本研究以台灣股票型共同基金持股資料為基礎，探討矩陣分解推薦系統方法在股票推薦與投資應用上的可行性與表現。研究將基金視為使用者、股票視為項目，以基金持股建構互動矩陣，並設計三種輸入規格，原始持股比例、報酬調整持股與主動調整幅度，搭配四個模型（Explicit MF、Explicit MF + SI、Implicit MF、Implicit MF + SI），共 12 組模型—輸入組合，分別從推薦排序指標與投資組合績效兩個層面進行系統性比較。實證期間涵蓋 2006 年第 1 季至 2025 年第 2 季，共計 78 季。
研究結果顯示三項主要發現。第一，隱性回饋模型在推薦排序指標上全面優於顯性回饋模型，NDCG@10 高出約 26.7%、ROC_AUC 差距達 75.6%，驗證了以 binary 持有訊號為預測目標、持股比例為信心權重的建模方式，在基金持股資料上更為貼切。第二，在投資績效層面，所有 12 組組合——包含原先預期最具前瞻性的主動調整幅度——均未能產生統計上顯著的多空（Q5−Q1）報酬；經 Newey-West 序列相關校正後，最高的 t 統計量絕對值僅約 1.4，且此一全面不顯著的結論對投資組合的聚合方式亦具穩健性。第三，推薦排序能力與投資績效之間存在明顯背離：推薦表現最佳的隱性模型，其投資績效並不顯著，顯示模型準確還原基金持股偏好的能力，未必代表該偏好對未來報酬具有預測力。此一背離對推薦系統於金融領域的應用具有方法論上的警示意義：僅以推薦指標優化模型未必能改善投資績效，兩個層面的表現必須分別檢視。

This study examines the feasibility and performance of matrix factorization recommender systems for stock selection and investment application, using the equity holdings of Taiwanese equity mutual funds. Treating funds as users and stocks as items, we construct a fund–stock interaction matrix and design three input specifications—the raw holding weight, the return-adjusted holding weight, and active rebalancing—paired with four models (Explicit MF, Explicit MF + SI, Implicit MF, and Implicit MF + SI), yielding 12 model–input combinations that are systematically compared along two dimensions: recommendation ranking metrics and investment portfolio performance. The sample spans 2006Q1 to 2025Q2, comprising 78 quarters.
The study reports three main findings. First, implicit-feedback models substantially outperform explicit-feedback models on recommendation ranking, with NDCG@10 higher by approximately 26.7% and a ROC_AUC gap of 75.6%, confirming that treating the binary holding signal as the prediction target while using the holding weight as a confidence weight is better suited to fund holdings data. Second, in terms of investment performance, none of the 12 combinations—including active rebalancing, the input expected ex ante to be the most forward-looking—generates a statistically significant long–short (Q5−Q1) return; after a Newey-West correction for serial correlation, the largest absolute t-statistic is only about 1.4, and this uniformly insignificant result is robust to the portfolio aggregation scheme. Third, recommendation ranking ability and investment performance diverge: the implicit models with the strongest recommendation performance deliver insignificant investment returns, indicating that a model's ability to accurately recover fund holding preferences does not imply that those preferences predict future returns. This divergence carries a methodological caution for applying recommender systems in finance: optimizing models on recommendation metrics alone need not improve investment performance, and both dimensions must be examined separately.

Agarwal, V., Gay, G. D., & Ling, L. (2014). Window dressing in mutual funds. Review of Financial Studies, 27(11), 3133–3170.
Agarwal, V., Mullally, K. A., Tang, Y., & Yang, B. (2015). Mandatory portfolio disclosure, stock liquidity, and mutual fund performance. Journal of Finance, 70(6), 2733–2776.
Chen, H.-L., Jegadeesh, N., & Wermers, R. (2000). The value of active mutual fund management: An examination of the stockholdings and trades of fund managers. Journal of Financial and Quantitative Analysis, 35(3), 343–368.
Coval, J., & Stafford, E. (2007). Asset fire sales (and purchases) in equity markets. Journal of Financial Economics, 86(2), 479–512.
Cremers, K. J. M., & Petajisto, A. (2009). How active is your fund manager? A new measure that predicts performance. Review of Financial Studies, 22(9), 3329–3365.
Daniel, K., Grinblatt, M., Titman, S., & Wermers, R. (1997). Measuring mutual fund performance with characteristic-based benchmarks. Journal of Finance, 52(3), 1035–1058.
Gabaix, X., Koijen, R. S. J., Richmond, R. J., & Yogo, M. (2025). Asset embeddings (NBER Working Paper No. 33651). National Bureau of Economic Research. https://doi.org/10.3386/w33651
Grinblatt, M., & Titman, S. (1989). Mutual fund performance: An analysis of quarterly portfolio holdings. Journal of Business, 62(3), 393–416.
Grinblatt, M., & Titman, S. (1993). Performance measurement without benchmarks: An examination of mutual fund returns. Journal of Business, 66(1), 47–68.
Hung, P.-H. (2025). Window dressing, fund performance, and fund flows: Insights into Taiwan's equity mutual funds. Asia-Pacific Journal of Financial Studies, 54(4). https://doi.org/10.1111/ajfs.70012
Kacperczyk, M., Sialm, C., & Zheng, L. (2005). On the industry concentration of actively managed equity mutual funds. Journal of Finance, 60(4), 1983–2011.
Kacperczyk, M., Sialm, C., & Zheng, L. (2008). Unobserved actions of mutual funds. Review of Financial Studies, 21(6), 2379–2416.
Lakonishok, J., Shleifer, A., Thaler, R., & Vishny, R. (1991). Window dressing by pension fund managers. American Economic Review, 81(2), 227–231.
Lou, D. (2012). A flow-based explanation for return predictability. Review of Financial Studies, 25(12), 3457–3489.
Musto, D. K. (1999). Investment decisions depend on portfolio disclosures. Journal of Finance, 54(3), 935–952.
Satone, V., Desai, D., & Mehta, D. (2021). Fund2Vec: Mutual funds similarity using graph learning. In Proceedings of the Second ACM International Conference on AI in Finance (ICAIF '21) (Article 16, pp. 1–8). Association for Computing Machinery. https://doi.org/10.1145/3490354.3494381
Wermers, R. (2000). Mutual fund performance: An empirical decomposition into stock-picking talent, style, transactions costs, and expenses. Journal of Finance, 55(4), 1655–1695.
Wermers, R., Yao, T., & Zhao, J. (2012). Forecasting stock returns through an efficient aggregation of mutual fund holdings. Review of Financial Studies, 25(12), 3490–3529.
池祥萱、林煜恩、周賓凰（2007）。基金績效持續與聰明錢效果：台灣實證。管理學報，24(3)，307–330。
莊惠菁、管中閔（2020）。共同基金卓越績效的認定與評估：新逐步檢定法的應用。證券市場發展季刊，32(1)，1–32。
陳獻儀、楊淑玲、張眾卓、陳婉榕（2014）。基金績效歸屬與基金流量之關聯性。臺大管理論叢，24(2)，283–308。
盧秋玲、張清發、沈仰斌（2017）。積極比例是否解釋台灣共同基金績效？證券市場發展季刊，29(3)，1–38。