When Richer Information Does Not Improve Allocation: Machine Learning and Portfolio Optimisation in NFT-Ecosystem Tokens

Abstract Whether richer information systematically improves portfolio allocation remains an open question in extreme-risk digital asset markets. This paper examines whether combining machine-learning-based return forecasts with intraday-informed volatility estimation enhances portfolio performance in NFT-ecosystem tokens. While prior evidence in traditional equity markets suggests that integrating predictive models with OHLC-based modified mean–variance (MMV) optimization can improve risk-adjusted returns, it is unclear whether such gains survive in highly volatile and non-stationary environments. Using a two-stage framework that separates return prediction from allocation, we compare classical mean–variance (MV), OHLC-based MMV, and naïve diversification strategies for portfolios of NFT-ecosystem tokens. Our results show that machine learning forecasts generate substantial economic value; however, incorporating intraday volatility into the optimization stage does not yield systematic improvements in risk-adjusted performance. Instead, the benefits of refined risk estimation are conditional on the portfolio’s correlation structure and the forecasting architecture employed. Overall, portfolio performance in NFT-ecosystem token markets is driven primarily by return predictability rather than increasingly sophisticated volatility modeling. The findings highlight a fundamental distinction between statistical improvements in risk measurement and their economic usefulness in allocation decisions in highly unstable financial markets.