamics of (1) protocol-level flows, and (2) the topology and weights of credit-exposure links. The DeXposure-FM is empirically validated on two machine learning benchmarks; it consistently outperforms the state-of-the-art approaches, including a graph foundation model and temporal graph neural networks. DeXposure-FM further produces financial economics tools that support macroprudential monitoring and scenario-based DeFi stress testing, by enabling protocol-level systemic-importance scores, sector-level spillover and concentration measures via a forecast-then-measure pipeline. Empirical verification fully supports our financial economics tools. The model and code have been publicly available. Model: https://huggingface.co/EVIEHub/DeXposure-FM. Code: https://github.com/EVIEHub/DeXposure-FM.

💡 Research Summary

The paper introduces DeXposure‑FM, the first time‑series, graph‑based foundation model designed to measure and forecast inter‑protocol credit exposures in decentralized finance (DeFi). Leveraging the newly constructed DeXposure dataset—comprising 43.7 million weekly entries that span over 4,300 protocols, 602 blockchains, and 24,300 unique tokens—the authors capture a weighted, directed multigraph where nodes represent protocols and edges encode token‑mediated credit exposures derived from token composition and valuation dynamics.

DeXposure‑FM’s architecture consists of a large‑scale graph‑tabular encoder (based on GraphPFN) that jointly processes tabular features (TVL, token holdings, chain‑bridge flows) and graph structure, followed by three task‑specific heads: (1) binary classification of edge existence, (2) regression of edge weights, and (3) multi‑step prediction of node‑level TVL changes. The encoder is initialized with publicly available pretrained weights, while the heads are trained from scratch. Training employs Adam optimization, time‑based forward splits, early stopping, and gradient clipping to handle the non‑stationary, heteroskedastic nature of DeFi data.

Evaluation is conducted on two bespoke benchmarks. The first benchmark assesses multi‑step forecasts of edge‑level exposures and aggregate network statistics such as density, concentration, and sector connectivity. DeXposure‑FM outperforms strong baselines—including Graph‑PFN, the temporal graph neural network ROLAND, and a persistence model—by 12‑23 % on MAE and achieves substantially higher R² scores for edge‑weight regression. The second benchmark, “predict‑then‑measure” stress testing, uses the model’s forecasted exposure graph to simulate system losses under predefined shock scenarios (e.g., stablecoin price crashes) and compares them to realized losses. The model‑driven simulations explain over 85 % of the actual loss, and correctly identify the most systemic protocols and spill‑over pathways.

Beyond raw predictive performance, the authors translate model outputs into a suite of macro‑prudential tools. By feeding the predicted graph into standard network‑analysis metrics (centrality, PageRank, beta‑diversity), they generate protocol‑level systemic‑importance scores, sector‑level spill‑over maps, and early‑warning concentration indices. These tools are validated empirically, showing strong alignment with observed market stress events.

All code, model weights, and the DeXposure dataset are openly released on HuggingFace and GitHub, ensuring reproducibility and facilitating community‑driven extensions. The paper outlines a future roadmap: expanding the data pool, incorporating richer financial signals beyond TVL, implementing continuous drift monitoring and periodic retraining, exploring architectural enhancements, and fostering an open‑source ecosystem through competitions and collaborative development.

In summary, DeXposure‑FM advances the state of the art in DeFi risk analytics by jointly modeling temporal dynamics and graph structure at massive scale, delivering superior forecasting accuracy, and providing actionable systemic‑risk metrics for regulators, protocol developers, and investors.