Quant finance has a metaphysics problem. We have a different one.

The premise

Every market model you’ll ever write makes a metaphysical choice before the first line of code. You can tell what it assumes by looking at what its primitives are. If the primitives are assets, positions, balance sheets, and flows — it’s assuming that the world is made of things, and that change is what happens to things. If the primitives are transitions, episodes, and evolving dynamics — it’s assuming the world is made of processes, and that things are temporary stabilizations inside flows.

The first view is material monism. Quant finance runs on it by default. Factor models, equilibrium-ish pictures, positioning models, microstructure-as-plumbing — all of them start from entities and ask what those entities do. This is measurable, tractable, and strong at capturing constraints and mechanics. It is terrible at capturing the things that make markets hard: reflexivity, narrative formation, regime change, liquidity spirals, crowding, path dependence.

The second view is process monism. It says things are temporary stabilizations inside flows; entities are slow-moving eddies, not primitives; persistence is repeated process, not static substance. It fits markets better than the material-monist view on the dimensions that matter. We want to adopt it as the design prior under our models, explicitly, and see what that buys us.

This is not a philosophy post. It’s a commitment about what our code takes as its primitives.

The four positions

Cross two axes: what is real (things or activities?) and where the primitive lives (inner/interpretive or outer/structural?). You get a 2×2 that lets you place any market model on a single diagram.

Most quant work sits in the top-left. Most macro-narrative work sits in the bottom-left. Most crisis-era writing is an uneasy blend of the right column, without ever saying so.

We think the strongest anchor for market research is the bottom-right: reflexive-process monism. Everything reduces to evolving feedback loops between interpretation and structure. Structure constrains what narratives can survive. Narratives change what structures get built. The two co-produce temporary realities that look stable for a while and then aren’t.

Why this fits markets

Markets are temporal all the way down. There is no market “state” you can photograph that isn’t already changing when you take the picture. Factors, narratives, regimes, and even firms are persistent patterns rather than primitives — every one of them dissolves on a long enough horizon. The material-monist view recovers this by bolting dynamics onto a static picture (“regime-switching GARCH”) after the fact. The process-monist view starts with dynamics and treats stability as the interesting thing to explain.

Three places where the commitment matters in practice:

• Regime detection is transitions-first, not state-first. If you anchor on process, you stop asking “what state is the market in?” and start asking “what transition is it going through, and how far?” The former is a classification problem. The latter is a trajectory problem. The credit cycle we isolated is intelligible only as a trajectory — its phase matters more than any single snapshot.
• Reflexivity is native, not bolted on. Soros-style feedback loops between price and belief are structurally invisible to an entity-first model (the entity exists before its observers) and structurally native to a reflexive-process model (the entity and its observers are co-defined by their interaction). If you want reflexivity to show up in the math, you have to put it in the primitives.
• Metastability replaces equilibrium. A process model expects long stretches where nothing moves much, punctuated by fast reorganizations. That matches market reality better than any equilibrium-plus-shocks decomposition, and it makes transition risk the first-class object instead of the exception.

What it buys you, and what it costs

What it buys:

• Regime changes, contagion, liquidity spirals, and narrative formation all become first-class phenomena, not anomalies to be patched in.
• Path dependence stops being a nuisance and becomes a feature you can exploit.
• You get a natural home for the kind of state-machine regime taxonomy we’re also considering — it’s hard to justify a transitions-first state machine on top of an entity-first ontology.

What it costs:

• It is easy to make beautiful and slippery. Process philosophy can turn into a cathedral of elegant verbs with no observables attached. If the framework can’t be tied down to transition matrices, persistence measures, entropy, coupling, and dispersion — measurable things that can fail a test — it’s decoration. This is the failure mode we’re most worried about.
• It imposes a discipline most libraries don’t support. Most statistical tooling assumes i.i.d. draws from a fixed distribution, which is exactly the assumption a process-monist model refuses to make. Time-varying everything gets tedious fast, and the standard errors are harder.
• It invites smuggling in untestable claims. “Markets are reflexive processes” sounds profound and commits you to nothing. We want every use of the frame to cash out in a thing we could measure differently and be wrong about.

The practical stack we’d run under this

• Ontology: process monism, leaning reflexive. Primitives are transitions, episodes, and evolving couplings — not states, positions, or factors.
• Psychology layer: cognitive-architecture metaphors for naming the kinds of dynamic organization markets fall into. DMN, SN, CEN, limbic as mnemonic labels only.
• Measurement layer: transition probabilities, persistence statistics, entropy of state histories, cross-asset coupling changes, attention concentration, dispersion trajectories, crowding stress.
• Use case: regime detection, transition forecasting, strategy selection conditioned on regime. The credit cycle fits in here as one long-period transition variable among several faster ones.

How we’d know it was working

This scenario is different from the other two in that there’s no single pre-registered test that makes it true or false. The commitment pays off over time, across models, if the things we build under it consistently do better at the tasks that material-monist models struggle with: anticipating transitions, estimating regime-conditional distributions, and not blowing up during phase changes.

If, a year from now, our best models are still entity-first with regime-switching patches, we were wrong to elevate the ontology. If they’re transitions-first and the patches are on the entities instead, the commitment did its job.

This is the scenario least likely to produce a “result.” It’s the one most likely to shape the shape of everything else we publish.