Swing Trading Guide: $35K to $8M in 21 Months

Summary

Kevin Xu's breakdown of his swing trading approach -- single stocks, held for days to weeks, built on a small number of high-conviction principles. The core thesis: attention moves price before fundamentals do, and volume is the primary indicator because it measures how many people care about a stock at any given moment.

Key concepts:

• Volume as primary indicator: volume tells you whether enough participants are in the trade to move the price. High volume on a move = real. Low volume = noise.
• Catalysts over fundamentals: a catalyst is a "story that gives hope" -- earnings, FDA approvals, macro events, sector rotation. Catalysts create short-term asymmetry because they concentrate attention.
• Attention → price: price moves when people pay attention. Fundamentals matter on long time horizons; attention matters on swing time horizons.
• Hold period: days to weeks: not day trading (too fast, fees eat you) and not investing (too slow, ties up capital). The sweet spot where catalysts play out.
• Risk management is the strategy: position sizing, stop losses, and knowing when you're wrong matter more than being right.

Agentic Trading System Design

The founder flagged this domain as a potential small bet -- not the specific strategy, but the broader question: what would an always-on agentic trading system look like?

This maps naturally to a data pipeline architecture:

1. Data pipeline — ingest volume data, price action, news feeds, social sentiment, SEC filings. This is fundamentally a [[01-projects/data-marketplace/index|data marketplace]] problem: signal detection is a data product, and the quality of inputs determines the quality of outputs.
2. Signal detection — identify catalysts in real time. NLP on news, volume spikes relative to historical baselines, unusual options activity. Each signal type is a [[06-reference/concepts/compounding-knowledge|compounding knowledge]] asset -- the system gets better at distinguishing real catalysts from noise with every trade.
3. Decision — given a signal, should we act? This is where [[06-reference/concepts/systems-over-goals|systems over goals]] matters most. A systematic decision framework (position sizing rules, entry/exit criteria, risk limits) removes the emotional component that kills most traders.
4. Execution — place the trade. An [[SOUL.md|always-on agent]] reacting to signals in real time has a structural advantage over a human who sleeps, works, and checks charts intermittently.
5. Feedback — did the trade work? Why or why not? Log every decision and outcome.
6. Compound learning — the feedback loop trains the next iteration. This is the real edge: not any single trade, but the system's ability to learn from thousands of trades faster than a human could.

Founder's Caveat

Not endorsing Xu's specific strategy or claims. The $35K→$8M headline is attention-grabbing by design -- survivorship bias is real, and for every success story there are thousands of quiet wipeouts. What's interesting here is the domain characteristics: data-intensive, clear feedback mechanisms, real-time signal processing, and a use case where an always-on agent has a genuine structural advantage over manual execution.

Open Questions

• Feasibility: can a retail-scale agentic system generate alpha, or is this a domain where institutional infrastructure (co-location, proprietary data, market-making relationships) is table stakes?
• Regulatory: what are the SEC/FINRA constraints on algorithmic trading at retail scale? Pattern day trader rules, wash sale implications, reporting requirements.
• Starting small: paper trading first. Build the pipeline, run it in simulation mode for 3-6 months, measure whether the signals have predictive value before risking capital.
• Broker API access: which brokers offer robust APIs for programmatic trading? Alpaca, Interactive Brokers, TD Ameritrade -- what are the tradeoffs?
• Data costs: real-time market data, Level 2 quotes, news feeds, social sentiment APIs -- what does the minimum viable data stack cost per month?
• Edge definition: if the system can't articulate what its edge is (speed? information synthesis? emotional discipline?), it probably doesn't have one.