增強式學習（Reinforcement Learning）透過與環境不斷的互動來學習，以達到極大化每一期報酬的總和的目標，廣泛被運用於多期的決策過程。基於這些特性，增強式學習可以應用於建立需不斷動態調整投資組合配置比例的動態資產配置策略。 本研究應用Deep Q-Learning演算法建立動態資產配置策略，研究如何在每期不同的環境狀態之下，找出最佳的配置權重。採用2007年7月2日至2017年6月30日的美國中大型股的股票ETF及投資等級的債券ETF建立投資組合，以其日報酬率資料進行訓練，並與買進持有策略及固定比例投資策略比較績效，檢視深度增強式學習在動態資產配置適用性。 / Reinforcement learning learns by interacting with the environment continuously, in order to achieve the target of maximizing the sum of each return. It has been used to solve multi-period decision making problem broadly. Because of these characteristics, reinforcement learning can be applied to build the strategies of dynamic asset allocation which keep reallocating the mix of portfolio consistently. In this study, we apply deep Q-Learning algorithm to build the strategies of dynamic asset allocation. Studying how to find the optimal weights in the different environment. We use Large-Cap, Mid-Cap ETFs and investment-grade bond ETFs in the U.S. to build up the portfolio. We train the model with the data of daily return, and then we measure its performance by comparing with buy-and-hold and constant-mix strategy to check the fitness of deep Q-Learning.

京都大学 / 0048 / 新制・課程博士 / 博士(工学) / 甲第18582号 / 工博第3943号 / 新制||工||1606(附属図書館) / 31482 / 京都大学大学院工学研究科建築学専攻 / (主査)教授 加藤 直樹, 教授 門内 輝行, 教授 神吉 紀世子 / 学位規則第4条第1項該当 / Doctor of Philosophy (Engineering) / Kyoto University / DFAM

京都大学 / 新制・課程博士 / 博士(工学) / 甲第25275号 / 工博第5234号 / 京都大学大学院工学研究科建築学専攻 / (主査)教授 三浦 研, 教授 金多 隆, 教授 牧 紀男 / 学位規則第4条第1項該当 / Doctor of Philosophy (Engineering) / Kyoto University / DFAM

科学研究費補助金 研究種目:一般研究(B) 課題番号:01470008 研究代表者:平尾 公彦 研究期間:1989-1990年度

本文研究重點係企業併購法之資產收購與公司法第一八五條之爭議問題。首先對公司法第一八五條作性質上之認定，並論及公司法第一八五條本身要件之差異，及該條與企業併購法第二十七條之適用問題。其次，亦對事實合併原則加以討論，該原則之重點在於強調交易所產生之經濟效果，而非交易之名稱，當重大資產交易所產生之經濟效果已與合併之效果相同時，即應賦予股東相關之權利，而不能再以單純之資產買賣視之；至於資產收購與公司分割之比較而言，在法律設計日趨多元化與當事人以特約約定之情形下，資產收購與公司分割間之差距日漸縮小。 / 再，本文同時探討公司進行重大交易時決策權力之配置，先說明一般合併時之權力配置問題，復討論在特殊合併型態時公司決策權力配置之問題，另分析於公司讓與主要部分營業或財產時，美國法與我國法在公司內部決策權力配置之異同，並探討不同決策權力配置之設計所可能產生之優、缺點。復以主要部分營業或財產之認定標準為研究重心，在公司讓與主要部分之營業或財產時，通常引發之爭議即該交易究為單純之資產買賣，係公司一般日常業務，僅由公司董事會決定即可，或已構成資產收購而產生與合併類似之效果，故應賦予公司股東相當於合併時得主張之否決權與股份收買請求權？又因賦予股東相關之決議權與股份收買請求權，對公司而言，通常會增加合併之成本，例如召開股東會之相關費用，與股東行使股份收買請求權時應給付予股東之現金，將使公司現金流量縮減而可能產生失敗之結果，故公司多希望規避相關法規之程序，而不願意將該交易行為認定為合併，以減少現金支出，惟此種解釋雖能顧及公司經營之效率，卻對股東權益保障不足，使股東失去公平收回投資之機會，是本文期能透過美國與我國在相關實務見解與學說之綜合分析，產生一可資參考之認定標準。 / 最後，本文欲討論公司讓與主要部分之營業或財產後之法律效果，主要針對收購公司是否應承擔被收購公司債務之問題進行相關之分析與檢討，在交易相對人不同之認知下，公司若未遵循公司法第一八五條所進行之交易效力問題，簡言之，在公司從事公司法第一八五條之重大交易時，應如何劃定股東之權益維護與相對人之交易安全間之平衡點。

