Its favorite trade is no trade.
A cautious trading-research and approval workflow agent. It researches, backtests, and paper trades conservative daily-bar trend-following ideas, then prepares human-reviewed trade tickets for connected brokerage workflows. Built to behave like a disciplined quant assistant, not a gambling bot.
How it works
One cycle per day, after the close. It reads the market, manages open risk, looks for a high-quality setup, and — most days — recommends doing nothing. For public use, it should start in paper mode and graduate only through explicit review.
Daily bars for a broad-equity watchlist. Stale or missing data means the asset is simply not traded that day.
No new longs unless the broad market is healthy — benchmark above its 200-day average and breadth intact.
Trend structure, volume, volatility band, breakout or pullback, an ATR stop, and a minimum 2:1 reward-to-risk.
Risk-based sizing, single-asset and correlation-cluster caps, and a haircut for positions that double up on the same risk.
It renders the full trade ticket with reasoning, sizing, and risk notes. No review, no action.
Every signal — including explicit "avoid" decisions — and every trade is logged with full reasoning.
Risk rules · enforced in code
Conservative defaults sit on top of absolute ceilings written into the risk modules. Code wins over config — a typo can never widen a limit past its hard cap.
| Risk per trade | max 1% of equity |
| Daily drawdown | halt at 5% |
| Total drawdown | shutdown at 10% |
| Single-asset allocation | max 20% |
| Correlation cluster | max 50% |
| Open positions | max 3 |
| Reward / risk | min 2:1 |
| Confidence score | min 70 / 100 |
| Stop loss | mandatory — no stop, no trade |
Access model
JerryQuant can become a paid agent workflow for other builders or operators, but the clean version is simple: charge for access to research, setup, alerts, paper reports, and review tooling. Do not charge as if it guarantees trades or returns.
A small wallet-native fee or subscription can unlock the dashboard, reports, and setup support.
Every new user should begin with paper mode so the agent proves its behavior before brokerage access.
Users connect their own accounts and keep custody. The agent should never receive withdrawal permissions.
Approval gates, logs, and clear disclosures keep the product a tool, not an unmanaged money manager.
Edge controls
Trend-following bleeds in bear and choppy tapes. New longs are blocked unless the broad market is in good health — the gate only ever sits the agent out, it never forces a trade.
A chandelier trailing stop lets winners run, a time stop frees dead capital, and partial profit-taking banks gains at a set R multiple, then ratchets the stop to breakeven.
Names that move together share one cap, and a highly correlated new position is sized down — so three look-alike bets can't quietly become one triple-sized one.
Walk-forward out-of-sample windows, a parameter-sensitivity sweep to catch curve-fitting, and a bootstrap Monte Carlo of returns and drawdowns. Pure analysis — it never trades.
Safety