The strategy that keeps MAKING BILLIONS to INSTITUTIONAL traders: PEAD.

Finance-focused summary (PEAD: Post-Earnings Announcement Drift)

Presenter / context

• Marco Conti: ex-market maker (7 years) who claims “more than 30 million euro” profit for a large investment bank; now runs an automated trading-focused hedge fund/channel.
• Topic: PEAD (Post-Earnings Announcement Drift)—a widely documented market inefficiency where prices continue drifting after earnings.

---

Market inefficiency explained (PEAD)

PEAD definition

• After an earnings release, the stock price continues to drift in the direction of the earnings surprise for several weeks:
  • If earnings beat consensus → stock tends to keep going up.
  • If earnings miss → stock tends to keep going down.

Duration

• Drift lasts roughly ~60 trading days (per cited research).

Why it persists (under-reaction / frictions)

• Investors update models gradually over weeks (institutions) and retail reacts emotionally to headlines.
• Sell-side analysts revise estimates over days.
• Limits to arbitrage include:
  • Idiosyncratic risk
  • Transaction costs
  • Risk manager exposure caps

---

Academic research timeline cited

• 1968: Ball & Brown introduced the phenomenon through accounting income evaluation (early evidence markets don’t fully react immediately).
• 1989: Bernard & Thomas formalized drift using an earnings-surprise sorting framework:
  • Positive abnormal returns in 41 out of 48 quarters
  • Average drift around ~60 days
• 2006: Livnat & Mendenhall updated methodology comparing analyst forecasts vs time-series forecasts, and defined standard entry/exit windows used in modern implementations.
• 2021: “Thing” reviewed 200+ papers across 50 years, concluding “the effect persists globally.”

---

Trading strategy framework (extracted rules)

A “complete trading strategy” is described as:

Entry condition (concordant signal)

1. Compute earnings surprise = (reported EPS − consensus estimate)
2. Require the price reaction direction around the earnings event.
3. Concordant means the surprise direction matches price reaction direction:
   • Surprise above consensus + stock reacts positively → Long
   • Surprise below consensus + stock reacts negatively → Short
   • Disagreeing surprise/reaction → No trade

Timing / event mapping

• If earnings are announced after market close:
  • reaction day = next trading day
• If earnings are announced before market open:
  • reaction day = same day (market opens already reflecting news)
• Enter at the open of the following session after the full reaction has played out.

Exit condition

• Time-based exit only:
  • Hold for 60 trading days, then close.
• Baseline includes no stop-loss and no take-profit.

Position sizing

• Fixed fraction:
  • 10% of capital per position
  • Example: $100,000 account → $10,000 per position
• Rationale: avoid excessive single-stock exposure.

---

Data and tooling approach

Price data

• Daily data from TradeStation (used within MultiCharts per subtitles).

Earnings data fields required (4 columns)

1. Announcement date
2. Timing flag (before open vs after close)
3. Reported EPS
4. Consensus EPS estimate

• Surprise calculated as: reported − estimate

Free earnings source

• Zacks.com (includes estimates, surprises, and timing flag)

Backtest universe

• 20 US stocks across sectors: tech, consumer, financials, healthcare, industrial, energy
• Explicitly mentioned names:
  • Nike (NKE)
  • Starbucks (SBUX)
  • Eli Lilly (LLY)
• Example ticker used for walkthrough:
  • Google / Alphabet (GOOGL/Google) (ticker not perfectly specified in subtitles, but “Google” is used)
• Period: Jan 2018 to Apr 2026
• Total events: 658 earnings events across 20 stocks

Code generation method

• Uses large language models (explicitly mentions ChatGPT/Claude) to generate a single-script backtest.
• Claims it’s not “clean/efficient,” but fast for research.

