Price prediction models for alternative investments

The rise in interest in alternative investments with potential diversification and higher returns is driven by the limitations of traditional financial markets and the evolving needs of modern investors. In this doctoral thesis, we investigate pricing mechanisms for two alternative investing sectors: Environmental, Social, and Governance (ESG) investing, focusing on environmentally responsible publicly traded equities, and digital blockchain-based assets, using a unified methodological framework. The first part of this study explores the financial implications of Corporate Environmental Responsibility (CER) within S&P 500 firms over a fifteen-year period. We assess the impact of environmental performance on market valuation using risk-adjusted metrics based on a CAPM model under the efficient market hypothesis. The analysis, supported by panel data techniques including OLS and 3SLS estimation, introduces the concept of a “Green Premium”—a measurable trade-off between environmental responsibility and investor returns. Findings reveal that while CER correlates positively with operational metrics, such as sales and profit, it does not improve stock performance, suggesting a persistent market discount for environmental stewardship. The second part of the thesis shifts focus to blockchain finance, offering a comprehensive examination of the price behavior of Non-Fungible Tokens (NFTs) through modern machine learning approaches. Data extracted directly from blockchain platforms and transformed through advanced feature engineering are used to evaluate the predictive performance of Random Forest, XGBoost, and Multilayer Perceptron models. The final dataset was the output of an advanced feature transformation pipeline. The dataset was enriched by domain-informed manual feature synthesis and Deep Feature Synthesis, reshaped by Principal Component Analysis (PCA), and refined through Feature selection techniques implementation for optimal interpretability and dimensionality reduction. Of the models tested, XGBoost demonstrated the highest predictive accuracy, with historical pricing data emerging as the most influential predictor. Together, the two parts of this thesis contribute to expanding the literature on non-traditional asset classes by applying strong analytical methods across distinct investment domains. The findings provide sector-specific insights into what drives asset values and establish a transferable methodological framework for forecasting asset prices in emerging and heterogeneous markets.