Stop Guessing Your Daily Bias. Measure It.

Finance-focused summary (intraday bias research + trading framework)

Core concept: “intraday bias”

The video explains intraday bias—specific time windows within a trading day where markets historically tend to drift up or down on average. This is derived from historical measurement, not from “guessing” daily direction.

---

Presenter / credentials (in-video)

• Mati
  • Claims 7 years as a market maker
  • Claims making more than €30 million for a large investment bank
  • Now runs a hedge fund that automates trading strategies

---

Markets / instruments mentioned

• S&P 500 futures (used in the live example)
• Additional examples mentioned (no tickers clarified):
  • NASDAQ (spelled “NASB” in the summary)
  • Oil
  • Bitcoin
• No individual stock tickers were mentioned.

---

Timelines / data span / session assumptions

• Data loaded: Jan 1, 2018 → Sep 28, 2026
• Timeframe: 60 minutes
• Session: full 24-hour session
• Time references:
  • Cash open: 9:30 Eastern
  • Mentions European close (no exact time given)
  • Mentions an overnight session with fewer participants
• Strategy calculations repeatedly convert to Central Time.

---

Measurement methodology (step-by-step framework)

1) Define a measurement “window” trade

Input: a test time (initially 10:00 a.m. Central)

Rule:

• If bar time == test time:
  • Buy at the open of the next bar
  • Sell at the open of the bar after that
• Hold duration: exactly 1 hour
• No stops / no targets (explicitly for measurement, not a final strategy)

---

2) Optimize across all hourly windows

Optimization idea:

• Test entry hours from 0 (midnight) to 23
• The summary notes “steps of 100” (likely a scripting/rounding artifact); the intent is testing each hour across the day.

Output:

• One result row per hour tested
• Evaluates going long for 1 hour starting at that hour

---

3) Convert bias into a long/short strategy

• Use “better hours” as longs
• Use “worse hours” as shorts

Explicit heuristic example:

• “Only go long if the average trade is above 5 or 6” (example threshold shown as 6)

Example pairing logic:

• Long: a clustered period such as midnight → 3:00 a.m.
• Short: a worst-performing window such as 14:00 → 3:00 p.m.
• Later emphasis in the summary:
  • short at 9:00 a.m. Central

---

4) Add risk management (stops/targets) and optimize

• First pass: no stop-loss / take-profit
• Then add:
  • Stop-loss (example tested range: $0 to $10,000)
  • Take-profit (example tested range with steps of 500)
• Optimization reportedly yields 441 combinations.

Example “best” combination reported:

• Stop-loss = $4,500
• Take-profit = $2,000
• Mentions a top metric such as profit factor = 10.8 (subtitles read “108”)

---

5) Prevent overfitting via in-sample / out-of-sample split

Warning: picking best hours across the entire dataset can lead to overfitting—great backtest performance that fails live.

Proposed split:

• In-sample: Jan 1, 2018 → Dec 31, 2024
• Out-of-sample: Jan 1, 2025 → end (subtitles suggest it continues to end of 2026)

---

6) Validate the bias

The bias is treated as “real” if it works:

• In both in-sample and out-of-sample
• With similar equity curves
• With similar average trade
• With similar % winners

If out-of-sample is weak, the video’s guidance is not to assume future profitability.

---

Key findings / example numbers (from the example)

Hourly drift examples (1-hour hold, long trades; ~8-year backtest)

• Long at 00:00 (midnight): profit ~$13,650
• Long at 01:00: profit ~$21,000
• Consistent profits cited for going long from 22:00 to 06:00
  • Framed as reflecting the overnight anomaly / overnight drift
• Long at 09:00 a.m.: losses around ~$30,000
• Other poor-hour candidates mentioned:
  • 14:00
  • 21:00

---

Time-rotation strategy example (no stops/targets)

Based on intraday bias mapping:

• Net profit: ~$75,000
• Profit factor: ~1
• Percentage profitable: 49.22%
• Average trade (before cost): ~1672
• Mentions equity curves comparing long-only vs short-only components.

---

Risk management tuning (example results)

• Introducing stops/targets reduces net profit (intermediate step values not fully specified)
• “Best” combination from optimization:
  • Stop-loss $4,500
  • Take-profit $2,000
  • Mentions profit factor 10.8

---

Overfitting test outcome

• Optimizing over the full/entire period appears good initially
• But out-of-sample performance weakens
• Therefore, the selected hour-set is not robust

---

Explicit recommendations / cautions

• Don’t rely on gut feeling (“not just what the gut feeling is telling you”)
• Avoid overfitting:
  • Don’t optimize across all years and assume it will work live
  • Use in-sample vs out-of-sample validation
• Construct the bias strategy using the heuristic:
  • Only consider longs when average trade > 5–6 (example threshold: 6)
• Risk management is staged:
  • Start with measurement (no stops/targets), then add stops/targets and optimize

---

Disclosures / disclaimers

• The provided summary indicates no explicit “not financial advice” disclaimer appears in the referenced subtitles.

---

Presenters / sources

• Presenter: Mati
  • Market maker for 7 years, runs an automated hedge fund, builds a YouTube channel
• Referenced sources (general):
  • “Academic research” on intraday effects
  • Mentions intraday momentum and overnight anomaly
  • No specific paper titles are cited in the summary