---

Backtest results (key numbers)

Baseline results mentioned (configuration comparisons)

Configuration 1: entry + exit per research; concordant filter used

• Net profit: $131,000 on $100,000 capital
• Max drawdown: 17.4%
• Profit factor: 1.99
• Win rate: 61%
• Annualized return: ~12%
• Calmar ratio: ~0.68
• Interpretation: “working strategy” on a diversified basket; no optimization.

Configuration 2: remove concordant filter (direction from surprise only)

• Net profit: $251,000
• Max drawdown: 32.5%
• Calmar: 0.69
• Takeaway: higher returns, substantially worse risk.

Configuration 3: concordant off + filter small surprises (5% magnitude threshold)

• Net profit: $191,000
• Max drawdown: 26%
• Calmar: 0.67
• Takeaway: focusing on larger surprises helps vs conf2, but still not as good as the “best” version.

Configuration 4: both filters on (concordant + 5% threshold)

• Net profit: ~$82,000
• Max drawdown: 20%
• Calmar: 0.36
• Key insight: stacking filters broke performance; suggests partial overlap, and that removing “small-but-high-quality” signals can hurt.

Configuration 5: long-only (remove short leg), concordant filter on, no surprise threshold

• Net profit: $146,000
• Max drawdown: 20%
• Profit factor: 2.68
• Win rate: 63%
• Described as the “cleanest by the book version”
• Conclusion: PEAD still works, but shorts underperform.

---

Explicit asymmetry / short-leg behavior

• Short trades consistently lose:
  • Short leg P&L always between −$14,000 and −$19,000 (per subtitles).
• Given reasoning:
  • Positive surprises drift more reliably than negative ones (especially post-2010, large caps)
  • Bad news may be pre-announced; management guidance may front-run the miss; activists may short ahead of release
  • By release time, the negative reaction is partially priced in

---

Disclosures / cautions mentioned

• No explicit “not financial advice” disclaimer appears in the subtitles provided.
• Implicit caution: earnings data quality may be lower when using free sources vs paid providers.

---

Instruments / tickers mentioned

• Nike (NKE)
• Starbucks (SBUX)
• Eli Lilly (LLY)
• Google / Alphabet (GOOGL/Google) (used as an example; ticker details not perfectly specified)
• Broader US basket of 20 stocks across sectors (other tickers not named in subtitles)

---

Key performance metrics explicitly cited

• Profit: $131k, $251k, $191k, ~$82k, $146k
• Drawdowns: 17.4%, 32.5%, 26%, 20%
• Profit factor: 1.99, 2.68 (best long-only configuration)
• Win rate: 61%, 63%
• Annualized return: ~12%
• Calmar ratio: ~0.68, 0.69, 0.67, 0.36
• Also noted: “0.66 to 0.76 depending on configuration” (edge range)

---

Methodology / framework recap (step-by-step)

1. Collect daily prices + earnings event data (date, timing, reported EPS, consensus EPS).
2. Compute earnings surprise = reported EPS − consensus EPS.
3. Determine price reaction direction around the earnings announcement.
4. Entry
   • If earnings surprise direction == price reaction direction:
     • Long on positive surprise & positive reaction
     • Short on negative surprise & negative reaction
   • Else: skip (in the concordant version).
5. Timing rule
   • Enter at the open of the next relevant session after the market incorporates the earnings info (accounts for before-open vs after-close releases).
6. Exit
   • Hold 60 trading days, then close.
   • Baseline includes no stop loss / take profit.
7. Position sizing
   • Use 10% of capital per trade (example: $100k account → $10k per position).

---

Presenters / sources (mentioned at end)

• Presenter: Marco Conti
• Academic sources cited:
  • Ball & Brown (1968)
  • Bernard & Thomas (1989)
  • Livnat & Mendenhall (2006)
  • “Thing” (2021) (spelled in subtitles)
• Data source mentioned: Zacks.com
• Trading platforms mentioned: TradeStation, MultiCharts, TradingView
• LLM tools mentioned: ChatGPT and Claude (via “ChatGPT or Claude”)