sp500年化報酬率的問題，透過圖書和論文來找解法和答案更準確安心。 我們找到下列股價、配息、目標價等股票新聞資訊

長期投資研究：以黃金、不動產投資信託、債券型ETF、臺灣ETF及美國ETF為例

為了解決sp500年化報酬率的問題，作者林韋丞 這樣論述：

Investment has been a serious issue for those who want to have a better quality of life or who expect to be retired earlier. Fortune, to some extent, is the key to own freedom in this society. However, how to reach the door of freedom by a relatively safer way to invest? Based on this idea, this th

esis aims at the long-term investment with the target of gold, U.S. REITs, U.S. Bond ETF, Taiwan Stock ETF, and U.S. Stock ETF. To support the statement provided in this thesis, this thesis has collected historical data for 27-30 years from Bloomberg Professional and used the data to calculate the a

nnualized rate of return and standard deviation of each long-term investment of ten-year from different periods (e.g., from 1992 to 2001). By analyzing the result of each of the investing tools above, this thesis discovered that the annualized rates of return and standard deviations of gold, U.S. RE

ITs, Taiwan Stock ETF, and U.S. Stock ETF are not stable while those of the U.S. Bond ETF are steady. When putting all the investing tools together into one portfolio, this thesis unearthed that the annualized rate of return and standard deviation all turn out to be stable even in a financial crisis

. Thus, this thesis concluded that when doing a long-term investment, it is vital to have a well-selected portfolio. In this way, it is possible to beat the market.

利用 TCN 及 Residual LSTM 建立股票投資深度學習回測模型

為了解決sp500年化報酬率的問題，作者張博鈞 這樣論述：

　　過去深度學習應用於股票之研究已有很多，但大多應用僅止於預測股價，該類研究衡量方式大多為統計量，如均方差(MSE;Mean square error)、Accuracy 衡量、衡量股價變動方向的正確率，以上之研究多停留提升預測之精準度，較少人將研究拓展至將預測股價應用於實際交易，並檢驗何種模型產生之投資績效較好，本研究的貢獻在於從深度學習模型得到預測股價後，利用程式交易的方法進一步將預測股價應用於實際交易，並驗證投資績效好壞。過去研究曾對TCN及LSTM、GRU(Gated recurrent units network)優劣做廣泛性的比較，發現 TCN 優於傳統 LSTM、GRU，但尚未

有TCN 與加上殘差連接(residual connection)的 LSTM網路的比較。　　本研究將TCN及RES LSTM(Residual LSTM)比較，應用在四個常被交易的標的上(兩個指數、微軟股票、JPMorgan Chase 股票)，檢視兩者何者創造較高好的投資績效，同時與買進持有(buy and hold)的投資績效比較。　　研究結果發現，發現不論是預測次營業日(x =1)或次週(x =5)後收盤價、或是以TCN網路還是RES LSTM網路預測，這兩種參數變動，都得到神經元數越少，年化報酬率越高這個結論，但是神經元數越少，風險報酬比不一定越高。　　不考慮神經元數的 8 種參數組

合中，以年化報酬率而言有 5 個組合為 RES LSTM網路較好，但是考量到風險的獲利指標，風險報酬比，8 種組合中有 5 個組合為TCN網路較好。TCN 參數、訓練時間均較RES LSTM 少，但是平均年化報酬率較高、平均風險報酬比較高，也能產生較多的獲利高原，且在各參數組合中以風險報酬比比較，獲勝組數較多，大抵而言，以動量交易策略應用，TCN 網路結果優於RES LSTM 網路。