● Trading System — About

← Back to Dashboard
Status
  • Paper trading only: Every position shown on this dashboard is simulated. No brokerage account is connected and no real money is at risk. Fills, stops, and targets are modelled against real market data, but nothing is ever sent to a broker.
  • Starting balance: $10,000, reset to a clean baseline on September 1, 2026. Cash on hand is derived from the trade history — starting balance minus the cost of open positions, plus realized profit and loss — rather than kept as a running total, so it cannot drift away from the trades that produced it.
  • Evaluation window: Performance is scored over 28 days from the first pick, against a fixed set of criteria — beating SPY, positive return, drawdown under 15%, and a win rate of at least 60%.
Pick Philosophy
  • Two-stage process: A quantitative screener filters the universe down to candidates that clear every gate, then Claude selects the final picks each morning based on technical setup, macro context, and risk. How many it may open is set by the risk budget below, not by a fixed target — and Claude routinely returns fewer than the slots available, which is the intended behaviour rather than a shortfall.
  • Multiple styles: Picks span growth, value, dividend, blue chip, and speculative categories to maintain diversified style exposure across the portfolio.
  • Momentum required: Every pick must be trading above its 20-day and 50-day moving averages, at least keeping pace with SPY over the last 10 trading days, and not trading on dead volume — the last completed session must reach 0.8× the median of the prior 10 sessions.
  • Earnings blackout: No picks are placed within 5 calendar days of a company’s earnings announcement — in either direction — to avoid binary event risk.
  • Minimum 2.5× reward-to-risk: Every pick must offer at least 2.5 times the reward relative to the stop-loss risk before it is considered for entry.
  • Dynamic position sizing: Size is calculated from ATR-14 (Average True Range) and automatically scaled for the current VIX regime — smaller positions in elevated or crisis conditions, full size in normal markets.
  • 25% single-position cap: ATR sizing targets a fixed dollar risk, which makes the position’s dollar value an output rather than an input — a low-volatility name can be sized to 40% of the account while still risking only 1.5% of it. No single position may exceed 25% of the portfolio, whatever the risk math returns.
  • Risk budget sets the position count: The number of positions that may be open is not fixed. Total risk-to-stop across the book — portfolio heat — is capped at 8%, and the slots available on any morning are that remaining budget divided by the current per-trade risk. At the 1.5% per-trade risk of a normal-volatility market that is five slots; in calmer conditions risk rises to 2.0% and the budget buys fewer, in a crisis it falls to 0.5% and buys more — though wider stops and smaller positions bind first there. Capital deployed is separately capped at 80% of the account, and available cash is checked last — heat, then exposure, then cash, so a blocked morning names the limit that actually bound rather than reporting an empty wallet.
  • Dynamic sector concentration: Hot sectors (outperforming SPY by 3%+ over 10 days) can contribute up to 4 picks; weak underperforming sectors are capped at 1 pick to limit concentration risk.
System Overview
  • Universe: S&P 500 + S&P 400 — around 900 unique large- and mid-cap stocks scanned each morning after market open, deduplicated between the two indexes. Both constituent lists come from Wikipedia. The second source was previously an iShares Russell 1000 holdings file that had been failing on every run, silently cutting the universe to 574; the screener now raises an alert if the combined universe falls below 700 rather than screening a truncated list quietly.
  • Stage 1 screener: Hard filters remove illiquid and high-risk names — minimum price ($5), market cap ($2B+), average daily volume (1M+ shares), 52-week performance (not down more than 20%), and beta (under 4.0).
  • Stage 2 screener: Full technical, fundamental, and sentiment scoring — SMA trend filter, relative strength vs SPY, volume confirmation, RSI, ATR, and earnings blackout check. Candidates clearing the minimum score are ranked by how far above their 200-day moving average they trade, and the strongest advance to Claude. That is the only ordering in the system with evidence behind it: across eight years rebuilt to include stocks that later left the index, it beat random selection in seven of eight periods and survived the 2018, 2020 and 2022 drawdowns, while every other scoring component — technical, fundamental and insider — tested no better than chance. It is a relative edge and not downside protection: in falling markets it still loses money, about ten points less than picking at random does. The composite score decides whether a stock qualifies, not where it places.
  • AI-assisted final selection: Up to 60 qualifying candidates are passed to Claude, which analyses each in the context of current market conditions, sector momentum, and portfolio composition to select the final picks. The sector caps described above usually bind before that limit does. Claude may select fewer than the available slots, or none.
  • Risk management: VIX-based regime detection (low / normal / elevated / crisis) automatically adjusts per-trade position size and stop-loss width to keep portfolio risk consistent across volatility environments — from 2.0% of the account risked per trade in calm markets down to 0.5% in a crisis.
  • Stop losses: Set at 2.0× ATR from entry in normal conditions, tightening to 1.5× in low volatility and widening to 2.5× in elevated regime and 3.0× in crisis, so the stop sits outside routine daily noise rather than inside it.
  • Dead-money exit: A position still within ±1 ATR of its entry price after 15 trading days is closed to free the capital. Fifteen days rather than five: targets sit roughly 5 ATR away, so a stock drifting inside 1 ATR for a week is behaving normally on the way there, and exiting that early was closing more than half the book before the thesis had a chance to play out.
  • Nightly self-assessment: After the close, Claude reviews the day’s results and may adjust its own tuning parameters — risk percentages, ATR multipliers, screening thresholds. Every proposed change is bounds-checked before it is written, and the parameters governing paper-versus-live mode cannot be changed by the assessment at all.
  • Weekly review: Every Friday the system re-evaluates all open positions — trailing stops up on winners, recalculating price targets, and closing positions where the original thesis has broken down.
  • GFV protection: Settlement dates are tracked for all recently closed trades. Stop-loss exits that would trigger a Good Faith Violation in a cash account are blocked until sale proceeds have settled (T+1).