本研究探討長期投資人之最適資產配置問題，並著重於通貨膨脹風險之分析。第一部份討論確定提撥退休金制度下，機構投資人或高所得自然人如何擬定投資策略規避通貨膨脹風險，達到極大化期末財富效用期望值。此研究擴展Battocchio與Menoncin (2004)所建構資產模型，不僅探討市場風險，亦考量通貨膨脹不確定性與基金費用誘因、下方風險保護兩機制，研究對資產配置行為之影響，並依動態規劃方法求得投資策略公式解。第二部份則強調下方風險之重要性，檢視在最低保證收益下，長期投資人跨期資產配置之財富管理議題，並回顧Deelstra et al.(2003)之模型架構，依平賭方法求得投資策略公式解，研究結果顯示基金投資策略可表示為最適CRRA(γ,T)型態共同基金與最低收益避險之組合。另一方面，如何估計通貨膨脹風險亦為本文強調之重點。Campbell和Viceira (2001)首次納入通貨膨脹風險並探討跨期投資議題，結論市場缺乏通貨膨脹連動投資標的時，投資人將減碼長期債持有比例。Brennan和Xia (2002)假設通貨膨脹率服從Ornstein-Uhlenbeck過程，結論投資人之避險需求隨持有債券到期日與投資期限改變。但以上結論未將通貨膨脹學習機制納入模型，因此，在第三部份提出依學習機制修正之投資策略可顯著增加財富效用，並分析在不同參數設定下，學習機制對於期末財富效用之影響。 / In this study, we study three essays of asset allocation problem for long term investors, which means that in this discourse we emphasis the importance of inflation risk. In the first topic, we derive the dynamic optimal investment strategy of the defined contribution pension schemes which include two mechanisms of partial floor protection and incentive fees and their benchmarks. We find investors should hold high proportion of stock index fund to hedge the inflation risk; moreover, the ratio of incentive fees to the setting of benchmark will change the optimal investment trend of underlying assets. In the second topic, we introduce the optimal investment portfolio with minimum guarantees and show that the fund manager should adjust the optimal weights of underlying assets with the ratio of the guarantee fund's value to the value of fund. Finally, this work focuses on how to precisely predict the dynamics of inflation rate. We apply learning method to adjust the prediction of inflation process and we use numerical analysis to study the effect of learning mechanism under different parameter setting.

