pead-tracker

PEAD Tracker (Post-Earnings Announcement Drift)

Stage 0 - Decision Memory (load FIRST)

Before analysis, load prior decisions so verdicts stay consistent across sessions:

• mcp__aifolimizer__get_cross_ticker_lessons with max_lessons=3 - portfolio-level win/loss patterns
• For any name you issue a per-ticker BUY/SELL/TRIM/HOLD on, also load mcp__aifolimizer__get_ticker_decision_history (ticker=…, max_decisions=5) and mcp__aifolimizer__get_ticker_reflection (symbol=…, n=3).

Reconciliation rule: if a prior decision exists and your new read flips it, state explicitly WHY it changed (new data / catalyst / price move). Never silently contradict a logged decision - that drift is exactly what this prevents.

Research basis

Bernard & Thomas (1989): stocks keep moving in direction of earnings surprise after earnings are public. Drift window: ~60 trading days (~85 calendar days) from report date. Expected abnormal return by size: small firms +5.1%, mid +4.3%, large +2.8%.

How to run

1. Call mcp__aifolimizer__get_profile - account types and capital context
2. Call mcp__aifolimizer__get_portfolio - full holdings list
3. Call mcp__aifolimizer__get_earnings_results with all held symbols, quarters=2 - get report dates, surprise %, outcome for last 2 quarters
4. Call mcp__aifolimizer__get_fundamentals for all symbols - market cap (determines expected drift magnitude), analyst target
5. Call mcp__aifolimizer__get_technicals for symbols with active drift - pct_from_52w_high, RSI, trend to assess whether drift is still running or exhausted

Call steps 3-5 in parallel after step 2 resolves.

Drift window logic

Use ONE clock - calendar days - throughout, to avoid mixing trading-day and calendar-day units. The ~60-trading-day Bernard-Thomas window ≈ 85 calendar days; we measure everything against that 85-day calendar window.

For each holding with a recorded earnings report:

• Compute calendar days since report date (use today's date)
• Active window: 0-85 calendar days since report
• Late window: 55-85 calendar days (drift fading - last chance to ride or exit)
• Expired: > 85 calendar days (no PEAD edge remaining)

Drift direction:

• outcome == "beat" + positive surprise_pct → positive drift expected, bias HOLD/ADD
• outcome == "miss" + negative surprise_pct → negative drift expected, bias TRIM/EXIT
• outcome == "meet" → no directional drift signal

Output structure

1. Drift summary table

Render markdown table. Columns:

Ticker	Report Date	Days Since	Surprise %	Outcome	Drift Window	Expected Edge	Action

• Expected Edge: use market cap to estimate remaining drift. Large-cap (>$100B): 2.8% over full window. Mid ($2B-$100B): 4.3%. Small (<$2B): 5.1%. Pro-rate on the same 85-day calendar clock: days_remaining = max(0, 85 - calendar_days_since), then edge_remaining = full_edge × days_remaining / 85. The max(0, …) clamp means an expired name shows 0% remaining edge, never a negative number.
• Action: ride (beat, early window), exit soon (beat, late window), trim (miss, any window), flat (meet or expired)

2. Active plays (≤5 bullets)

Only holdings still in drift window. Format per line:

TICKER · beat/miss X.X% surprise · Day N of ~85 (calendar) · ~Y% drift remaining · action

3. Expired positions check

List holdings where PEAD window just closed (85-100 days). Note: "PEAD edge gone - hold/trim on fundamentals only."

4. No-data gaps

List symbols where get_earnings_results returned empty or null dates. "TSX coverage gap" if .TO ticker. Do not fabricate dates.

5. Recommendation (Canadian tax-aware)

For each ride or trim action that conflicts with current portfolio weight:

• Which account to act in (TFSA > Non-Reg for gains, Non-Reg for harvesting losses)
• Max position cap: if adding via PEAD signal, cap at 5% incremental add - this is a momentum overlay, not a conviction change

After output - log decisions

For each Active play with action ride (ADD) or trim/exit soon (TRIM/EXIT):

• If the action is ride (ADD), first call mcp__aifolimizer__get_positioning_signals with symbols=[ticker]. If crowding_score >= 70 the drift is already consensus-crowded (late entry = negative expected alpha) - downgrade to HOLD or cap the incremental add hard; this is a momentum overlay, not a conviction buy. Favor names with crowding_score <= 30.
• Optional drift confirmation: call mcp__aifolimizer__get_earnings_commentary with ticker=<ticker>. Use mgmt_tone_trend (relative vs prior quarters), NOT the absolute tone. improving after a beat (or deteriorating after a miss) = the language confirms the drift direction → keep conviction. Divergence (beat but deteriorating tone) → downgrade one notch. Skip if relative is false or source is None.
• Then call mcp__aifolimizer__log_recommendation with skill="pead-tracker" (the param is skill, not skill_used - that belongs to log_trade_decision), ticker, action (ADD/HOLD/TRIM/SELL), conviction (HIGH/MED/LOW per surprise magnitude + days remaining), target_pct + stop_pct (% from entry - the schema takes percentages, not absolute prices), rationale (1-line citing surprise % + drift days remaining). Skip flat (no edge). Feeds forward win-rate / track-record loop.

Rules

• Always use get_profile first - never hardcode accounts or capital
• Never fabricate report dates - only use dates from get_earnings_results
• Expected drift is statistical, not guaranteed - present as probabilistic edge, not certainty
• Under 500 words
• If fewer than 3 holdings have earnings results in last 85 days, say so explicitly - don't pad

Gotchas

• get_earnings_results returns yfinance earnings_history - has date field but TSX (.TO) tickers sparse. Note gap, don't fabricate.
• "Days since report" requires comparing report date to today. Use currentDate from context. Do not assume.
• Large-cap beats often get priced faster (institutional speed) - 2.8% drift estimate is conservative and may already be gone by day 5 for mega-caps. Flag if stock already ran >2% since report.
• Miss + negative drift = same logic inverted. Don't add to a PEAD-miss name just because you like the company.
• surprise_pct from yfinance is EPS-only. Revenue miss can override EPS beat (stock sold off despite beat). Cross-check get_news_headlines if price moved opposite to EPS outcome.
• get_technicals cached 1h - RSI/trend signal may lag. Mention cache timestamp.
• PEAD effect strongest in first 20 trading days post-report, fades after. Strongest entry = day 1-14.
• Crypto holdings: no earnings. Skip. List in section 4.