Fuzzy Inference for Assessing Hydro-financial Risk inInstitutional Equity Holdings: A Decision Support ToolFramework

Sammanfattning

Water scarcity is a growing concern for institutional investors with long horizons and globally diversified portfolios. Freshwater is a fundamental input for economic production and value creation, yet water is also finite and unevenly distributed, and increasingly stressed by climate change, population growth, and ecosystem degradation. The financial consequences operate through operational disruption, increasing investment costs, business disruption and stranded assets, regulatory pricing and reputational exposure. Water-related risks are characteristically nonlinear, location specific and difficult to anticipate based on historical data, which is why traditional econometric methods struggle to assess this issue. The data available to institutions is largely linguistic or semi-quantitative, produced under different methodologies, and treating the data as crisp point estimates discards the underlying uncertainty. The OECD and TNFD have called for institutions to develop internal analytical capability that can operate under these conditions rather than postpone risk assessment. This thesis tries to fill this gap by developing a rule-based Interval Type-2 fuzzy inference framework that combines water-stress exposure, sectoral water dependency, and relative financial performance into an interpretable portfolio-level risk signal, the Hydro-Financial Risk Signal. Linguistic boundaries are calibrated empirically through fuzzy c-means clustering, and research derived knowledge from the data, and the IT2 system propagates uncertainty through to the output. The framework is applied to Andra AP-fonden’s (AP2’s) listed equity portfolio of scored 1235 holdings, representing 192.1 billion SEK. Sector and within sector patterns are consistent with independent ENCORE-based studies. The main purpose of the framework is to act as a decision support tool for further analysis and assessment of holdings.