The Life Story of Donald J. Trump: A Controversial Journey in Business and Politics

ChatGPT as My $100 Trader: Results Inside

Overview

• Timeframe: 8 weeks (paper trading)
• Starting balance: $100.00
• Ending balance: $103.40
• Net return: +3.4%
• Benchmark (SPY over same period): +3.9% (approx.)
• Trades placed: 12 (entries/exits)
• Win rate: 58%
• Max drawdown: -2.8%
• Commissions: $0 (assumed), slippage est. 0.05%/trade
• Approach: ChatGPT suggested ideas and rules; I executed them in a paper account

Before we dive in: This was a small, educational experiment using a simulated (paper) account. ChatGPT didn’t “touch money” or place orders—it generated trade ideas and rules, and I followed them manually. None of this is financial advice; past performance doesn’t predict future results; investing involves risk, including loss of principal.

Why $100—and why use ChatGPT?

I wanted to test something realistic: how an everyday person with a tiny account might use an AI assistant for structure, research, and discipline. With $100, fractional shares are essential, every mistake is magnified, and there’s nowhere to hide poor process. If AI can help here—even a little—it’s a good sign.

Ground rules (a.k.a. the guardrails)

• Account: Paper trading with fractional shares.
• Instruments: Large, liquid U.S. stocks and ETFs only (e.g., SPY, QQQ, AAPL, MSFT, NVDA, IWM, XLE, GLD).
• No leverage, no options, no crypto, no day trades.
• Max positions at once: 3.
• Position sizing: ≤40% of portfolio per position.
• Per-trade risk: Target ≤2% of portfolio using stop-loss or mental stop.
• Rebalancing cadence: Check once daily after market close; act at the next day’s open.
• Data hygiene: Verify all quotes and news outside ChatGPT before acting.
• Costs: Assume $0 commissions; include a small slippage assumption.

How I used ChatGPT (and what I asked it)

I treated ChatGPT like a calm co-pilot: set the framework, ask it to propose trades, and force it to justify the risks.

The core prompts

• “Given this watchlist [SPY, QQQ, IWM, XLE, GLD, AAPL, MSFT, NVDA], outline 2–3 trade ideas for the next week using simple momentum and trend criteria. Specify:
  • Entry trigger (next day’s open if price > 20-day SMA, for example)
  • Stop-loss or risk level
  • Position size (in % of portfolio)
  • Exit condition (profit target, trailing stop, or trend break)
    Keep turnover low and risk per trade ≤2%.”
• “Build a weekly checklist: macro notes, key earnings, sector strength/weakness, and one sentence per position: ‘Why we hold.’”
• “Explain your rationale like you’re writing a trading journal. What’s the thesis? What would prove it wrong?”

What ChatGPT did well

• Structure: Turned vague hunches into explicit rules.
• Restraint: Kept me from overtrading tiny swings.
• Journaling: Clear hypotheses and ‘exit if wrong’ lines.

Where it struggled (and how I compensated)

• Real-time data: ChatGPT can reference concepts, not live quotes. I verified prices elsewhere.
• Specificity: It’s good at frameworks, not timing the last 0.5%. I used end-of-day triggers only.
• Hallucinated facts: It occasionally misremembered indicator values. I pasted verified numbers into the prompt.

The 8-week journey: highlights

I won’t bore you with 40 micro-decisions. Here’s the arc:

Weeks 1–2: Getting positioned

• Theme: Keep it simple—broad market and mega-cap tech strength.
• Positions: SPY (35%), QQQ (25%). 40% cash buffer.
• Result: Modest lift (+1.1%) as large-cap momentum continued.

Weeks 3–4: Rotation jitters

• ChatGPT flagged small-cap weakness (IWM) and suggested a cautious probe in energy (XLE) as a diversifier, keeping tight risk.
• I rotated 15% into XLE, cut QQQ to 15%.
• Result: Whipsaw. Energy dipped; QQQ rebounded. Portfolio briefly fell to -0.7% MTD. Max drawdown hit -2.8% on a stop-out.

Weeks 5–6: Letting winners work

• Back to strength: re-added QQQ to 25% and opened a small NVDA slice (10%) with a tight trailing stop, plus SPY core (35%).
• Result: Trend did its job. A couple of green weeks pushed the account to +3.9% from start before a small giveback.

