The shelf
Clients shop around
19%
of high-net-worth individuals use a single wealth firm, down from 39% in 2019. Access to something distinctive decides who keeps the relationship.
Capgemini, World Wealth Report 2026
Prepared for WELREX · September 2026
Proven investment concepts, hard evidence and AI-accelerated research, built into strategies your relationship managers can offer and explain.
Confidential · For investment professionals only
The challenge
The shelf
19%
of high-net-worth individuals use a single wealth firm, down from 39% in 2019. Access to something distinctive decides who keeps the relationship.
Capgemini, World Wealth Report 2026
The evidence
−73%
median fall in Sharpe ratio when 215 bank-built strategies moved from backtest to live trading. A strong backtest is common. One that survives is rare.
Suhonen, Lennkh & Perez, Journal of Portfolio Management, 2017
The explanation
46%
of UK financial firms say they only partly understand the AI they use. Your RMs still have to explain every holding to the client.
Bank of England & FCA, AI in UK financial services, 2024
VantageVest was built around these three problems
Who we are
Financial statements, valuation, capital allocation, earnings surprises
Price trend, momentum, market breadth, volume
Volatility, drawdown, correlation, regime statistics
Each model weighs its evidence into one calibrated score. Testing sets the weights, and AI is used only where the tests support it.
Rebalanced daily to weekly. Liquid names, trading costs modelled up front, every holding traceable to the evidence behind it.
How the factory works
Start from published evidence and a reason the effect should persist.
Stops here ifthere is no economic rationale
Success and kill criteria are written down before the first test runs.
Stops here ifcriteria can't be fixed in advance
19 years and three bear markets, net of costs, with fundamentals dated by filing.
Stops here ifit fails its pre-registered gates
A separate review hunts for look-ahead, survivorship and timing artefacts.
Stops here ifa leak is found; the result is retracted
Live signals, scored against acceptance bands fixed before launch.
Stops here ifresults leave the band
Staged go-live with kill-switches, reconciliation and a full audit trail.
Stops ifthe hard drawdown limit is hit
Where AI fits
AI agents build and run the simulations, write up the results and review each other's work for leaks. People approve what ships.
Language models turn filings, transcripts, news and analyst videos into structured, time-stamped inputs. Live today in our analyst-intelligence models.
An AI component joins a live decision only after it is tested against a simpler rule. In Atlas the rule won, so Atlas runs on logic you can explain line by line.
Every input is time-stamped. Filings count from their release date, prices from their close.
AI output feeds scores and rules. It never sends an order on its own.
Model versions, prompts and inputs are fingerprinted, so any past decision can be reproduced.
People approve every model, every parameter change and every go-live.
Featured model · two sleeves
Each sleeve covers the other's blind spot. The core's exposure control is built for slow bear markets; the crisis sleeve is built for fast crashes.
~2,500 names, screened daily
Quality screens, ranked on capital discipline and earnings momentum
Market breadth and a drawdown ladder set how much is invested
Buys a basket of quality names after market stress peaks, with stops and holding limits on every episode
Limits the crisis sleeve's share of capital
Long-only, unlevered, one account
Atlas · simulated results, 2007–2025
| Calendar year | Atlas | S&P 500 TR |
|---|---|---|
| 2008 · financial crisis | −14.8% | −36.8% |
| 2020 · pandemic crash | +38.2% | +18.4% |
| 2022 · rate shock | −14.4% | −18.2% |
Simulated past performance. Past performance is not a reliable indicator of future results. USD, 4 Jan 2007 – 31 Dec 2025 (4,779 trading days), net of modelled trading costs and before any management or platform fees, dividends reinvested. Sharpe ratios use total returns without deducting cash, for every series. The conservative case removes the extra return from a rebalancing band that we do not underwrite. Benchmark: SPY total return. The universe is drawn from current listings, so results carry survivorship bias.
Atlas · evidence quality
Yes. It follows a written specification and published research; live pass/fail bands were set before paper trading.
All of them are logged: 58 documented experiments on Atlas, 28 of them rejected or deferred.
300 nearby settings, walk-forward tested, all kept an out-of-sample Sharpe above 0.9. The final settings used all 19 years.
Spread, market impact and a cap of 10% of daily volume are modelled. About 1.9% a year of cost drag is already deducted.
A one-day delay on every signal weakens the results, but they stay positive.
Some. The universe comes from current listings. We disclose it and can re-run on delisting-inclusive data.
Single-digit to low-$20s millions at current settings, enough for a pilot. Larger-cap variants can raise it.
Always. Paper and live results are reported on their own, against pre-set bands, never spliced onto the backtest.
Risk & governance
The band is the 5th–95th percentile of Atlas's simulated six-month returns. Falling below it raises a formal underperformance flag. Below the band with a 12-month Sharpe under −0.56, we declare the model is not working.
Offering Atlas at WELREX
Weekly target weights by API or file. WELREX keeps discretion and implements across its custodians.
Atlas beside your in-house strategies, with factsheets and client-ready reporting for every RM.
An actively managed certificate via an issuing partner: one ISIN, held at any custodian.
Built for WELREX: risk-tiered Atlas variants, risk overlays for your thematic portfolios, other markets.
You keep discretion, suitability and the client relationship. Terms are open: licence, revenue share or performance-linked.
Why VantageVest
contact@vantagevestai.com
For investment professionals only; not for retail distribution. This presentation is for discussion purposes. It is not investment advice, a financial promotion to retail clients, or an offer to buy or sell any security. All performance figures are simulated past performance from a backtest (January 2007 – December 2025), net of modelled trading costs, before fees and including dividends; they do not reflect actual trading, and the research universe carries survivorship bias. Simulated past performance is not a reliable indicator of future results. The value of investments can fall as well as rise.