ML should target volatility
James Ross of SynthdataCo argued that volatility modeling is the key primitive for AI-native finance, recommending ML efforts prioritize volatility as the foundation across equities and crypto rather than surface-level signals. That thesis reframes risk modeling as the central lever for robust automated strategies. (x.com)
Synth operates as Bittensor Subnet 50 (SN50) and markets itself as a predictive engine that generates synthetic price-path data and probability distributions focused on asset volatility. (subnetalpha.ai) James Ross is featured as Synth’s founder/operator in multiple interviews where he describes the subnet’s architecture for forecasting volatility with Monte Carlo-style simulations and continual 24‑hour volatility models run by competing miners. (youtube.com) Presentations and podcast episodes with Ross detail commercial routes: on‑chain propagation of forecasts into prediction markets, API integrations and monthly asset expansions, and prop trading on options venues; supported assets mentioned include BTC, ETH, SOL and gold. (youtube.com) A May 16, 2025 arXiv paper on foundation time‑series models explicitly evaluates realized volatility forecasting and concludes that large cross‑series time‑series models can improve volatility prediction when fine‑tuned. (arxiv.org) Peer‑review and industry analyses show mixed results: an IEEE study reported cases where deep learning provided up to double the economic gains versus HAR‑style volatility baselines, while other applied comparisons find a well‑tuned HAR model remains competitive in many settings. (ieeexplore.ieee.org)
