SkySail Strategies Outperforms Wall Street's 30-Year Risk Standard and Reduces Market Risk Forecast Error by 20 to 40 Percent With Proprietary AI Inference Model
Every morning, before the opening bell, trading desks around the world answer the same question: how much could the market move today?
The answer is one of the most consequential numbers in finance. It decides how large a position a fund is allowed to hold, where a bank sets its limits, how much a portfolio can carry through the night, where a trader places a stop. Trillions of dollars sit behind it. And for more than thirty years, it has come from the same family of models, built in the early 1990s and largely unchanged since.
SkySail Strategies, a quantitative investment firm focused on applied quantitative research and investments, decided that number could be predicted far better. Its Rubix VM system, built around a custom AI inference model the firm calls Rubix RPT, predicts how far a market will move with 20 to 40 percent less error than the industry's standard models, at every horizon from one hour to a full trading day.
"The industry has measured risk the same way since 1993 because nobody has been able to reliably beat it. We have."— Brian Devens, Managing Director
A number with a long history
The models Wall Street uses to forecast market risk descend from a single idea. In 1982 the economist Robert Engle showed that market turbulence clusters: a volatile day is more likely to be followed by another volatile day than by a calm one. The insight was so useful it earned him the Nobel Prize in Economics two decades later. Researchers refined it through the 1980s, and by 1993 the version most risk desks still run today had taken its final shape. The following year J.P. Morgan published its own variant and gave it away free, and the approach became the standard for measuring risk across the industry.
It has held that position through the dot-com crash, the 2008 financial crisis, the pandemic and every shock since. Regulators built their rules around it. Risk departments built their software around it. A generation of traders came up never knowing another way.
It did not survive that long by accident. The model has held its place for three decades because it is very hard to beat. Academics have proposed alternatives by the hundred. Banks and quantitative funds with enormous research budgets have tried to improve on it. Most challengers turned out to be no better than the original, and the few that were better were better by a hair. In a field where an edge of a percent or two is worth fortunes, thirty years at the top is an extraordinary record. That is why the industry still runs on it, and why any new approach is not being measured against something antiquated but against the standard that has outlasted everything thrown at it.
The approach works the way its origin suggests. It looks at how much the market moved yesterday, last week and last month, and projects that turbulence forward. On most days it does its job well. Its one structural weakness is that it can only describe the market that already happened, and that weakness surfaces on exactly the days when getting risk right is worth the most.
"These models are very good at telling you what yesterday looked like. They were never built to tell you what today is about to become."— Brian Devens, Managing Director
Rubix was built to predict the market as it is, hour by hour, from the market in front of it rather than the one behind it. It is built for every trading day, the quiet ones and the violent ones alike, and it holds its accuracy when the market changes character, which is where the old models lose theirs.
Where the number is actually used
The bank risk desk is where the industry learned to think about risk, and the 1993 lineage still runs inside that machinery. But the people who live and die by the next few hours are somewhere else: on trading floors, at proprietary trading and futures firms, on options desks, inside systematic funds. For them, how far price can travel before the close is not a compliance figure. It is the difference between a good year and a bad one.
That world has always moved faster on risk than the banks. It adopted a newer generation of realized-volatility models, the kind hedge funds and quantitative desks now treat as the state of the art, because accuracy at short horizons is money. And for the most common job of all, deciding how far the market can move in the next hour or two, most of that world still leans on a trader's rule of thumb, average true range, that dates to 1978.
Rubix beats all of them. The 1993 bank standard that has resisted thirty years of challengers, the newer realized-volatility models that hedge funds treat as the state of the art, the trader's rule of thumb, at every horizon from one hour to a full day. In practice that means positions sized to the day the market is actually going to have, stops that sit where they belong, and capital that is neither wasted defending against moves that never come nor exposed to the ones that do. Over time, that precision is what steadier risk-adjusted returns are made of.
"Everything in investing is built on one number: how much can this move against me. Position size, where the stop goes, how much you can hold overnight, all of it sits on top of that number. Get it wrong and nothing above it works. We decided to get it right."— Brian Devens, Managing Director
A different kind of AI story
Most conversations about artificial intelligence in finance are about predicting price: which stock will rise, which way the index will turn. SkySail took a different view of where the technology belongs.
Devens, a systems engineer who specialized in probabilistic and stochastic algorithms, built the firm around a conviction that the most valuable number in markets is not where price is going but how far it can go. Rubix RPT is not a general-purpose AI pointed at a price chart. It is an inference model SkySail built from the ground up for that single job: predicting how far a market can travel over the coming hours. Inside the Rubix VM system, it does that job 20 to 40 percent more accurately than the models the industry has relied on since 1993.
The firm is deliberately quiet about how. What it will say is what the model can do: give it any market with a price history and it predicts how far that market can move. The same engine that predicts risk on the Nasdaq predicts it on crude oil, on gold, on a currency pair, with nothing changed but the market in front of it.
Inside SkySail, Rubix VM is a foundational model: it does not trade on its own, and it now operates inside NQ-GX7. How SkySail Algorithms Earn Capital explains how foundations like it are built, proven and put to work.
"Every market has a distance it can travel in the next few hours, and almost nobody can predict it well. Rubix can, and it does it in every kind of market, calm or violent, trending or stuck, without losing its accuracy when the character of the market changes."— Brian Devens, Managing Director
| Move predicted | Limit failed | ||||||
|---|---|---|---|---|---|---|---|
| Year | Rubix | Realized-vol | GARCH | Rubix | Realized-vol | GARCH | |
| 2023 · a quiet year | 98% | 114% | 127% | 18% | 10% | 5% | |
| 2024 | 95% | 98% | 95% | 22% | 25% | 24% | |
| 2025 · through May | 94% | 78% | 62% | 22% | 34% | 55% | |
That reach is turning into demand. As SkySail extends Rubix into new markets and new applications, the interest has come from the part of the industry that manages serious capital and has the most to gain from a better risk number.
"It's an exciting time for us. Rubix has been remarkably effective, and as we deploy it across more markets and more applications we're seeing real traction and interest from family offices and funds. They understand better than anyone what a more accurate risk number is worth."— Brian Devens, Managing Director
The number, answered
Every morning the question gets asked. For thirty years the industry answered it with a model built in 1993, not out of habit but because nothing anyone built could beat it. The biggest banks in the world built their risk on it. The sharpest quantitative funds tried to improve on it and mostly failed.
SkySail built an AI that answers the question better than the standard that outlasted all of them, and put it to work on the number that everything else in investing depends on.
"That model held for thirty years because it earned it. Beating it was never going to be easy, and that's exactly why it was worth doing. It was the best available. It isn't anymore."— Brian Devens, Managing Director