1 / 64

Security Investment Analysis ─ Knowledge Discovery and Computational Intelligence 證券投資分析─知識發現與計算智慧

Security Investment Analysis ─ Knowledge Discovery and Computational Intelligence 證券投資分析─知識發現與計算智慧. 葉怡成 中華大學 資訊管理系 Prof. I-Cheng Yeh Department of Information Management Chung-Hua University. 大綱. 證券投資分析─選股與擇時 知識發現與計算智慧 以知識發現建構選股模型 以計算智慧建構選股模型 未來的研究方向. 1. 證券投資分析─選股與擇時.

yuma
Download Presentation

Security Investment Analysis ─ Knowledge Discovery and Computational Intelligence 證券投資分析─知識發現與計算智慧

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Security Investment Analysis ─Knowledge Discovery and Computational Intelligence證券投資分析─知識發現與計算智慧 葉怡成 中華大學 資訊管理系 Prof. I-Cheng Yeh Department of Information Management Chung-Hua University

  2. 大綱 • 證券投資分析─選股與擇時 • 知識發現與計算智慧 • 以知識發現建構選股模型 • 以計算智慧建構選股模型 • 未來的研究方向

  3. 1. 證券投資分析─選股與擇時 1-1 為何選股與擇時如此重要? 1-2 證券投資分析 1-3 為何專家選股與擇時老是行不通? 1-4 計量投資模型 1-5 驗證選股模型的原則

  4. 1-1 為何選股與擇時如此重要? 投資的目標 1.最大化報酬 2.最小化風險 Ex. 投資100萬,年報酬7%與20%的30年差距 7%:760萬 20%:2億4千萬 相差31倍

  5. 1-2 證券投資分析 What are the key problems in decision-making of stock investment? • 選股 (stock selection):which to buy/sell • 擇時 (market timing):when to buy/sell

  6. How do experts make decisions in stock selection? Fundamental Analysis • Value Factor: Cheap > Expensive (P/E ratio, P/B ratio) • Growth Factor: Earning > Deficit (ROE) • Scale Factor: Small > Large (Market Capital) • Moment Factor: Winner > Loser (Last quarter return) • Liquidity Factor: Cold > Hot (Turnover, Trading volume)

  7. 最有效的選股因子:價值因子與成長因子 權益證券的本質: 淨值與盈餘

  8. Performance of Value Factor and Growth Factor

  9. How do experts make decisions in Market Timing? Technical Analysis • Moving average (MA) • MACD • KD • RSI • OBV

  10. Technical Analysis: Moving Average Approach MA(1)>MA(50) 買入點 MA(1)<MA(50)賣出點 MA(1)<MA(50)賣出點

  11. Performance of Moving Average

  12. 1-3 為何專家選股與擇時老是行不通? • 專家也是凡人Ⅰ ─ 學習的偏誤 • 專家也是凡人Ⅱ ─ 貪婪與恐懼 • 專家也是凡人Ⅲ ─ 自私與代理人效應 • 學習偏誤的處方 ─ 計量投資模型 • 貪婪恐懼的處方 ─ 自我紀律

  13. 1-4 計量投資(quantitative investment ) • A quantitative investment is an investment in which investment decisions are determined by numerical methods rather than by human judgment. • If the whole procedure is done by human judgment or intuition, an investment process will be labeled as a “fundamental” one. • If it is purely done by computer-based models, the process can be classified as “quantitative”. • 以知識發現與計算智慧建構選股模型

  14. 1-5 驗證選股模型的原則 • 避免資料操弄偏差(data-snooping bias):模型要簡單才有普遍性。 • 避免短期偏差:模型要歷經長期考驗才有普遍性。 • 避免先視偏差:模型不可「偷看」歷史資料。 • 避免存活偏差(survivorship bias):模型不可「忽略」下市個股。 • 避免微型股偏差:模型要考慮實際操作的可行性。 • 考量成本原則:模型要考慮交易成本的侵蝕。 • 合理風險原則:模型要考慮風險與報酬的取捨。

  15. What do you see?(資料操弄偏差)

  16. 資料操弄偏差 • The particular parameters that researchers work with are often chosen because they have been shown to be related to returns. • For instance, suppose that you were asked to explain the change in SAT test scores over the past 40 years in some particular state. Suppose that to do this you searched through all of the data series you could find. After much searching, you might discover, for example, that the change in the scores was directly related to the jackrabbit population in Arizona. We know that any such relation is purely accidental; but if you search long enough and have enough choices, you will find something even if it is not really there. • Needless to say, the researchers on these matters defend their work by arguing that they have not mined the data and been very careful to avoid such traps by not snooping at the data to see what will work.

  17. 2. 知識發現與計算智慧 2.1 Information system level 2.2 Knowledge Discovery 2.3 Computational Intelligence

  18. 2.1 Information system level • Data Level: Data retrieval (DBMS) • Information Level: Information generating, What-if analysis (MIS) • Knowledge Level: Knowledge discovering (Knowledge Discovery) • Intelligence Level: Intelligence creating (Computational Intelligence)

  19. 2.2 Knowledge Discovery • Knowledge Discovery = Search for valuable information in large volumes of data. • KD methods mainly include: • Regression analysis • Neural networks • Inductive Decision Tree

  20. Regression analysis

  21. Neural networks

  22. Inductive Decision Tree

  23. 2.3 Computational Intelligence • Computational Intelligence involves iterative development or learning. Learning is based on empirical data. It is also known as non-symbolic AI and soft computing. • Computational Intelligence methods mainly include: • Evolutionary computation • Other bio-inspired computing Particle swarm optimization Ant colony optimization Artificial life Artificial immune systems

  24. Evolutionary computation applies biologically inspired concepts such as populations, mutation, and survival of the fittest to generate increasingly better solutions to the problem.