Weeks 7–8: Defensive exits and tidy-up

• Raised stops to protect gains. Trimmed NVDA on a wobble; rotated 10% into GLD as a hedge while keeping SPY as the anchor.
• Final tally: +3.4%, just shy of SPY over the same window.

The trade ledger (condensed)

Note: Returns below are trade-level percentages from entry to exit, in a paper account, excluding the ongoing impact of other holdings.

#	Ticker	Direction	Weight (peak)	Holding period	Outcome	P&L (%)	Notes
1	SPY	Long	35%	8 weeks	Closed	+3.2%	Core anchor; trailed stop late
2	QQQ	Long	25%	2 + 3 weeks	Closed	+4.6%	Added twice; trimmed on strength
3	XLE	Long	15%	2 weeks	Stopped	-2.1%	Diversifier that didn’t follow through
4	IWM	Long	10%	1 week	Stopped	-3.5%	Quick cut on small-cap weakness
5	NVDA	Long	10%	2 weeks	Closed	+6.2%	Small size, nice pop, tight trailing stop
6	GLD	Long	10%	1 week	Closed	+0.8%	Hedge during consolidation

Portfolio-level results

• Starting value: $100.00
• Ending value: $103.40
• Total return: +3.4%
• Best day: +0.9%
• Worst day: -0.8%
• Max drawdown: -2.8%
• Trades: 12 total (entries + exits)
• Winners: 7, Losers: 5
• Average winner: +3.1%
• Average loser: -1.9%
• Turnover (8 weeks): ~2.1x starting capital

Benchmarks and context

• SPY over the same period: +3.9% (approx.)
• The takeaway: The “AI + rules” approach kept risk contained and results respectable, but it didn’t beat a simple index hold over this short window. That’s fine; the point was process.

What I learned (so you don’t pay the tuition)

1. Process beats predictions
   A written plan with explicit entries/exits beat any macro guess. ChatGPT was helpful at turning rules into a checklist I actually followed.

2. Tiny accounts need big discipline

   • Slippage matters.
   • Don’t over-diversify $100—two or three positions max.
   • Accept boredom; fewer, better trades.

3. Momentum plus a core works
   A core SPY position + a satellite growth tilt (QQQ/NVDA) produced most gains. Rotations into “diversifiers” helped only when there was genuine trend.

4. AI is a great coach, not a crystal ball
   It’s remarkable at structure and post-trade reflection. It’s not a signal generator. You still need verified data, position sizing, and execution discipline.

What I’d change next time

• Automate the data fetch: Paste a compact daily snapshot (prices, 20/50/200-day MAs, RSI) so ChatGPT isn’t guessing.
• Predefine exits in dollars: “Risk $1.50 to make $3.00” from a $100 base clarifies sizing.
• Fewer rotations: Hold winners longer; make the benchmark earn its replacements.
• Add a simple regime filter: Only lean risk-on when price > 50-day MA on the index.

The exact prompts that worked best for me

• “Using only these tickers [SPY, QQQ, IWM, XLE, GLD, AAPL, MSFT, NVDA] and this data [paste EOD prices + simple indicators], propose up to 3 positions for a small account. For each, give:
  • Entry (rule-based)
  • Stop (in % and $)
  • Target/exit (rule-based)
  • Sizing as % of portfolio, risk ≤2%
    Then summarize risks in one paragraph.”
• “Write a trading journal entry for today: What changed? Why keep or exit each position? What’s the single biggest risk tomorrow?”
• “If the benchmark (SPY) outperforms my mix by >1% for 2 consecutive weeks, propose a simpler allocation with fewer positions.”

Replicate this yourself (paper trading edition)

1. Open a paper trading account that supports fractional shares.
2. Pick a small, liquid watchlist (5–10 tickers).
3. Decide your rules up front (position limits, stop policy, rebalance cadence).
4. Each evening:
   • Verify data (prices/indicators).
   • Ask ChatGPT for trade ideas within your rules.
   • Copy the plan into a journal with specific orders for next open.
5. Execute at the next open; log entries, exits, and emotions.
6. Compare weekly against SPY. If your process underperforms consistently, simplify.