本論文包含台灣壽險業資產配置國際化及匯率避險兩個主題。首先，探討台灣壽險業積極向主管機關申請核准提高國外資產配置比率，與美國投資人偏好投資自己國家資產不一致的現象，是台灣壽險業資產配置不得不的策略，或是國際化的迷思? 以25家壽險公司2004 年至2008年財務資料實證結果發現，國外資產納入投資組合對壽險業投資績效有利。如果將樣本公司依據所有權區分為本資公司與外資公司，資料顯示，本資公司國際資產配置較為積極，惟其報酬績效與外資公司差異並不顯著。研究亦發現，2008年美國次貸風暴顯著負向影響台灣壽險公司國外投資報酬，即提高國外資產配置雖可提高報酬，惟匯率風險、信用風險及系統風險暴露亦相對提高，建議壽險公司於追球較高報酬同時，應同時加強風險管理。另實證亦發現，資產規模愈大公司之投資報酬率相對較遜，建議於追求保費市占率成長時，應重視投資報酬績效的實質提升。 第二部分探討2004年至 2008年台灣壽險業國外投資匯率風險管理策略對投資績效影響。以整體產業觀察，匯率避險對投資報酬率有正面效果；本資公司避險策略相較外資公司積極，報酬率亦相對較優；股票上市公司有財報揭露股價波動之壓力，經理人有較強誘因採取避險策略，投資報酬率相較優於股票未上市公司，惟差異並不顯著。實證結果支持Glen and Jorion (1993) and Campbell et al. (2010)避險可以降低匯率風險提升投資報酬績效之研究結論，2006年實施之34號會計公報，顯著影響本資公司與上市公司之避險行為。 / In this study, we study two essays on international asset allocation and the currency hedging problem for Taiwan life insurer industry. In the first essay, we investigate the high percentage of foreign investments placed by Taiwan life insurers and how this phenomenon is at odds with the bias for investing at home common among American investors. The holdings of 25 Taiwan life insurance companies, between the years 2004 and 2008, are scrutinized with a view towards evaluating home bias and its financial impact. We find that foreign investment has proven profitable for the life insurance industry. However, if the life insurance industry is divided into two categories according to its ownership structure, i.e., domestic-owned and foreign-owned companies, and that while the performance of investments made by domestic-owned life insurers differs from that of foreign-owned life insurers, the difference is insignificant. We also found that global financial turmoil in 2008 had a massively negative impact on the foreign investments of Taiwan life insurance companies and firm size and return on investment is negative correlated, suggesting that life insurers should focus on enhancing investment performance and risk management. In the second essay, we examine the currency hedging strategy and its impact on the performance of Taiwan life insurance industry investments from 2004 to 2008. We find that currency hedging strategies have yielded positive results, overall, for the industry. However, if the life insurance industry is divided into two categories according to its ownership structure, i.e., domestic-owned and foreign-owned companies, the results show that the currency hedging strategies employed by the domestic-owned companies enjoy advantages over those of foreign-owned firms. If the sample is further divided into those publicly listed on the TAIEX and others, our results show that a hedging strategy has positive effects on listed company. Our findings support the work in Glen and Jorion (1993) and Campbell et al. (2010), which reveal that hedging strategies improve foreign investment returns and can reduce currency risks in comparison to non-hedging strategies. Our empirical results indicate that SFAS No. 34 has a significant effect on currency hedging behavior among domestically owned and listed companies.

隨著我國壽險業資產比率快速增加，投資績效不但影響眾多保戶的權益，更影響整體經濟之安定，但面臨現今全球金融和經濟環境劇烈動盪，加上壽險同業間激烈競爭的情況，我國壽險公司如何訂定投資策略做好資產配置，對壽險公司的營運健全而言相當重要。現行保險法第一百四十六條限制壽險公司的投資上限，但法令限制對壽險公司資產配置的影響為何，本研究將透過Markowitz的平均數－變異數投資組合模式對我國整體壽險業及國泰人壽、南山人壽、新光人壽及富邦人壽探討之；並以夏普指數、崔納指數及詹森指數，評估上述四家壽險公司之資金運用績效；另藉由分析壽險業之資金成本是否低於實際投資率，以達成研究壽險業長期資產配置之穩健度。 本研究主要結論如下：1.運用Markowitz 投資組合模型所推導出的效率前緣，以最大Sharpe Measure評估，不受法令限制下所建立之最佳投資組合，較有受法令限制下所建立之最佳投資組合的期望報酬率高，且分散風險的效果較佳。2.整體壽險業及前四大壽險公司之實際投資報酬率皆低於其實際投資組合之期望報酬，顯示壽險業於資金運用的靈活度及績效性有改善的空間。3.以績效評估指標求出前四大壽險公司的資金運用績效，發現新光人壽在此三項評估指標皆位居最後；國泰人壽在評估中皆名列前茅。4.新光人壽的平均實際報酬率低於平均資金成本，應控管好資金成本並加強投資組合之績效；富邦人壽平均實際報酬率高過平均資金成本最多，顯示富邦人壽在資金成本控管及投資績效有良好之表現。整體壽險業的實際報酬率亦高於平均資金成本，顯示我國壽險業於營運狀況正常。 / With the life insurance companies’ assets ratio rapidly increasing, the investment performances affect not only the right of a number of policy holders, but also affect the economic stability. However, facing the dynamic global financial and economic environment and the keen competition in the domestic life insurance industry, the life insurance companies need to adopt the proper investment strategy. Law of Insurance 146th p restricts the investment upper limit of the life insurance company. This research will use Markowitz MV model to discuss the influence of this investment restriction on life insurance companies’ asset allocation by the samples of Life industry, Cathay Life Insurance, Nan Shan Life Insurance, Shin Kong Life Insurance, and Fubon Life Insurance, and evaluate the performances of these four life insurance companies by Sharpe ratio, Treynor ratio, and Jensen’s measure. This research also analyze the cost of capital and real rate of return of these companies to examine the stability of life insurance industry’s long term asset allocation. The conclusions of this research are as follows: 1.Evaluated by the Markowitz efficient frontier and the Sharpe measure, there is the higher expected rate of return and better diversification with no investment restriction. 2.The actual rates of return of the life insurance industry and the above four life insurance companies are below the expected rates of returns of their portfolio evaluated be the Sharpe measure, which means the life insurance industry need to prove their capital allocation. 3. Comparing the performance of the life insurance companies by the performance indicator, we find the then Shin Kong Life Insurance is the last, while Cathay Life Insurance has a good score. 4. We also find the real rate of return of Shin Kong Life Insurance is lower than its cost of capital, which means Shin Kong Life Insurance need to adjust its cost of capital and the investment performance. Meanwhile, Fubon Life Insurance is the excellent in controlling the cost of capital and investment. The real rate of return of the Life insurance industry is higher than its cost of capital, and that shows the Life insurance industry has normal operation.

