Earnings flash dip catch

财报接飞刀 — Post-Earnings Flash-Dip Catch

Purpose: 财报夜股价插针,在高概率的位置接刀、在阻力位卖出。 The #1 rule (born from the MRVL $170 error): 接刀位锚定最近的台阶, 不是旧基座;并用 实现波动 ≈ 隐含 × 0.65 做硬校验。详见最后的 Post-Mortem。

---

When to invoke

✅ Trigger when: a clear ticker + earnings/post-earnings context ("财报接飞刀", "盘后捞货", "财报后怎么接", "catch the dip on X earnings").

❌ Do NOT trigger for: code "catch", error handling, flaky-test catches, or any non-equity use of "catch/dip/knife".

---

STEP 1 — Pull the data (exact tool calls)

Run these mcp__yfmcp__* tools. Do them in parallel where independent.

1. yfinance_get_ticker_info → read these fields:
   • currentPrice, regularMarketPrice, previousClose, regularMarketDayLow/High
   • postMarketPrice, postMarketChangePercent, marketState (POST/PRE/REGULAR)
   • fiftyTwoWeekHigh, fiftyDayAverage, twoHundredDayAverage
   • forwardPE, trailingPE, priceToSalesTrailing12Months, targetMedianPrice, targetMeanPrice, numberOfAnalystOpinions, recommendationKey
   • earningsTimestamp (when it reports), beta, averageVolume
2. yfinance_get_price_history with period="5d", interval="30m", prepost=true → this is the after-hours tape; read the actual flush low/high from the post-4pm-ET bars. Also pull period="3mo", interval="1d" for the shelf (Step 3).
3. yfinance_get_option_dates → pick the weekly that expires just after the print (for the straddle/expected move) AND the next monthly (for durable walls).
4. yfinance_get_option_chain for both dates. Option chains are large and will exceed the token limit → they save to a file. Parse with python/jq:
   • Top open-interest call strikes = call walls (resistance / sell-into).
   • Top open-interest put strikes = put walls (support / dip-catch).
   • ATM straddle mid (nearest-strike call_mid + put_mid) ÷ spot = expected move %.

Caveat to always state: the price-alert cron checker reads the regular-session price (regularMarketPrice/last_price) — it will NOT see tonight's after-hours flush. After-hours catching = broker extended-hours limit orders, not the bot.

---

STEP 2 — Read the stats (what matters)

Field	What it tells you
postMarketPrice vs previousClose	the actual earnings gap % (the move that happened)
straddle expected move %	what options PRICED IN (1σ). Realized ≈ this × 0.6–0.7
fiftyTwoWeekHigh, run vs 50/200DMA	how extended → bigger run = bigger "sell the news" risk
forwardPE, priceToSales, targetMedianPrice	is it already above analyst targets? (premium = fragile)
beta	high beta (>2) = wider AH whipsaw

---

STEP 3 — Find the SHELF (the critical step I got wrong)

The primary catch is anchored to the most-recent higher-low consolidation shelf — the last zone where price went sideways for 2–5 days building higher lows, right before the final impulsive leg up into earnings.

Algorithm:

1. From the daily 3mo chart, walk BACK from the pre-earnings high.
2. Find the last multi-day base where the daily lows cluster (the launchpad of the final leg). That cluster's low = the shelf floor = primary catch.
3. Use the MOST RECENT shelf. The base migrates UP with price. If a new shelf formed 3 days ago, that is the anchor — NOT a shelf from 3 weeks ago.

MANDATORY guardrail (this catches the MRVL error):

The shelf must be within ~1 realized move of the pre-earnings close, i.e. shelf ≳ close × (1 − implied×0.65). If your candidate shelf is more than ~1.3× the realized move below the close, it is the WRONG (stale) shelf — go find the newer, higher one.

Example check: MRVL close $208, implied 13.3% → realized ≈ 8.6% → expected floor ~$190. A $170 "shelf" is −18% = 2× too far → reject it, it's the old base.

---

STEP 4 — The catch/sell formula

realized_move   = implied_straddle_pct × 0.65          # vol-risk-premium haircut
expected_floor  = pre_earnings_close × (1 − realized_move)

CATCH 1 (primary, ~50%) = the shallower (higher) of { most-recent shelf , expected_floor }
                          ≈ also the top put wall just under spot
CATCH 2 (deep, ~30%)    = next put-wall cluster below  (gap-fill; fills only if it keeps falling)
CATCH 3 (tail, ~20%)    = full 1σ implied straddle low  (rare; usually does NOT fill)

STOP                    = CLOSING break below the start of the latest leg up
                          (a structural level, NOT a vol-derived number)

SELL 1 / 2 / 3          = the stacked CALL walls above (trim 1/3 at each); max-pain
                          and prior 52wH/ATH are magnets

