Earnings Event Trader

Earnings Event Trader

A single, specific setup: a quality name beats EPS but the market punishes it anyway. The "beat-but-down" reaction is empirically a slow-fade — the price often recovers over the next several days as the noise (guidance, one-line items, sector rotation) gets parsed. This strategist takes that bet.

Required: _shared/intent-schema.md, _shared/circuit-breakers.md. See references/decision-rules.md.

Strategy summary

• Universe: 10 mega-cap quality names from allowlist.
• Trigger: symbol reported earnings in the prior session (AM today or PM previous trading day), actual_eps > estimated_eps, current price gapped DOWN ≥ 2% from prior close.
• Entry: buy at next regular-hours open or current price (limit at +0.1%).
• Exit: +3% from entry OR 5 trading days, whichever first.
• Sizing: fixed 25% of buying power.
• Cadence: daily, once (10am ET tick).

6-phase protocol

Phase 1 — Pre-flight (strategist-side)

Read ~/.hermes/profiles/autotrader/state/earnings-event-trader.json. If today's date already has an entry attempt logged → skip to phase 4 (exit logic for any open positions).

Phase 2 — Compose context

regime-intelligence with: "Is this an earnings season where beats are getting rewarded, or punished broadly? One line."

special-situations with: "Any quality names that reported recently with notable management commentary worth flagging?"

Both into notes. The regime response can suppress entries (if response includes "beats_punished_broadly", skip phase 4 entries entirely — the strategy thesis is broken).

Phase 3 — Signal gather

calendar = get_earnings_calendar(start_date=<yesterday>, days=1, filter="high_market_cap")

For each entry in calendar.results where symbol in allowlist:

• Pull recent EPS: get_earnings_results(symbol=symbol) — read most recent quarter's actual_eps and estimated_eps.
• Pull recent prices: get_equity_quotes(symbols=[symbol]) for current price and prior session close.

Compute:

beat = actual_eps > estimated_eps
gap_pct = (current_price - prior_close) / prior_close * 100
trigger = beat AND gap_pct <= -gap_down_threshold_pct

Phase 4 — Decide

Exits first (any open earnings positions):

for pos in open_positions:
    current_price = quotes[pos.symbol].last_trade_price
    pnl_pct = (current_price - pos.entry_price) / pos.entry_price * 100
    age_days = today - pos.opened_at (trading days)
    if pnl_pct >= exit_target_pct:
        emit sell intent; reason: "target_hit"
    elif age_days >= exit_max_days:
        emit sell intent; reason: "max_age"

Entries (only if context didn't say beats_punished_broadly):

for symbol in triggered_symbols:
    if symbol in open_positions: skip
    notional = buying_power * 0.25
    qty = floor(notional / current_price)
    if qty == 0: skip
    emit buy intent (limit, limit_price=current_price*1.001, reason: f"beat ${actual} vs ${est}, gap {gap_pct:.1f}%")
    break  # one entry per tick (max_concurrent_positions enforced)

Phase 5 — Hand off to executor

Phase 6 — Emit & persist

Log entry attempts (regardless of fill) so phase 1 next tick can short-circuit.

Interactive mode

• "Any beat-but-down setups today?" — run phases 1-4 in review mode, report what triggered (zero is the common answer).
• "Did NVDA beat?" — call get_earnings_results(symbol="NVDA"), report.

What this strategist will never do

• Trade on a miss. Beats only.
• Trade on a gap UP. The fade is one-directional.
• Buy pre-earnings. The strategy is post-event.
• Hold past 5 trading days. Force exit.
• Trade outside allowlist.