grad-event-study
用事件研究法算公告前後的異常報酬，量化市場反應
中文翻譯
--- name: "grad-event-study" description: "用事件研究法衡量公司或市場事件前後的異常報酬與累積異常報酬（CAR）。當使用者需要量化公告對市場的影響、設計事件窗口與估計窗口，或問到「這個事件有沒有影響股價」「怎麼計算異常報酬」「市場對這個公告的反應是什麼」時，就用這個 skill。" metadata: category: "WP-31 量化方法" tags: ["event-study", "abnormal-return", "CAR", "market-reaction", "event-window", "finance"] --- # 事件研究法 (Event Study) ## Overview 事件研究法（Fama 等人，1969；MacKinlay，1997）把實際報酬和預期（正常）報酬的模型相比，分離出可歸因於特定事件的異常報酬。事件窗口內的累積異常報酬（CAR）量化了市場的整體反應。 ## When to Use - 衡量市場對財報公告、併購、政策變動或監管事件的反應 - 檢定半強式的市場效率 - 量化公司揭露資訊的經濟顯著性 - 比較市場對不同事件類型或公司特徵的反應 ## When NOT to Use - 事件日期不明確，或資訊是逐漸洩漏的 - 有干擾事件和事件窗口重疊 - 該公司股票流動性差，有很多零報酬的日子 - 事件早就被廣泛預期，在事件窗口之前就已完全反映在價格上 ## Assumptions ``` IRON LAW: Event study validity requires that the event was UNANTICIPATED — if the market priced it in before the event window, abnormal returns will be zero even if the event matters. ``` 關鍵假設： 1. 事件日期可以精確識別，而且事件是出乎意料的 2. 事件窗口內沒有干擾事件 3. 正常報酬模型在估計窗口內設定正確 4. 市場微結構效應（交易清淡、買賣價差跳動）不會扭曲報酬 ## Methodology ### Step 1 — Define Event and Windows 找出事件日（第 0 天）。設定估計窗口（例如 [-250, -11]）來估計正常報酬。設定事件窗口（例如 [-1, +1] 或 [-5, +5]）來捕捉反應。 ### Step 2 — Estimate Normal Returns 使用市場模型：Ri,t = αi + βi × Rm,t + εi,t，在估計窗口內估計。替代方案包括固定平均報酬或 Fama-French 因子。模型設定見 `references/`。 ### Step 3 — Compute Abnormal and Cumulative Abnormal Returns AR = 事件窗口內每一天的實際報酬 - 預期報酬。CAR = 事件窗口內 AR 的加總。跨公司計算 CAAR（累積平均異常報酬）。 ### Step 4 — Statistical Testing 用參數檢定（橫斷面 t 檢定、Patell 檢定）和無母數檢定（符號檢定、秩檢定）檢定 H₀: CAR = 0。兩種都報告，確保穩健。 ## Output Format ```markdown ## Event Study: [Event Description] ### Window Design | Window | Period | Rationale | |--------|--------|-----------| | Estimation | [-250, -11] | [rationale] | | Event | [-1, +1] | [rationale] | ### Abnormal Returns | Day | AR (%) | t-stat | |-----|--------|--------| | -1 | x.xx | x.xx | | 0 | x.xx | x.xx | | +1 | x.xx | x.xx | ### Cumulative Abnormal Returns | Window | CAR (%) | t-stat | p-value | Significant? | |--------|---------|--------|---------|-------------| | [-1, +1] | x.xx | x.xx | x.xx | [Yes/No] | ### Cross-Sectional Analysis - [If applicable: regression of CAR on firm characteristics] ### Limitations - [Note any confounding events or assumption violations] ``` ## Gotchas - 事件日期集中（例如全產業的法規）會違反橫斷面獨立性 - 估計窗口太短，會讓正常報酬的參數很吵 - 事件窗口太長，會提高出現干擾事件的機率 - 低價股和流動性差的證券，會人為墊高異常報酬 - 市場模型假設 beta 固定，結構性斷裂會讓這個假設失效 - 發表偏誤：發現 CAR 為零的研究很少被發表 ## References - MacKinlay, A. C. (1997). Event studies in economics and finance. *Journal of Economic Literature*, 35(1), 13-39. - Fama, E. F., Fisher, L., Jensen, M. C., & Roll, R. (1969). The adjustment of stock prices to new information. *International Economic Review*, 10(1), 1-21. - Kolari, J. W., & Pynnönen, S. (2010). Event study testing with cross-sectional correlation of abnormal returns. *Review of Financial Studies*, 23(11), 3996-4025.
