EAGER: Gaining Visibility into Supply Network Risks with Large-Scale Textual Analysis

全球化和追求精益生产大大增加了当代供应链的脆弱性。曼谷发生的破坏性事件可能会导致北京的生产停止，进而阻碍波士顿的产品交付。这些事件以各种各样的，有时是意想不到的形式出现：自然灾害，劳工抗议，公用事业中断，网络攻击，政治变化和流行病爆发，仅举几例。它们可能导致连锁供应链故障和业务连续性中断，并可能造成严重的近期和长期物理，财务和声誉后果。有效管理供应链中断的主要障碍是对相互关联的供应网络结构和相关风险状况的可见性有限。例如，美国国防部特别强调，2015年供应链结构和风险的可见性有限，属于高战略风险项目。这个早期概念探索性研究资助（EAGER）项目直接解决了这一挑战。它开发了预测分析，风险学习和缓解策略，可供公司和组织随时部署，以提高其供应链风险的可见性和控制。数据驱动的方法对于具有复杂供应链的组织（如跨国公司和政府机构）特别有用，可以更好地衡量供应网络风险的分布和影响，并积极管理此类风险，提高供应链的弹性。虽然现有的理论假设供应网络结构和相关的风险状况的完美的知识，本研究遵循双管齐下的方法，直接占和减轻有限的可见性。首先，它开发了一系列的实证模型，增加了供应链风险及其驱动因素的可见性，特别强调网络驱动的风险相互依赖性。首次将联合收割机自动文本分析（主题编码和情感分析）与高维统计分析相结合，从大规模定性数据中分离出供应风险信息。具体来说，该项目将量化社交媒体上企业披露和用户生成内容的语言，以1）描述风险分布和相互依赖性，2）量化这些风险对直接和次级供应链合作伙伴的影响，以及3）识别早期预警因素并开发风险事件的预测模型。第二，利用从实证结果中获得的见解，该项目开发了新的风险学习和缓解量化模型，解决并解释了有限的可见性。一类模型关注于在供应网络知识完备但风险知识不完备的情况下的最优风险学习。另一类模型侧重于在一般不完全信息下设计最优风险缓解策略。该模型解决了有效性的直接（采购过剩库存和多源）与间接（供应合同）缓解策略的博弈论框架。

Globalization and the quest for lean production have significantly increased vulnerability of contemporary supply chains. A disruptive event in Bangkok can stop production in Beijing, and in turn hamper product delivery in Boston. Such events materialize in various, sometimes unexpected forms: natural disasters, labor protests, utility outage, cyber-attacks, political shifts, and epidemic outbreaks, to name a few. They can lead to cascading supply chain failures and business continuity interruptions, with potentially severe near- and long-term physical, financial, and reputational consequences. The key barrier to the effective management of supply chain disruptions is the limited visibility into the interconnected supply network structure and the associated risk profiles. For example, the Department of Defense has specifically highlighted limited visibility of supply chain structures and risks as items of high strategic risks in 2015. This EArly-concept Grant for Exploratory Research (EAGER) project directly addresses this challenge. It develops predictive analytics, risk learning and mitigation strategies that can be readily deployed by firms and organizations to increase visibility and control of their supply chain risks. The data driven approach is particularly useful to organizations with complex supply chains, such as multinational firms and governmental agencies, to better measure the distribution and impact of supply network risks, and proactively manage such risks and achieve better supply chain resilience. While existing theories assume perfect knowledge of supply network structure and the associated risk profiles, this research follows a two-pronged approach to directly account for and mitigate limited visibility. First, it develops a series of empirical models that increase visibility into supply chain risks and their driving factors, with a particular emphasis on network-driven risk interdependencies. It is the first research to combine automated textual analysis (topic coding and sentiment analysis) and high-dimensional statistical analysis, to isolate supply risk information from large scale, qualitative data. Specifically, the project will quantify the language of corporate disclosures and user generated content on social media to 1) characterize risk distributions and interdependencies, 2) quantify the impacts of these risks on both immediate and sub-tier supply chain partners, and 3) identify early-warning factors and develop predictive models for risk events. Second, leveraging insights gained from the empirical results, this project develops new quantitative models on risk learning and mitigation that address and account for limited visibility. One class of models focuses on optimal risk learning given complete knowledge of supply network but incomplete knowledge of risks. The other class of models focuses on designing optimal risk mitigation strategies with general incomplete information. The model addresses effectiveness of direct (procuring excess inventory and multi-sourcing) versus indirect (supply contracts) mitigation strategies in a game-theoretic framework.