本文包含關於估計風險與動態資產配置的兩篇研究。第一篇研究主要就當須估計的投資組合其投入參數具有高維度特質的觀點下，探究因忽略不確定性通膨而對資產配置過程中帶來的估計風險。此研究基於多重群組架構下所發展出的新投資決策法則，能夠確實地評價不確定性通膨對資產報酬的影響性，並在應用於建構大規模投資組合時，能有效減少進行最適化投資決策過程中所需的演算時間與成本。而將此模型應用於建構全球ETFs投資組合的實證結果則進一步顯示，若在均值變異數架構下，因建構大型投資組合時須估計高維度投入參數而伴隨有大量估計風險時，參數估計方式建議結合採用貝氏估計方法來估算資產報酬的一階與二階動差，其所對應得到的投資組合樣本外績效會比直接採用歷史樣本動差來得佳。此實證結果亦隱含：在均值變異數架構下，穩定的參數估計值比起最新且即時的參數估計資訊對於投資組合的績效來得有益。同時，若當投入參數的樣本估計值波動很大時，增加放空限制亦能有利投組樣本外績效。 第二篇文章則主要處理當處於對數常態證券市場下時，投資組合報酬率不具有有限動差並導致無法在均值變異數架構下發展出最適化封閉解時的難題。本研究示範此時可透過漸近方法的應用，有效發展出在具有放空限制下，考量了估計風險後的簡單投資組合配置法則，並且展示如何將其應用至實務上的資產配置過程以建構全球投資組合。本文的數值範例與實證模擬結果皆顯示，估計風險的存在對於最適投資組合的選擇有實質的影響，無估計風險下得出的最適投資組合，不必然是存有估計風險下的最適投資組合。此外，實證模擬結果亦證明，當存有估計風險時，本文所發展的簡單法則，能使建構出的投資組合具有較佳的樣本外績效表現。 / This dissertation consists of two essays on dynamic asset allocation with regard to dealing with estimation risk as being in different uncertainties in the mean-variance framework. The first essay concerns estimation errors from disregarding uncertain inflation in terms of the need in estimating high-dimensional input parameters for portfolio optimization. This study presents simplified and valid criteria referred to as the EGP-IMG model based on the multi-group framework to be capable of pricing inflation risk in a world of uncertainty. Empirical studies shows the proposed model indeed provides a smart way in picking worldwide ETFs that serves well to reduce the amount of costs and time in constructing a global portfolio when facing a large number of investment products. The effect of Bayesian estimation on improving estimation risk as the decision maker is subject to history sample moments for input parameters estimations is meanwhile examined. The results indicate portfolios implementing the Stein estimation and shrinkage estimators offer better performance compared with those applying the history sample estimators. It implicitly demonstrates that yielding stable estimates for means and covariances is more critical in the MV framework than getting the newest up-to-date parameters estimates for improving portfolio performance. Though short-sales constraints intuitively should hurt, they do practically contribute to uplift portfolio performance as being subject to volatile estimates of returns moments. The second essay undertakes the difficulty that the probability distribution of a portfolio's returns may not have finite moments in a lognormal-securities market, and thus leads to the arduous problem in solving the closed-form solutions for the optimal portfolio under the mean-variance framework. As being in a lognormal-securities market, this study systematically delivers a simple rule in optimization with regard to the presence of estimation risk. The simple rule is derived accordingly by means of asymptotic properties when short sales are not allowed. The consequently numerical example specifies the detailed procedures and shows that the optimal portfolio with estimation risk is not equivalent to that ignoring the existence of estimation risk. In addition, the portfolio performance based on the proposed simple rule is examined to present a better out-of-sample portfolio performance relative to the benchmarks.