Core mental model (never violate):

• 跌多深取决于财报质量,而质量盘前不可知 → 主仓必须放在高概率的台阶, 绝不押在波动率尾部 (1σ)。
• 好财报 → 跌到最近台阶就被买回(浅跌,接刀1 成交)。
• 烂财报 → 才会捅穿台阶到 1σ(接刀2/3 + 可能触发止损)。
• 所以"接刀1 成交、深接没成交"是正常且正确的——别去追那个不会来的深价。

---

STEP 5 — Option walls & max pain (post-print nuance)

• Call wall (high call OI) = resistance the dealers defend → sell-into zone.
• Put wall (high put OI) = support the dealers defend → dip-catch zone.
• Max pain = strike where most options expire worthless = dealer magnet into OPEX.
• After the print, the pre-earnings WEEKLY walls evaporate (IV crush + those were earnings bets that close/expire). Trust the MONTHLY walls for durable structure; re-read max pain after the gap.
• Dealers can't fully hedge in the thin after-hours book → price overshoots the "fair" gamma level by a few %, then mean-reverts at the 9:30 ET open. That overshoot is the wick you catch with a resting limit order, not by watching.

---

STEP 6 — Output template

# $TICKER 财报接飞刀 — 接/卖阶梯

**现价**: $X (盘后 $Y, marketState)  · **前收**: $Z
**财报**: [beat+raise / in-line / miss] — [一句话实际数字]
**预期波动**: 隐含 ±A%(跨式) → 实现估计 ±B% (×0.65) → 预期下沿 ~$F
**最近台阶**: $S (建于 [日期区间])   **本段 leg 起点**: $L

| 触发 | 动作 | 依据 |
|---|---|---|
| 跌破 $CATCH1 | 接刀1 · 主仓 50% | 最近台阶 ∩ 实现波动下沿 ∩ put 墙 |
| 跌破 $CATCH2 | 接刀2 · 30% | 下一道 put 墙(仅大盘走弱) |
| 跌破 $CATCH3 | 接刀3 · 20% | 1σ 满幅(罕见,常不成交) |
| 收盘破 $STOP | 止损清仓 | 跌破本段 leg 起点 = 结构破坏 |
| 涨破 $SELL1/2/3 | 各减 1/3 | call 墙 / max pain / ATH 磁吸 |

**怎么挂**: 券商 extended-hours **限价**买单(别用市价),GTC,明早确认仍有效。
**一句话**: [浅跌买台阶 / 深跌看质量 / 破 leg 起点走人]

Then set the ladder as alerts in the private repo ssurmic/invest-watchlist (price-alert/scripts/add_alert.py TICKER below|above PRICE --note "..."), commit + push. NEVER push personal levels to the public claude-investment-skills repo.

---

Worked example + POST-MORTEM (MRVL, 2026-05-27) — the guardrail

• Pre-earnings close $208 (ATH $218, ran +172% in 3mo). Weekly implied ±13.3%.
• My error: set CATCH 1 at $170, anchored to the stale 5/8–5/19 base, and treated the full 1σ ($174) as the expected drop. AH low was only ~$187–189 → never filled, missed nothing (the print was beat+raise so the dip was always going to be shallow).
• Correct: realized ≈ 13.3%×0.65 = 8.6% → floor ~$190; AND the most-recent shelf was $188–190 (5/20–5/22). Both → $190, exactly where it bottomed. The user's "Jacko" call of ~$189 was right because it anchors to the recent shelf.
• Lesson encoded: the mandatory guardrail in Step 3 (shelf ≳ close×(1−implied×0.65), reject anything >1.3× realized below close) makes the $170 anchor impossible.

---

Anti-patterns (DO NOT)

• ❌ Anchor the catch to an old/lower base when a newer higher shelf exists. (The base migrates UP with price — always use the most recent shelf.)
• ❌ Put primary size at the full 1σ implied straddle low. That's a ~16%-probability tail; realized moves run ~0.65× implied. Primary goes at the shelf.
• ❌ Use the pre-earnings weekly option walls after the print (they evaporate).
• ❌ Use the cron alert bot to "catch" the after-hours flush (it reads regular-session price). Use broker extended-hours limit orders.
• ❌ Market-buy the after-hours wick (thin book = terrible fill). Resting limit only.
• ❌ Set the stop at a vol-derived number; set it at the structural leg-start close.

Related skills

• jackal-state-machine — gap magnitude → State (3 pullback / 4 deep / 5 break)
• option-wall-analysis — max pain + walls deep-dive
• jackal-earnings-playbook — the 5-phase intraday earnings choreography
• price-alert — set the ladder as Telegram alerts (private repo)