Oracle Counterpoint: Relationships between On-chain and Off-chain Market Data

Paper: arXiv 2303.16331

Authors: Zhimeng Yang, Ariah Klages-Mundt, Lewis Gudgeon

Abstract

We investigate the theoretical and empirical relationships between activity in on-chain markets and pricing in off-chain cryptocurrency markets (e.g., ETH/USD prices). The motivation is to develop methods for proxying off-chain market data using data and computation that is in principle verifiable on-chain and could provide an alternative approach to blockchain price oracles. We explore relationships in PoW mining, PoS validation, block space markets, network decentralization, usage and monetary velocity, and on-chain Automated Market Makers (AMMs). We select key features from these markets, which we analyze through graphical models, mutual information, and ensemble machine learning models to explore the degree to which off-chain pricing information can be recovered entirely on-chain. We find that a large amount of pricing information is contained in on-chain data, but that it is generally hard to recover precise prices except on short time scales of retraining the model. We discuss how even noisy information recovered from on-chain data could help to detect anomalies in oracle-reported prices on-chain.

Complexity vs Empirical Score

• Math Complexity: 5.5/10
• Empirical Rigor: 6.5/10
• Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: This paper presents a novel approach to decentralized price oracles by analyzing the mutual information between on-chain network features and off-chain market prices using graphical models and ensemble ML. While empirically sound and highly original in its application to DeFi resilience, the long-term predictive power remains limited without frequent retraining.

Research Flowchart

  flowchart TD
    A[Research Goal: Proxy Off-Chain Crypto Prices via Verifiable On-Chain Data] --> B[Data Inputs: On-Chain Features<br> Mining, Staking, Gas, Velocity, AMMs]
    B --> C[Feature Analysis: Graphical Models & Mutual Information]
    C --> D[Model Training: Ensemble Machine Learning Models]
    D --> E[Key Finding 1: High pricing info present on-chain; precise recovery needs frequent retraining]
    D --> F[Key Finding 2: On-chain proxies enable anomaly detection for oracle price feeds]