本研究旨在運用混頻面板資料（Mixed-frequency Panel Data）與結構化機器學
習方法，探討台灣半導體產業每股盈餘（EPS）與毛利率（Gross Margin）之即時預
測（Nowcasting）問題。隨著全球經濟與政治環境快速變化，半導體產業作為數位
經濟核心，其財務表現對市場投資與政策制定具有高度影響力。然而，傳統財務預
測方法多依賴低頻歷史財報資訊，難以及時反映當期經濟環境與產業動態，亟需具
備即時性與高預測準確度之方法加以改進。
本研究運用稀疏群組正則化技術，處理高維結構化的混頻資料，並納入台灣上
市企業特有之月營收資料，結合宏觀經濟變數，包括利率、匯率、通膨率與工業生
產指數等，建構可即時預測 EPS 與毛利率之模型。月營收資料作為台灣企業每月
強制公告之高頻營運資訊，能有效補強傳統財務報表資訊之時效性不足，提升模型
對企業營運動態之即時掌握能力。
實證部分以台灣半導體產業為研究對象，評估不同變數組合對 EPS 與毛利率
預測效果之影響，並與傳統 AR(1) 基準模型進行比較，檢驗所提模型之預測效能。
研究結果顯示，納入月營收資料與總體經濟變數後，能顯著提升 EPS 與毛利率預
測準確度，提供更具時效性與實務參考價值之預測資訊。整體成果除具學術創新意
涵，亦可協助投資人、企業管理者及政策制定者在快速變動的市場環境中，做出更
為前瞻且有效之決策。

This study applies mixed-frequency panel data and structured machine learning
techniques to nowcast two key financial indicators—earnings per share (EPS) and gross
margin—for Taiwan’s semiconductor industry. In an era of rapidly changing global
economic and political environments, the financial performance of the semiconductor
sector, a cornerstone of the digital economy, plays a critical role in shaping investment
strategies and policy decisions. However, conventional forecasting methods relying on
low-frequency historical financial reports often fail to capture real-time economic shifts
and industry dynamics, highlighting the need for more timely and accurate predictive
models.
This research leverages sparse-group regularization techniques to handle high
dimensional, structured mixed-frequency data. In particular, it incorporates the unique
monthly revenue disclosures mandated for listed companies in Taiwan, combined with
key macroeconomic variables such as interest rates, exchange rates, inflation rates, and
industrial production indices, to construct models for real-time EPS and gross margin
forecasting. Monthly revenue data, as high-frequency operational information, effectively
enhances the model’s ability to track current business performance and respond to market
changes in a timely manner.
An empirical analysis is conducted on Taiwan’s semiconductor industry to evaluate the
impact of different variable combinations on the predictive performance for EPS and
gross margin. The proposed models are also benchmarked against traditional AR(1)
models to assess forecasting accuracy. The results demonstrate that incorporating monthly
revenue data and macroeconomic indicators significantly improves predictive accuracy,
providing more timely and practical insights. The findings offer theoretical contributions
to the academic literature and valuable guidance for investors, corporate managers, and
policymakers in making forward-looking and effective decisions in a dynamic market
environment.

王薪棉（2023）。《台灣電子產業鏈概況》。台北：台北外匯市場發展基金
會委託研究計畫。
Andreou, E., Ghysels, E., & Kourtellos, A. (2010). Regression models with
mixed sampling frequencies. Journal of Econometrics, 158(2), 246–261.
Babii, A., Ball, R. T., Ghysels, E., & Striaukas, J. (2023). Panel data
nowcasting: The case of price-earnings ratios. arXiv preprint
arXiv:2307.02673.
Babii, A., Ghysels, E., & Striaukas, J. (2021). Machine learning mixed
frequency models. CEPR Discussion Paper No. DP16532.
Babii, A., Ghysels, E., & Striaukas, J. (2022). Machine learning panel data
models with mixed frequencies. arXiv preprint arXiv:2203.11172.
Ball, R., & Ghysels, E. (2018). Automated earnings forecasts: Beat analysts or
combine and conquer? Management Science, 64(4), 1574–1594.
Bown, C. P., & Wang, E. (2024). Semiconductors and modern industrial policy.
Peterson Institute for International Economics Working Paper 24-1.
Chudik, A., & Pesaran, M. H. (2011). Common correlated effects estimation
of heterogeneous dynamic panel data models with weakly exogenous
regressors. Journal of Econometrics, 165(1), 68–90.
Foroni, C., Marcellino, M., & Schumacher, C. (2015). U-MIDAS: MIDAS
regressions with unrestricted lag polynomials. Journal of the Royal Statistical
Society: Series A (Statistics in Society), 178(1), 57–82.
[10] Ghysels, E., Santa-Clara, P., & Valkanov, R. (2006). Predicting volatility:
Getting the most out of return data sampled at different frequencies. Journal
of Econometrics, 131(1–2), 59–95.
[11] Ghysels, E., Sinko, A., & Valkanov, R. (2007). MIDAS regressions: Further
results and new directions. Econometric Reviews, 26(1), 53–90.
[12] Liu, J. S., & Weng, Y. C. (2016). On predicting the semiconductor industry
cycle: A Bayesian model averaging approach. Applied Economics, 48(30),
2847–2863.
[13] Liu, W.-H., & Weng, S.-S. (2016). On predicting the semiconductor industry
cycle: A Bayesian model averaging approach. Empirical Economics.
[14] Simon, N., Friedman, J., Hastie, T., & Tibshirani, R. (2013). A sparse-group
lasso. Journal of Computational and Graphical Statistics, 22(2), 231–245.