TLDR: A new research paper introduces TLJD, a novel AI model that predicts city-level Foreign Direct Investment (FDI) by analyzing judicial data from over twelve million adjudication documents. This approach offers a more reliable alternative to traditional economic data, which can be prone to manipulation. TLJD builds an index system of 380 judicial performance indicators, transforms them into a tabular dataset, and uses a Mixture of Experts model to predict FDI. Experiments show TLJD outperforms existing methods, demonstrating the effectiveness of judicial data and revealing regional variations in how judicial performance influences investment decisions.
A new research paper introduces a novel approach to predicting city-level Foreign Direct Investment (FDI) by leveraging judicial data, offering a more reliable alternative to traditional economic indicators. This method, called Tabular Learning on Judicial Data (TLJD), aims to support local governments in making more informed economic decisions, aligning with the United Nations Sustainable Development Goal of promoting sustained economic growth.
Historically, city-level FDI predictions have relied heavily on economic data such as GDP. However, such data can be susceptible to manipulation, leading to less accurate forecasts and potentially misleading policymakers. To address this critical issue, the researchers propose using large-scale judicial data, which provides a transparent and verifiable source of information reflecting local investment security and returns.
The core of this innovative approach involves building a comprehensive index system for evaluating judicial performance. This system comprises 380 indicators, categorized into four key types: Procedural Justice (fairness of decision processes), Distributive Justice (fairness of decision outcomes), Judicial Efficiency (efficiency of judicial decisions), and Judicial Characteristics (other features of judicial performance). These indicators are derived from over twelve million publicly available adjudication documents from China, covering the period from 2016 to 2019. The unstructured judicial documents are transformed into a structured tabular dataset, where each row represents a city in a specific year and contains the values of these 380 indicators.
The TLJD method itself is designed with two main components: Indicator Feature Encoding and City-Level FDI Prediction. The Indicator Feature Encoding module transforms the numerical judicial performance indicators into embeddings, capturing complex relationships between them. It uses both a row encoder to describe each sample and a column encoder to map global similarities between indicators. These are then fused and processed by transformer layers with arithmetic attention to incorporate contextual information.
For the City-Level FDI Prediction, TLJD employs a Mixture of Experts (MoE) model. This model consists of four specialized expert models, each focusing on one of the four judicial performance indicator types. A gating network dynamically assigns weights to these expert models, allowing the system to adjust the importance of different judicial aspects based on regional variations. The final FDI prediction is a weighted sum of the outputs from these expert models.
The effectiveness of TLJD was rigorously validated through extensive experiments on two real-world application scenarios: cross-city prediction (estimating missing historical FDI data) and cross-time prediction (forecasting future FDI). The results demonstrated TLJD’s superior performance compared to ten state-of-the-art baselines, achieving high accuracy with R2 scores reaching at least 0.92. This highlights the significant potential of using judicial data for accurate city-level FDI prediction.
An important finding from the research is the analysis of expert weights, which revealed regional variations in how judicial performance influences FDI. For instance, the Procedural Justice expert showed greater influence in China’s eastern coastal regions, while the Distributive Justice expert was more dominant in central, western, and northeastern regions. Furthermore, cities with lower FDI tended to have a larger average weight for the Distributive Justice expert, suggesting investors in these areas prioritize practical judicial outcomes. Conversely, cities with higher FDI showed a greater emphasis on Procedural Justice, indicating that investors in well-established investment environments value the fairness and transparency of judicial procedures more.
Also Read:
- AI Enhances Startup Investment Decisions with Proven Framework
- Advanced AI Model Enhances Long-Term Stock Market Forecasting
This research marks a significant step forward by introducing judicial data as a reliable and practical alternative for FDI prediction, addressing the limitations of traditional economic data. The methodology and findings presented in this paper, available at https://arxiv.org/pdf/2507.05651, have potential applications beyond China, contributing to global economic growth and sustainable development goals.
