Direction is the question everyone asks. Volume and volatility are where the market actually answers first.
Our model doesn’t try to guess up-or-down out of thin air. It reads how much is trading and how hard prices are swinging — two signals that move ahead of price — and uses that to build a view on where the market is likely to lean next.
Chasing price direction head-on is fighting the hardest part of the market
On a heavily traded stock, tomorrow’s direction is close to the most efficient, most competed-over number in finance. Every large player, every algorithm, every headline gets priced in almost instantly. Treated in isolation, next-bar direction behaves close to a coin flip — which is exactly why so many models that try to predict it head-on quietly fail out of sample.
That’s not a flaw in the modeling. It’s a property of the market itself. The fix isn’t a smarter way to guess direction — it’s asking a different, more answerable question first.
What’s hard
Raw next-bar direction on a liquid name
Near-random, heavily arbitraged
What’s tractable
Volatility clustering, volume conviction
Persistent, structured, measurable
Volume and volatility carry the structure that price alone hides
Volatility clusters. Calm markets tend to stay calm; turbulent markets tend to stay turbulent. That persistence is one of the most reliable, well-documented regularities in all of finance — far more stable than the sign of the next return.
Volume moves first. A shift in conviction — accumulation, distribution, capitulation — usually shows up in how much is trading before it shows up in where price ends up. Volume is the market thinking out loud before it commits.
Put together, a read on “how much is trading and how hard” turns out to carry real information about which way the tape is likely to lean next — even though the direction signal alone barely does.
Volatility
Regime persistence
Calm → calm, stress → stress
Volume
Conviction, ahead of price
Early signal of a shift underway
Why a quantum-inspired model, specifically
Volume and volatility don’t move in clean, straight lines — they interact, lag each other, and shift character as the market moves between calm and stressed regimes. Capturing that kind of tangled, non-linear behaviour is exactly where classical statistical models start to strain.
We built our reservoir on principles borrowed from quantum physics — the same ideas used to model complex physical systems that carry memory and shift between distinct states. It lets the model hold a much richer, higher-dimensional picture of “what kind of market this is right now” than a conventional model working with the same handful of inputs, and it’s naturally sensitive to the moment a market flips from one regime into another — which is precisely when direction becomes easier to read.
Think of it less as a black box that predicts prices, and more as a very sensitive instrument for reading market state — and state is what direction is eventually made of.
Classical models
Linear or shallow non-linear
Struggle across regime shifts
Our approach
High-dimensional, physics-based state
Naturally regime-aware
Tested where it’s easy to get caught lying: out of sample
Every result is produced walk-forward — the model only ever sees the past, is tested on data it has never touched, and is re-trained forward in time as new data arrives. No look-ahead, no cherry-picked window. We run this across multiple, genuinely different market regimes rather than a single friendly year, because a model that only works in one type of market is a coincidence, not an edge.
Volatility targeting wins
In steady uptrends, scaling exposure down when volatility rises and up when it’s calm captures the drift with a smoother ride.
Direction + volatility together win
When there’s no clean trend, combining a directional lean with volatility-based sizing is what separates signal from chop.
Adaptive combination wins
The same combined approach that works sideways carries through corrections, because it isn’t betting on a trend that isn’t there.
Figures above summarize internal research across multiple tickers and multi-year windows and are shown for illustration of methodology, not as a forecast of future returns.
What we don’t publish, and why
The exact physical model, its internal parameters, and the way it turns raw volume and volatility into a market read-out are the result of years of dedicated research. That architecture is proprietary and we don’t publish it — the same way a trading desk doesn’t publish its execution logic or a fund doesn’t publish its factor weights.
What we do share, openly, is what the model does, why the approach makes sense, and how it performs under honest, out-of-sample testing. That’s what should earn your trust — not a diagram of the internals.
Questions worth asking
Does the model predict exact prices?
No. It produces a probabilistic read on market state — direction lean, expected volatility — which is then translated into position sizing and exposure through a trading strategy, not a price target.
Why not just predict direction directly, like most models try to?
Because direction alone is the noisiest, most competed-over signal in the market. Volume and volatility carry more real structure, and a good read on market state is what makes a directional lean trustworthy in the first place.
Is this financial advice?
No. This page describes a research methodology and forecasting approach. It isn’t investment advice, and past or backtested performance is never a guarantee of future results.