본문영역 바로가기 메인메뉴 바로가기 하단링크 바로가기

KISDI 정보통신정책연구원

KISDI 정보통신정책연구원

검색 검색 메뉴

KISDI News

  • KISDI Develops Semiconductor Market Forecast and Risk Index for the AI Era: Enhancing Export Forecasting and Supply Chain Response

    • Pub date 2026-04-23
    • File There are no registered files.
※ URL(Korean): https://www.kisdi.re.kr/bbs/view.do?bbsSn=114974&key=m2101113055776&pageIndex=1&sc=&sw=

KISDI Policy Report (25-09)
Developing an Issue Analysis and Forecasting Model for the Semiconductor Market Using Structured and Unstructured Data

KISDI Develops Semiconductor Market Forecast and Risk Index for the AI Era: Enhancing Export Forecasting and Supply Chain Response

.................................................................................................................................................
▲ Deep learning-based TFT model outperforms conventional econometric models in predicting semiconductor market turning points
▲ DRAM exports projected to grow faster than the average growth cycle of the past decade, driven by AI and HBM demand
▲ Development of a semiconductor market risk index using policy, diplomatic, supply chain variables, and news text data
▲ Proposal for an early warning system based on an integrated supply chain network combining structured and unstructured data

The Korea Information Society Development Institute (KISDI, President Sangkyu Rhee) recently published its Policy Report (25-09), *Developing an Issue Analysis and Forecasting Model for the Semiconductor Market Using Structured and Unstructured Data*. The report analyzes structural changes and key issues in the semiconductor market and presents a semiconductor export forecasting model together with a market risk index.

The study focuses on the semiconductor industry, where demand has surged with the rapid expansion of AI technologies. It combines a state-of-the-art deep learning algorithm, the Temporal Fusion Transformer (TFT), with news text mining to develop export forecasting models and market, policy, and diplomatic risk indices. It also presents an empirical analysis of semiconductor supply chain networks using actual inter-firm transaction data.

Although the semiconductor industry accounts for approximately 20 percent of Korea's total exports and serves as a strategic sector for economic security, it faces increasingly complex geopolitical risks, including U.S.-China technology competition, tariff measures, and global supply chain restructuring. Under these conditions, conventional econometric models based primarily on historical linear trends have become less effective in capturing rapidly changing market dynamics. The report therefore emphasizes the need for forecasting models that are resilient to structural changes and for real-time risk monitoring systems integrating both structured data, such as export statistics and macroeconomic indicators, and unstructured data, including news text.

Using the deep learning-based TFT model to analyze semiconductor export trends, the study projects that DRAM exports will continue to increase through the first half of 2026. The projected growth trajectory exceeds the average upward slope observed during previous expansion cycles over the past decade, suggesting the possibility of achieving export performance comparable to earlier boom periods. The model also effectively captures the surge in semiconductor demand driven by AI, which conventional econometric models tended to underestimate. Variable importance analysis further shows that the semiconductor market risk index developed in this study provides explanatory power comparable to traditional macroeconomic variables such as exchange rates and GDP.

The news text-based semiconductor market risk index applies a log-normal standardized scale in which 100 represents normal conditions and 130 indicates a high-risk level corresponding to the top one percent of observations. This design enables consistent interpretation across different time periods, industries, and risk factors. An analysis of five major components—policy, diplomacy, market conditions, macroeconomy, and external shocks—shows that risk levels were elevated among major memory and foundry firms during the 2019 memory price correction and Japan's export restrictions. By contrast, during 2024–2025, amid expanding AI demand and tariff-related uncertainty, relatively higher risks were identified among mid-sized and late-entering firms as well as equipment manufacturers and wholesale businesses.

The study also constructed semiconductor supply chain networks for 2016, 2020, and 2024 using actual transaction data provided by NICE Information Service, thereby complementing the limitations of existing news-based network analyses.

Jaeyoung Jang, Research Fellow at KISDI, noted, “The semiconductor industry has entered a new phase driven not simply by business cycles but by the emergence of AI as a new technological paradigm. Future export forecasting and policy formulation should therefore rely on advanced forecasting systems capable of learning structural market changes and emerging risks in real time, rather than extrapolating past trends.”

The report recommends institutionalizing the semiconductor market risk index as a core component of Korea's industrial Early Warning System. Specifically, it proposes establishing a policy coordination mechanism that automatically triggers joint review meetings among the Ministry of Trade, Industry and Energy, the Ministry of Economy and Finance, and the Ministry of Science and ICT when the risk index exceeds 130. It also recommends designating a lead organization, such as the Ministry of Trade, Industry and Energy or the Korea Institute for Industrial Economics and Trade, to oversee the production, verification, and publication of the index, while strengthening data linkages with international supply chain cooperation frameworks.

In addition, the report emphasizes the need to strengthen proactive risk management by integrating the TFT-based export forecasting model with the semiconductor market risk index to create a comprehensive early warning system. Such a system would enable combined analysis with real-sector indicators, including stock prices, spot and futures prices, production, and inventory levels.

The report is available for download from the KISDI website ([www.kisdi.re.kr](http://www.kisdi.re.kr)).