英文原文
--- name: "grad-event-study" description: "Apply event study methodology to measure abnormal returns and cumulative abnormal returns (CAR) around corporate or market events. Use this skill when the user needs to quantify the market impact of announcements, design event and estimation windows, or when they ask 'did this event affect stock price', 'how do I calculate abnormal returns', or 'what is the market reaction to this announcement'." metadata: category: "WP-31 量化方法" tags: ["event-study", "abnormal-return", "CAR", "market-reaction", "event-window", "finance"] --- # 事件研究法 (Event Study) ## Overview The event study method (Fama et al., 1969; MacKinlay, 1997) isolates the abnormal return attributable to a specific event by comparing actual returns against a model of expected (normal) returns. Cumulative abnormal returns (CAR) over an event window quantify the total market reaction. ## When to Use - Measuring market reaction to earnings announcements, M&A, policy changes, or regulatory events - Testing semi-strong form market efficiency - Quantifying the economic significance of corporate disclosures - Comparing market reactions across different event types or firm characteristics ## When NOT to Use - The event date is ambiguous or the information leaked gradually - Confounding events overlap with the event window - The firm's stock is illiquid with many zero-return days - The event was widely anticipated and fully priced before the event window ## Assumptions ``` IRON LAW: Event study validity requires that the event was UNANTICIPATED — if the market priced it in before the event window, abnormal returns will be zero even if the event matters. ``` Key assumptions: 1. Event date is precisely identifiable and the event was unexpected 2. No confounding events occur within the event window 3. The normal return model is correctly specified during the estimation window 4. Market microstructure effects (thin trading, bid-ask bounce) do not distort returns ## Methodology ### Step 1 — Define Event and Windows Identify the event date (day 0). Set estimation window (e.g., [-250, -11]) to estimate normal returns. Set event window (e.g., [-1, +1] or [-5, +5]) to capture the reaction. ### Step 2 — Estimate Normal Returns Use the market model: Ri,t = αi + βi × Rm,t + εi,t estimated over the estimation window. Alternatives include constant mean return or Fama-French factors. See `references/` for model specifications. ### Step 3 — Compute Abnormal and Cumulative Abnormal Returns AR = Actual return - Expected return for each day in the event window. CAR = sum of ARs over the event window. Compute CAAR (cumulative average abnormal return) across firms. ### Step 4 — Statistical Testing Test H₀: CAR = 0 using parametric tests (cross-sectional t-test, Patell test) and non-parametric tests (sign test, rank test). Report both for robustness. ## Output Format ```markdown ## Event Study: [Event Description] ### Window Design | Window | Period | Rationale | |--------|--------|-----------| | Estimation | [-250, -11] | [rationale] | | Event | [-1, +1] | [rationale] | ### Abnormal Returns | Day | AR (%) | t-stat | |-----|--------|--------| | -1 | x.xx | x.xx | | 0 | x.xx | x.xx | | +1 | x.xx | x.xx | ### Cumulative Abnormal Returns | Window | CAR (%) | t-stat | p-value | Significant? | |--------|---------|--------|---------|-------------| | [-1, +1] | x.xx | x.xx | x.xx | [Yes/No] | ### Cross-Sectional Analysis - [If applicable: regression of CAR on firm characteristics] ### Limitations - [Note any confounding events or assumption violations] ``` ## Gotchas - Clustering of event dates (e.g., industry-wide regulation) violates cross-sectional independence - Short estimation windows produce noisy normal return parameters - Long event windows increase the probability of confounding events - Penny stocks and illiquid securities inflate abnormal returns artificially - The market model assumes constant beta — structural breaks invalidate this - Publication bias: studies finding zero CAR are rarely published ## References - MacKinlay, A. C. (1997). Event studies in economics and finance. *Journal of Economic Literature*, 35(1), 13-39. - Fama, E. F., Fisher, L., Jensen, M. C., & Roll, R. (1969). The adjustment of stock prices to new information. *International Economic Review*, 10(1), 1-21. - Kolari, J. W., & Pynnönen, S. (2010). Event study testing with cross-sectional correlation of abnormal returns. *Review of Financial Studies*, 23(11), 3996-4025.