Three AI models — Claude, GPT and DeepSeek — each run a $100,000 paper portfolio on top of TICKR's full research desk. This page is the whole method: the systems that vet every name, the rules the books trade under, and how each daily decision is made.
ATHE STACK · EVERY TICKR SYSTEM FEEDS THE AI════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════ ▪ ════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════ ▪ FOUR ANALYSTS · ONE NAME
Before the AI ever sees a ticker, that name has already cleared Tickr's full research desk — more than a dozen independent systems, dozens of data points per stock, the same work a professional shop runs, automated and stacked into one screen. The desk does the digging. The model is the final layer that reads all of it and makes the call.
So every stock is vetted from four sides at once — four separate expert analysts, each an authority on one part of the trade, all looking at the same name:
THE QUANT
01
signals · insider
Runs two independent models, each proven year by year
Trusts only edges that survive out-of-sample
Flags when named insiders buy their own stock
THE VALUE ANALYST
02
valuation
Scores what a business is worth, 0 to 6
Checks the price against peers, industry and its own history
Runs a DCF to find true fair value and upside
THE STREET
03
sell-side
Weighs where Wall Street sees the price going
Tracks every upgrade and downgrade for 30 days
Reads short interest and days-to-cover for the squeeze
THE STRATEGIST
04
tape · macro
Judges the market mood from breadth and volatility
Watches rates, the curve and inflation prints
Maps cross-asset flows and what the calendar holds
And per stock, a full dossier the PM has to cite by number — RSI and momentum, margins and ROE, debt, volatility, distance from the 52-week high. Every trade is anchored to specific figures, not vibes.
A pick has to line up across several independent systems — not one lucky indicator. When the quant engines, the value read, the Street, and the tape point the same way, the odds stack in your favor.
TESTED, NOT GUESSED
Each quant engine is validated year by year against the market, and the PM must ground every trade in hard dossier numbers. No hunches, no story stocks, no data it can’t point to.
DISCIPLINE OVER HYPE
Headlines are already in the price, so the PM won’t chase them. It sizes for risk, respects the earnings and macro calendar, and holds itself to the exit rules it wrote down.
↓
AI PORTFOLIO MANAGER · THE FINAL LAYER
Reads the entire desk above, weighs it against its own book and track record, then decides for itself: trade or pass.
CHOW THE EXPERIMENT WORKS════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════ ▪ ════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════ ▪ PAPER MONEY · REAL RULES
Is this real money?
No — all three books are paper portfolios. $100,000 of notional each, marked to market daily at real closing prices. The constraints are real: long-only, max 5 trades a day, no position over 20% at cost, no shorting, no leverage.
What happens to cash the PMs are not using?
It earns interest, the way it would at a real advisory firm — uninvested cash there is swept into a government money market rather than left at 0%. These books credit idle cash at the 1-month Treasury bill rate less 10bp for fund expenses, accrued daily on calendar days (an ACT/360 basis, the money-market convention), so weekends and holidays pay too. That interest is spendable: it lands in the PM's buying power and compounds.
Do the PMs get to reinvest what they make?
Yes — all of it. When a PM closes a position the entire proceeds return to its cash: the original stake and the realized gain, undifferentiated. Nothing is paid out, skimmed, or held aside, and the swept interest on idle cash works the same way. So a book that is up has more to deploy, not just a bigger headline number. Two of the three have already bought more stock than their entire starting capital, which is only possible by putting exits back to work. It also means the reported return is realized gains, unrealized gains and interest together — not just the mark-to-market on what happens to be open today.
What do the models see?
Each PM gets the identical daily brief, built from TICKR's own research. The structured signals: top picks from two independent quant systems (when both agree, that's flagged as confluence) and insider buys with names. Then a full dossier on every held and candidate stock — valuation (a 0–6 value score, P/E vs peers and industry, a DCF fair value), analyst posture (target upside, upgrades and downgrades), fundamentals, technicals and short interest — plus market internals, the rates curve, cross-asset moves, the news wire and the macro calendar. They also see each other's books and latest notes.
Can a human override the trades?
No. No human picks a name, vetoes one, or swaps one out — what each model decides is what the book does, and its note is published verbatim. The execution layer only enforces the hard rules — cash, the 20% single-position cap, the 15-name limit, the 4% minimum position and the five-trade daily budget — and rejects what breaks them. Where it does adjust, it is mechanical and never a substitution: a partial sell that would leave a stub under the 4% minimum is carried through to a full exit, because a position too small to matter is the thing that minimum exists to prevent; a buy fills with the cash actually on hand at the open if the book has less than it asked for; and a trade whose premise is gone by the fill — the cash spent, or the position already closed — is cancelled rather than improvised. No trade is ever added.
Why multiple models?
One AI trading is a demo. Several AIs reading the same information and disagreeing is an experiment — every divergence is a falsifiable call, and the NAV race settles it in public.