國內壽險公司面臨資金不斷累積與同業間激烈競爭，加上全球經濟動盪不安，國際金融偶有黑天鵝事件發生，尤以2008年金融海嘯重創全球產業為最，壽險業亦難以倖免，肇致壽險業者經營益顯艱困。因此，為維繫公司永續經營，規劃長期財務穩健性至關重要。是故，如何訂定投資策略與妥適資產配置，並兼顧風險管理，為當前壽險公司重要課題。 本研究分析郵政壽險資產配置行為，歸納影響公司資產配置之內、外在因素，例如流動性風險、利率風險、信用風險、資本適足率、匯率風險及法令規範等因素，皆影響資產配置策略。另為建構最佳資產配置，提升獲利，逐步改善財務結構，在現行法令限制下，運用Markowitz之投資組合理論為分析工具，導出效率前緣線，再運用夏普指標績效分析，來建立最佳投資組合。 另外分析壽險公司必須正視未來「國際會計準則」（International Financial Reporting Standards, IFRS）IFRS 4第二階段適用公平價值評估負債，利差損問題會更加嚴重，對業務發展與負債評價將產生巨大衝擊。尤其我國壽險業發行商品大都以長年期終身險為主，若壽險商品負債評價與資產不一致時，其缺口將因利率變化影響損益波動。 / Abstract Domestic life insurers are in a challenging environment with increasing asset size to manage and fierce competition within the industry. Moreover, the world economy is going down a bumpy path. Every now and then in the global financial system, we encounter a black swan event. Among them, the financial tsunami of 2008 hit global industries most severely. The financial tsunami of 2008 also left life insurers having an increasingly difficult time running the business. It is crucial to have sound long-term financial plans in order to ensure business sustainability. Therefore, how to form an investment strategy, determine asset allocation and manage risks at the same time becomes a critical issue for life insurers. The research studies the asset allocation behavior of Chunghwa Post insurance sector and lists both the internal and the external factors affecting asset allocation. Factors like liquidity risk, interest rate risk, credit risk, capital adequacy, currency risk and regulations all have some influence on the asset allocation strategy. Meanwhile, the research constructs efficient frontier with Markowitz Portfolio Theory and adopts Sharpe ratio as the performance measure to build an optimal portfolio under current regulations with the goal of optimizing asset allocation, boosting profits and gradually improving the financial structure. The research also studies the tremendous impact of IFRS 4 on business development and liability valuation of life insurance companies. The implementation of IFRS 4 Phase II will require fair value measurement of liabilities, which will exacerbate the negative interest spread problem. When the liability valuation approach of insurance products is not in line with asset valuation, the gap will intensify the income fluctuations from interest rate movements, especially for domestic life insurers whose main products are long-term whole life policies.