| < Back to Master Index | Next: Chapter 1 > |
A primary manifestation of long-term climate change is the increased intensity, frequency, and duration of extreme weather events. This trend already poses a significant negative impact on global agriculture, thereby leading to a long-term threat to worldwide food security.
Extreme weather events will not only impact the physical world; instead, they will also affect economic and financial markets through a complex transmission pathway. This transmission will firstly be reflected in the shock of commodity futures. Then, it will impact the financial performance of corporations in the related field and ultimately influence the stock price and credit market.
Existing research tended to investigate these two aspects in isolation, and showing a polarization, one group focuses on the physical impact of weather-related risk on agricultural yield, another group only analyses the correlations of these events with financial market, this broken chain in study leads to limited understanding of the complete picture of how physical shock transferred to systematically economic and financial risk.
This study aimed to address the four core research gaps identified in the current literature.
This study put efforts into bridging the methodology gap by constructing highly homogeneous spatial-temporal units to improve the accuracy and reliability of the cross-region comparison at a fundamental level. In addition, we addressed the challenge of interaction and non-linear gap by implementing an advanced machine learning model deeply, investigating the interaction and threshold effects of compound events using a data-driven method, and overcoming the linear assumption limitation of the traditional statistical model.
Furthermore, this research broke through the integration gap by building a systematic quantitative framework, bridging the yield response signal in the physical world and the pricing mechanism of the financial market. Finally, this study addressed the explainability gap by adopting an XGBoost model combined with SHAP values, to open the machine learning “black-box”, quantify the global- and local-level contributions of each feature, thus considerably improving the explainability of the model.
This research provides a theoretical basis and quantitative tools for decision-makers to develop and implement climate adaptation and mitigation strategies.
For example: