Black-Scholes: Global Truth, Local Fallacy

Every options textbook published in the last 30 years opens the same way: "Black-Scholes assumes constant volatility, which is clearly wrong." Then it spends 400 pages trying to fix it. Heston adds stochastic volatility. Merton adds jumps. Dupire builds local volatility. SABR models the smile. Rough volatility models the microstructure.

They all asked the same question: "BSM is broken. How do we replace it?"

Nobody asked the better question: "If BSM is so wrong, why does the entire global options market, across every asset class, every regime, and every decade, still quote in BSM implied volatility?"

The answer changes how you think about every pricing model ever built.

The Paradox Nobody Talks About

Black-Scholes-Merton is wrong. This isn't controversial. Volatility isn't constant. Returns aren't lognormal. Markets jump. The model's assumptions are violated every second of every trading day, in every market on Earth.

And yet:

• Every option on every exchange is quoted in BSM implied volatility.
• Every model (Heston, SABR, local vol, rough vol) calibrates to the BSM IV surface.
• Every risk system reports Greeks computed from BSM.
• Every trader communicates in BSM coordinates: "The 25-delta put is at 22.3."

This isn't inertia. It isn't convention for convention's sake. Something structural is hiding in plain sight, and once you see it, you can't unsee it.

BSM Is Not a Failed Model. It's the Origin of Model Space.

Step back and consider what BSM actually represents. Yes, it assumes things: continuous trading, a single source of randomness, constant rates. Those are the mechanical assumptions you need to make hedging work. But notice what it does not assume. No jumps. No stochastic vol. No skew. No regime intelligence. No view on how markets actually behave. It is the unique pricing solution you get when you take the bare mechanics of hedging and refuse to add a single belief on top of them.

In information-theoretic terms, BSM is the maximum-entropy solution: the risk-neutral measure that makes the fewest assumptions beyond the structural minimum required for replication. It is the price you get when you refuse to believe anything you don't have to.

This makes BSM something far more important than a pricing model. It makes it the center of the entire space of possible models. The geometric origin. The fixed point that every other model is measured against, not because it's accurate, but because it is the only model that carries no opinion.

Global Truth, Local Fallacy

Here's the core insight: BSM is globally true and locally false, simultaneously.

It's globally true in the sense that it is the ensemble mean across all possible market dynamics. If you averaged over every conceivable regime (high vol, low vol, jumps, no jumps, put skew, call skew, rough paths, smooth paths), you'd converge on BSM. It is the center of gravity of all theoretical possibilities.

It's locally false because no real market, at any specific moment, actually behaves like BSM predicts. There are always jumps, or stochastic vol, or skew, or some deviation from the assumptions. BSM is never the correct description of what's happening right now.

This is the duality that resolves the paradox. BSM works as a coordinate system precisely because it fails as a predictive model. You don't need the center of a space to describe any specific point in it. You need it to measure the distance to every point from a common reference.

The 50-Year Misdiagnosis

The entire post-BSM research program, from Heston (1993) to rough volatility (2018), was built on a misdiagnosis.

They observed that BSM doesn't match market prices. They concluded it was broken. They set out to fix it by building "better" models.

Heston's premise was straightforward: volatility isn't constant, option prices prove it, so relax that assumption.

Dupire took the same path from a different angle: the constant volatility assumption is clearly inadequate, so derive a deterministic volatility function that fits the smile.

Merton applied the same instinct to the paths themselves: Black-Scholes excludes jumps, so extend the model to include jump processes.

They all asked: "BSM is wrong. How do we build something better?"

The question they never asked: "BSM is always wrong. Why does the market still quote in it?"

That's not a bug. That's the signal.

They treated BSM as a patient. But BSM was never sick. It was never meant to describe any local market condition. It's the global reference frame: the model you get when you have zero information beyond no-arbitrage.

You don't "fix" the mean by replacing it with a data point. You center on the mean, then measure deviations from it.

Every "Better" Model Proves the Point

Here's the irony that the model builders never noticed: every model they built to "replace" BSM actually reinforced its centrality.

• Every model is calibrated to BSM implied volatility. Not to raw prices, but to BSM-transformed prices.
• Every model is measured by its deviation from BSM. We talk about smile, skew, and term structure, all defined relative to the BSM flat-vol baseline.
• Every model reverts to BSM when its extra parameters go to zero. Set vol-of-vol to zero in Heston, jump intensity to zero in Merton, and you get BSM back.
• Every model's P&L is reported in BSM Greeks.

They didn't replace BSM. They built a coordinate system around it. Every "improvement" is a more detailed map of how far today's market has wandered from the center.

Even Jim Gatheral, arguably the most sophisticated thinker in the post-BSM era and author of The Volatility Surface, acknowledged that BSM IV is a "convenient coordinate system." But he stopped at pragmatic utility. He spent his career searching for the One True Model that would fit the entire volatility surface across all strikes and maturities. He parameterized the surface in BSM coordinates while trying to eliminate the need for BSM. He held the answer in his hands and set it down.

