1952年Markowitz 提出平均數－變異數投資組合模型（Mean-Variance Model，簡稱MV 模型）後，開創了投資組合理論的先河，他認為風險與報酬是影響資產配置的兩大因素，其中Markowitz在估計共變異數矩陣時，使用樣本共變異數矩陣模型（Sample Covariance Model）做運算。雖然MV 模型具權威性，但仍存在估計誤差的問題，因此許多共變異數矩陣的估計方法應運而生，包括Litterman and Winklemann（1998）的高盛衰退率共變異數矩陣模型以及Ledoit and Wolf（2003）的單一指數濃縮估計法。本文比較各種共變異數矩陣的效率前緣（efficient frontier）；並採用全域最小變異組合（Global Minimum Variance Point），檢驗樣本共變異數矩陣模型、高盛衰退率共變異數矩陣模型及單一指數濃縮估計法所建構的投資組合，其績效是否優於市值加權的台灣50指數；且以滾動視窗（rolling window）方式，比較三種方法績效之異同優劣。本研究實證結果顯示三種方法相對於大盤均有較佳表現，各方法間則以單一指數濃縮估計法表現較佳。 / Markowitz indicated Mean-Variance Model and initiated the portfolio theory in 1952. He proved that risk and return are two important components to impact on asset allocation, and used sample covariance model to calculate covariance matrix. However, MV model exists estimation error. Therefore, many covariance matrix methods was proposed including Goldman Sachs decay rate covariance matrix model of Litterman and Winklemann（1998）, and shrinkage to single-index covariance matrix method of Ledoit and Wolf（2003）. This study compares the efficient frontier build by different covariance matrix methods. Also, this study adopts global minimum portfolio and rolling window to discuss performance of portfolio constructed by these three methods. The conclusion is that the performance of portfolio constructed by these three covariance matrix methods is better than market index, and shrinkage to single-index covariance matrix is the best method to construct portfolio.

本研究以985筆台灣科技業公司為樣本，並且使用企業投資指標作為指數加權基礎，探討以有形和無形資產投資規模進行基本面指數化的績效表現與可行性。我們發現即使在考慮了價值風險和規模風險之後，以研究發展費用相關指標建構的基本面指數仍可以產生超額報酬。此外，研究結果顯示部分的基本面指數具有市場擇時能力，能避免投資組合績效受到價格不效率的影響。在對樣本進行流動性的篩選以及考慮投資組合的交易成本之後，我們仍得到一樣的結果。 / We employ 985 companies in technology industry in Taiwan to examine the performance and feasibility of the fundamental indices constructed by corporate investment metrics (including both tangible and intangible investment). We find that the fundamental indices constructed by R&D expenditure-related metrics generate significant Fama-French alpha. Besides, evidence shows that parts of the fundamental indices have market timing ability to prevent performance dragged by price inefficiency. We draw a same conclusion after weeding out the companies with low liquidity and adjusting for transaction costs.

本研究以1992年至2017年計19,710筆台灣上市櫃公司為樣本，探討以企業乘數與資本報酬率為指標形成投資組合的績效表現與可行性。研究發現單純以低企業乘數指標組成的投資組合表現最佳，且能長期打敗大盤；以低企業乘數與高資本報酬率的綜合指標形成的投資組合表現次佳；而單純以高資本報酬率為指標形成的投資組合表現最差，且長期劣於大盤表現。研究結果亦包括企業乘數投資組合可以創造出顯著的超額報酬（Alpha），且其優異的表現可以歸因於投資人對於企業的未來盈餘表現預期錯誤，造成市場暫時出現錯誤定價的現象。 / We employ Taiwan’s listed companies from 1992 to 2017 as a sample to examine the performance and feasibility of forming a portfolio based on enterprise multiple and return on invested capital. We find that the portfolio which consists solely of low enterprise multiple stocks outperform the market in the long run; the portfolio formed by composite indicators of low enterprise multiple and high return on invested capital beats the market as well; while the pure high return on invested capital portfolio underperforms. We also find that the low-minus-high enterprise multiple portfolio generates Fama-French alpha, and its excellent performance can be attributed to investors' expectation errors of the company's future earnings performance, resulting in temporary market mispricing.