A bilingual research brief for When Winning Streaks Mislead: Regime Dependence, Tail Deferral, and Survivorship Bias in Retail Fixed Coupon Notes — SSRN Working Paper No. 7106058, Sinclair Huang, 2026. Why multi-year profit streaks in autocallable structured notes reveal far less than intuition suggests.
What the paper studies. A retail Fixed Coupon Note (FCN) is, in economic substance, a short position in a worst-of down-and-in put sold by the investor. The paper asks what a long no-assignment streak reveals about future tail risk. It uses streak illusion in a statistical, not psychometric, sense: regime-dependent tail deferral can generate a benign record that says little about resilience after a state change. The study does not observe investor beliefs and does not estimate behavioral bias — the term names an information-content problem, not a measured psychological effect.
Three findings. First, in a strong-bull single-regime benchmark, eight placements avoid assignment (Scenario D) about 78% of the time and the expected run before assignment is about 32 completed placements — placement counts, not calendar time. In purged forward test blocks, a streak-only model has mean block AUCs of 0.489 for the next note and 0.467 for a one-year event, indicating no stable ranking ability. Second, positive modeled returns rely on regime-conditional equity and variance risk premia; the negative pricing wedge in the sequential decomposition is a stylized residual, not an issuer-fee estimate. Third, a 1,260-trading-day simulation beginning in a bull state has a −8.9% median total return and a −71.8% 5th percentile.
The historical comparison. Across nine surviving semiconductor stocks (108,780 overlapping hypothetical contract-start/basket observations, 2000–2026), assignment clusters in 2000–02 and 2008, reaching 54.4%. Of fixed 504-trading-day paths, 56.8% have no Scenario-D settlement despite completing 11.5 notes on average; fixed 1,260-day paths have a +23.9% median and a −68.3% 5th percentile. A start-year block bootstrap gives a 95% interval of [+0.5%, +39.7%] for the five-year median, but only partly addresses dependence among overlapping windows. Synthetic casualties are an assumption-driven sensitivity analysis, not a survivorship-bias correction, and cross-industry results show a broad, non-monotonic association between volatility and assignment.
What follows. The evidence supports state-conditional risk disclosure; conclusions about investor understanding or choice would require separate investor-level data.
中文｜核心觀點
散戶端的固定配息票據（FCN），在經濟實質上是投資人賣出的 worst-of 下敲入賣權的空頭部位。本文要回答的是：一段長期「未履約」的連勝紀錄，對未來的尾部風險究竟透露了多少訊息。文中的「連勝幻覺」是統計意義、而非心理計量意義上的用語——在制度依賴的尾部遞延下，溫和市場可以產生一段亮眼紀錄，但它對狀態切換後的抗跌能力幾乎沒有資訊含量；本研究並未觀察投資人信念，也未估計任何行為偏誤。
三個發現：其一，在強勢多頭的單一制度基準下，連續八次配置約有 78% 機率未觸發履約，預期可完成約 32 次配置——這是合約次數、不是日曆時間；且在 purged forward 測試區塊中，僅用連勝訊號的模型平均 AUC 只有 0.489／0.467，顯示沒有穩定的排序能力。其二，模型中的正報酬來自制度條件下的股權與波動風險溢酬；序列分解中的負向定價楔差是風格化殘差，不是發行商費用的估計。其三，自多頭狀態起步的 1,260 交易日模擬，中位數總報酬為 −8.9%，第 5 百分位為 −71.8%。
歷史對照（九檔存續半導體股、108,780 個重疊的假想合約起點觀測）：履約集中於 2000–02 與 2008 年，最高達 54.4%；固定 504 交易日路徑中，56.8% 未發生履約結算；起始年區塊拔靴法給出五年期中位數的 95% 區間 [+0.5%, +39.7%]，但僅部分處理重疊視窗的相依性。合成傷亡樣本屬於假設驅動的敏感度分析，不是生存者偏誤的校正；跨產業結果顯示波動度與履約之間是廣泛但非單調的關聯。政策面，證據支持「依市場狀態的風險揭露」；關於投資人理解或選擇的結論，需要另行取得投資人層級的資料。