There Are No "Local" Models. Only Particular Cases.

The word "local" implies continuity, a smooth neighborhood of market conditions where a model works. But markets don't have smooth neighborhoods. They have discrete, disjoint regimes that shift without warning.

• A stochastic volatility regime (VIX goes from 12 to 40 in 48 hours).
• A jump regime (earnings announcement, FOMC decision, flash crash).
• A skew regime (equity put skew vs. rate call skew vs. crypto symmetric smile).
• A liquidity regime (market open vs. close, single stock vs. index).

Each regime is essentially a different world. No model trained in one survives the crossing into the next without re-calibration. Heston fits today, breaks tomorrow. SABR works for rates, fails for equities. Rough vol fits SPX microstructure, fails VIX dynamics.

What people call "local models" are really particular-case models: temporarily useful tools calibrated to one transient configuration of asset class, volatility regime, skew shape, and market microstructure. They are valid for that case and only that case.

Even the most persistent "stylized fact" in options, the OTM put skew, is not a law of nature. It's a particular case. Japanese rates in the 1990s had call skew. Crypto in 2021 was symmetric or call-heavy. VIX futures invert their term structure. Skew is not a model input. Skew is a market output: contingent, transient, and theoretically reversible.

What This Actually Means for Pricing

This isn't philosophy for its own sake. It has teeth.

BSM is the only model that doesn't care what the current regime is. It survives all markets because it fits none of them specifically. It has the lowest regression error across all theoretical possibilities, precisely because it doesn't overfit to any particular one.

Every other model (Heston, SABR, local vol, jump diffusion, variance gamma, rough vol) is a particular-case tool. Useful in its regime, obsolete in the next. And that's fine. All models are useful, but only for particular cases. No product "requires" a specific model. A 1-day ATM SPX option during a flash crash needs more sophisticated modeling than a 10-year barrier in calm rates. It's the case that determines the model, not the product.

Quote, hedge, and manage risk in BSM. Calibrate particular-case models to explain today's deviation from BSM. Switch or discard them the moment the case changes.

This is what the best trading desks already do. They just never say it out loud. The exotic desk uses local vol for today's smile. The vol arb desk uses Heston for vol-of-vol dynamics. Risk aggregates everything back into BSM Greeks. The trader quotes everything in BSM IV. They're particular-case switching without the philosophy.

High-Resolution Black-Scholes

There is one more implication, and it reframes everything.

The monster models at firms like Citadel Securities and Jane Street, the production engines with thousands of calibrated parameters that reprice entire options surfaces every few minutes, are not "beyond" Black-Scholes. They are Black-Scholes.

They solve the exact same problem BSM solves: "What is the unique no-arbitrage risk-neutral measure that makes the fewest assumptions beyond what we actually know?" The only difference is the size of the information set. BSM knows three things (spot, rate, dividend). The production engine knows 20,000 things (every option price on the surface, plus historical calibrations, plus microstructure signals).

If the entire options surface were perfectly lognormal with flat volatility, if the market truly had zero information beyond the BSM baseline, these engines would automatically converge on pure BSM. All the extra parameters (jump intensities, roughness, vol-of-vol) would go to zero. Not because someone coded BSM in, but because BSM is the unique maximum-entropy solution consistent with an uninformed market.

The 50,000-parameter production pricer is what Black, Scholes, and Merton would have built if they'd had access to today's options surfaces and GPU clusters. It's not post-BSM. It's BSM at higher resolution.

Possibility Space, Not Time

One final clarification, because the wrong analogy is seductive.

It's tempting to cast BSM as a "Big Bang" initial state that evolves over time into complex market structures. That analogy breaks down because BSM doesn't describe a time evolution. It describes a static geometry.

There is no "financial Big Bang" followed by an irreversible increase in complexity. There is only:

1. The space of all conceivable risk-neutral measures consistent with no-arbitrage.
2. Exactly one point in that space that encodes zero additional beliefs: BSM.
3. The actual market, which selects a different point in that space every day based on what participants collectively believe.

BSM is the pristine, perfectly ordered center of possibility space. One parameter, perfect symmetry, zero skew. The calibrated production model is a contorted point far from that center, twisted by thousands of specific market beliefs into a shape that matches today's surface. The distance between them is the total information the market has injected beyond the minimum.

Every tick of implied volatility away from the BSM flat-vol baseline is the market screaming: "We know something you're pretending not to know."

The Framework

Reduced to its essence:

• BSM is the universal model. Never true. Always relevant. The least-regression-error model across all theoretical possibilities. The fixed point of model space. The coordinate origin. The market's lingua franca.
• Everything else is a particular-case model. Temporarily useful. Each one valid for one specific configuration of asset class, regime, skew, and microstructure. Tools, not truths. Calibrated to today's deviation from BSM. Discarded when the case changes.

BSM is the only model that survives all markets because it fits none. Every other model fits one market and dies with it.