Paper Type
Short
Paper Number
PACIS2026-1264
Description
As geopolitical tensions and supply chain disruptions intensify, integrated circuit (IC) enterprises face unprecedented uncertainty, making accurate performance prediction critical and challenging. Existing approaches predominantly rely on historical financial time series and struggle to capture exogenous shocks and their cascading effects across firms. This study proposes EvoChain, a predictive framework that integrates evolutionary event graphs (EEGs) with supply chain network structure for firm-level performance forecasting. EvoChain extracts EEGs from IC industry news using NER and large language model–based relation extraction, and adopts a layered heterogeneous graph design: a GCN encodes each firm's local EEG subgraph into an event-evolution embedding, which then serves as the node feature for a GAT over supply chain network. An attention-based gating mechanism integrates these representations with historical financial sequences for jointly predicting four financial metrics. On 317 Chinese IC firms over 2013–2023, EvoChain outperforms baselines, and ablation confirms heterogeneous module contributions across metrics.
Recommended Citation
Yue, Chenchen and Geng, Ruibin, "Forecasting Enterprise Performance in the IC Industry: A Framework Combining Evolutionary Event Graphs and Supply Chain Propagation" (2026). PACIS 2026 Proceedings. 3.
https://aisel.aisnet.org/pacis2026/data_analtyics/data_anltics/3
Forecasting Enterprise Performance in the IC Industry: A Framework Combining Evolutionary Event Graphs and Supply Chain Propagation
As geopolitical tensions and supply chain disruptions intensify, integrated circuit (IC) enterprises face unprecedented uncertainty, making accurate performance prediction critical and challenging. Existing approaches predominantly rely on historical financial time series and struggle to capture exogenous shocks and their cascading effects across firms. This study proposes EvoChain, a predictive framework that integrates evolutionary event graphs (EEGs) with supply chain network structure for firm-level performance forecasting. EvoChain extracts EEGs from IC industry news using NER and large language model–based relation extraction, and adopts a layered heterogeneous graph design: a GCN encodes each firm's local EEG subgraph into an event-evolution embedding, which then serves as the node feature for a GAT over supply chain network. An attention-based gating mechanism integrates these representations with historical financial sequences for jointly predicting four financial metrics. On 317 Chinese IC firms over 2013–2023, EvoChain outperforms baselines, and ablation confirms heterogeneous module contributions across metrics.
Comments
05-DataAnalytics