Cryptocurrency Valuation: An Explainable AI Approach

Explainability matters in financial AI: The paper demonstrates that effective financial prediction models should be grounded in established economic theory (in this case, monetary theory) rather than being "ad hoc" black boxes, making their recommendations understandable and trustworthy to users.

Transparency through open-source release: The researchers distributed their trading algorithms as open-source software via Python Package Index, exemplifying responsible AI practice by allowing independent verification and scrutiny of their methods rather than keeping them proprietary.

Long-term vs. short-term predictability differs: The study found that while traditional valuation ratios had "little predictive power for short-term bitcoin returns," their proposed price-to-utility (PU) ratio "effectively predicts long-term bitcoin returns," suggesting that AI models may have different reliability depending on prediction timeframes.

Blockchain data enables new accountability: By utilizing "unique blockchain accounting methods" and the Bitcoin UTXO model, the researchers created a valuation approach based on verifiable on-chain data rather than opaque market signals, potentially reducing manipulation and improving model interpretability.

Practical validation of AI recommendations: The paper tested their model through an "automated trading strategy" that outperformed conventional approaches, demonstrating that explainable AI isn't just theoretically sound—it can produce measurable real-world results.