Common Risk Factors in Decentralized AI Subnets

Battle Count: 50

Highly relevant for quantitative trading in cryptocurrency and DeFi markets. The paper identifies a mechanically derived size premium (1.01%/day SMB) and momentum factor (0.68-0.75%/day) in a novel asset class of 128 decentralized AI subnet tokens. The structural derivation via AMM pricing (Proposition 1) provides a clear economic mechanism, and the halving natural experiment offers a rare causal test. However, the critical finding that the strategy is only implementable below ~$10K AUM due to AMM slippage (Proposition 2) severely limits practical applicability for institutional strategies. The paper is most relevant for small-cap crypto trading, DeFi market-making, and understanding factor dynamics in AMM-priced assets. The positive risk-return relationship (opposite of equity anomalies) and the mechanical size-liquidity link (0.93 correlation) are important for risk model calibration in DeFi.

Moderate complexity. The factor construction follows standard Fama-French methodology (tercile sorts, long-short portfolios, Fama-MacBeth regressions, GRS tests) well-documented in the literature. The AMM slippage calculation (Proposition 2) is straightforward: slippage = Δτ/τ. The halving regression discontinuity is a simple dummy variable regression. The main implementation challenges are: (1) accessing and processing Taostats API data for 128 subnets over 406 days, (2) handling subnet lifecycle events (deregistration/re-registration, startup mode), (3) computing exact AMM slippage for portfolio-level trades across heterogeneous pool depths, and (4) managing the expanding cross-section (63 to 124 subnets). No machine learning or complex optimization is required; the analysis is primarily econometric.

The paper uses publicly available blockchain data from the Taostats API, standard Fama-French methodology, and provides detailed formulas for portfolio construction, return computation, and slippage calculations. The AMM pricing mechanism is fully transparent and deterministic. However, the paper does not provide a GitHub repository or code. The Taostats API is publicly accessible. The methodology follows well-established asset pricing literature (Fama-French 1993, Liu-Tsyvinski-Wu 2022). The halving event is a verifiable on-chain event. One limitation is the use of full-sample betas in Fama-MacBeth regressions, which the author acknowledges introduces look-ahead bias.