  25. 以GA解最佳化的優點:不可微分與局部最大值函數以GA解最佳化的優點:不可微分與局部最大值函數

  26. Data structure of GA

  27. Algorithm of GA

  28. 3. 以知識發現建構選股模型 3.1 變數的處理─排序正規化 3.2 單變數分析─相關係數與排序法 3.3 迴歸分析 3.4 神經網路 3.5 迴歸樹 3.6 模型比較 3.7 小結

  29. 自變數 X1=第t+1季報酬率 X2=第t+1季ß值   X3=第t季負債/淨值比   X4=第t季股東權益報酬率 (ROE) X5=第t+1季成交量(百萬股)   X6=第t+1季週轉率   X7=第t+1季市值(季底)   X8=第t+1季收盤價(季底)   X9=第t淨值股價比(B/P) X10=第t益本比(E/P) X11=第t每股淨值   X12=第t每股盈餘 (EPS) X13=第t稅後淨利 X14=第t最新淨值股價比(B/P) X15=第t最新益本比(E/P) 因變數 Y=第t+2季報酬率 3.1 變數的處理─排序正規化

  30. 排序正規化 將自變數與因變數正規化,即將各股票的各變數分季排序,該季最大者其排序值Rank=1;最小者Rank=0,其餘依此內插。例如中位數的Rank=0.5。 優點: (1) 專注橫向資料比較 (2) 避免單季資料偏差 (3) 避免極端資料偏差 (4) 降低錯誤資料影響

  31. 3.2 單變數分析─相關係數與排序法各因子的第t+2季報酬率Rank值平均值的十等分圖 最重要變數 B/P與E/P

  32. 各因子對報酬率的相關係數 最重要變數 B/P與E/P 最重要變數 B/P與E/P

  33. 3.3 迴歸分析 最重要變數 B/P與E/P

  34. 3.4 神經網路 最重要變數 B/P與E/P 最重要變數 B/P與E/P

  35. 3.5 迴歸樹 最重要變數 B/P與E/P 最重要變數 B/P與E/P 最重要變數 B/P與E/P

  36. 3.6 模型比較 模型評估:季報酬率的比較

  37. 3.7 小結 • 迴歸分析在判斷變數的影響方向上並不正確。 • 神經網路常造成過度學習,但神經網路在判斷變數的影響方向上遠比迴歸分析正確。 • 迴歸樹以三分段表現最佳。 • 迴歸分析、神經網路、迴歸樹這三種方法的投資績效大致上相近。 • 各法都發現X14「最新淨值股價比」與X15「最新營餘股價比」是最重要變數。 • 基本分析所能獲得的超額季報率約5~6%。

  38. 4. 以計算智慧建構選股模型 4.1 系統原理 4.2 規則篩選法 4.3 評分排序法 4.4 模型比較 4.5 小結

  39. 4.1 系統原理 GA最佳化引擎 選股系統

  40. 4.2 規則篩選法 原理:測試與發現最能篩選出高報酬股票的規則。 (1) 知識結構:假設股票的篩選規則如下: IF X1 R1 C1 AND X2 R2 C2 AND X3 R3 C3 其中 • X1, X2, X3={1,2,…,15}分別代表益本比等選股變數的Rank值 • R1, R2, R3={<, >}分別代表 “<” 與 “>” • C1, C2, C3={1,2}分別代表 0~1之間的實數 例如 IF 4 > 0.2 代表 IF X4(淨值報酬率Rank值) > 0.2 AND 14 > 0.9 AND X14(新淨值市值比Rank值) > 0.9 AND 5 < 0.9 AND X5(成交量Rank值) < 0.9

  41. (2) 目標函數:應用篩選規則在每一季的股票資料庫,可產生每一季的投資組合,此投組的報酬率公式如下: 投組的報酬率 = 投組內所有股票的報酬率總和 ÷ 投組內所有股票的總數 利用一段期間內(通常是十年左右)的每一季的投組報酬率可以產生一個代表篩選規則在此期間內的年複利報酬率。 (3) 最佳化過程:GA可調整X1~X3, R1~R3, C1~C3等九個參數來進行交易模擬,以找出能最大化在此期間內的年複利報酬率的最優交易策略。

  42. 世代數為20,個體數為20優化過程 報酬率收斂

  43. 訓練與測試期間報酬率Rank平均值散佈圖

  44. GA產生的選股規則 最重要變數 X14(B/P)與X15(E/P)

  45. 4.3 評分排序法

  46. 世代數為20,個體數為20優化過程 報酬率收斂

  47. 訓練與測試期間報酬率Rank平均值散佈圖

  48. GA產生的選股函數

  49. 選股函數的係數 最重要變數 B/P與E/P 最重要變數 B/P與E/P 最重要變數 B/P與E/P 最重要變數 B/P與E/P

  50. 4.4 模型比較 模型評估:季報酬率的比較

More Related