I Let Claude Code Build a FREE 5-Minute Scalping Bot… The Results Were Insane

Finance-Focused Summary (AI Scalping Bot)

Experiment Setup (AI Scalping Bot)

• The creator gives Claude (Claude Code) model Opus 5 access to a real trading account with a ~2.5k USDT sub-account on Bybit.
• The mission spans 2 hours total, split into:
  • Phase 1 (Hour 1): find high-probability trades / research
  • Phase 2 (Hour 2): execute trades using a 5-minute scalping approach

Tools Integrated via MCP Servers

• Trader Kit
  • Used to install/backtest TradingView strategies and return strategy code.
• Custom MCP tool for TradingView’s library
  • Provides Claude access to TradingView’s built-in strategy/indicator library
  • Intended to reduce reliance on overly common “default” indicators (example: avoiding typical RSI and simple moving averages/EMAs setups)

---

Assets / Instruments Mentioned

• DEX USDT (live trade symbol mentioned)
• “BANK long USD” (described as “bank long USD”; ticker not clearly specified)
• BTC USDT (used as a target for the 1-hour timeframe strategy search)
• USD (pairing context for “bank long USD”)
• General references to Forex, Gold, and crypto coin pairs (coverage claims)

---

Backtesting Results (TradingView-Style)

Reported performance:

• Return / P&L: 1,968%
• Max drawdown: just under 30%
  • Later improvement claims mention drawdown reduction (see below)
• Number of closed trades: 111
• Profit factor: just above 2
• Equity curve: described as strong, with a recent drawdown and recovery

---

Strategy Logic / Methodology (TradingView Indicators + Rules)

Core Trend Direction: T3 Indicator

• If price is below T3 → look for shorts
• If price is above T3 → look for longs

Entry Timing / Momentum: Squeeze Momentum Indicator

• Uses oscillator behavior (green/black cross behavior noted)
• Black crosses are described as related to entry confirmation
• Framed as identifying likely breakout conditions (subtitles relate this to Bollinger Bands behavior)

Volatility Filters (Sideways-Market Avoidance)

• Shaking Volatility
  • Rule described: take entries when it is above 0
• Change of Volatility
  • Rule described: take entries when it is above 0

ADX Filter

• Rule described: ADX to be red above blue
• Exact parameter values not specified

Stops / Trailing Management (Stop ATR)

• Stop ATR is used in a trailing-stop manner:
  • described as setting the stop to the low of the indicator
  • used as a trailing stop loss
• Creator’s caution/disclaimer:
  • The creator generally dislikes trailing stops due to TradingView alert/execution delays
  • Wicks can worsen exits if signals/alerts lag
  • Preference: execute stops on the exchange (e.g., market orders) where possible to reduce “alert delay” risk

---

Claimed Incremental Improvements (Indicator Additions)

Claude’s improvements to the strategy (with reported best-setting comparisons):

• Max drawdown: 37% → ~21%
• Profit factor: 0.8 → 2
• Trades: 208 → 106
• Cause attributed to adding:
  • Change of Volatility
  • Shaking Volatility

---

Live Trading Results (Competition Run ~1 Hour)

Live trades executed:

• Total trades: 4
• Win rate: 100% (won every trade, small sample)

Reported outcomes:

1. DEX USDT long
   • Held ~8 minutes
   • Profit: $1.01
2. Volume breakout
   • Held ~6 minutes
   • Profit: $8.27
3. “Bank long USD”
   • Used regime expansion
   • ATR tripled
   • Volume spiked to 7.16× normal volume
   • Volume then “died”; trade nearly erased profit
   • Closed near-flat: profit $0.33
   • Claude opened a second trade after a pullback:
     • Held ~10 minutes
     • Profit: $0.33
4. Final trade
   • Profit: $5.71
   • Held duration not specified in the subtitles

---

Cautions / Recommendations from the Creator

• Despite strong backtest and live outcomes, the creator warns not to get overexcited:
  • They would not live trade this strategy based only on backtest quality.
  • Recommend forward testing for 1–3 months
  • Then test on a smaller account before scaling
• Skepticism about AI/LLM trading profitability:
  • Claim: 95–99% of traders lose money, implying poor odds for most approaches
  • Concern: LLMs may be trained on Reddit/books data, creating mismatch with live market conditions
• Trailing-stop risk:
  • Notes prior experience where TradingView alerts did not exit fast enough during wick/volatility events

---

Quantitative Risk Notes

• Early mention of live sizing/leverage:
  • Collateral used: $50
  • Leverage: 9x
• Backtest drawdown:
  • ~30% in a single-run report
  • Later improvement claims cite ~21% under “best settings”

---

Disclosures / Disclaimers

• No formal “not financial advice” line appears in the subtitles.
• Informal risk disclaimer (“real talk”):
  • The creator would not put AI their money under these conditions
  • Requires forward testing (1–3 months) and small-account verification before scaling

---

Presenters / Sources Mentioned

• Claude (Claude Code) — model “Opus 5”
• TradingView — strategies/backtesting platform
• Bybit — exchange/account used
• Indicator/tool authors referenced:
  • Lazy Bear — Squeeze Momentum indicator
  • David Tech — Shaking Volatility indicator
  • BJR117 — Change of Volatility indicator
• Creator/speaker: David (referred to as “Come on, David.”)