Are We Modeling the World, or Just Ourselves?

A Quiet Critique of Financial Predictive Culture

Jul 08, 2025

I’m all for introspection, but it shouldn’t come at the cost of seeing the outside world clearly. Image generated with Leonardo AI

Most financial models today are predictive. They forecast quarterly earnings, project cash flows, simulate downside scenarios. They extrapolate from the past to tell us something about the future.

But here’s a quiet, unsettling question:
Are we actually modeling the world—or are we just modeling our own assumptions?

This question matters more than ever in a world shaped by sustainability forces. Climate change, social transitions, regulatory shocks—they’re not linear, they’re not smooth, and they rarely behave like the historical data would suggest.

And yet, many models continue to treat them as if they do.

Let’s unpack why this happens—and what we can do about it.

The Comfort of Predictive Thinking

Prediction is seductive. It gives us a sense of control. If you’ve worked in finance, you know the feeling: Excel spits out a neat figure, a valuation multiple, a 5-year CAGR. You feel anchored.

But here’s the catch: the more precisely we predict using the wrong structure, the more confident we become in the wrong conclusion.

This is especially true when sustainability enters the picture. Models trained on past data—data that mostly reflects a world of cheap energy, weak regulation, and slow environmental feedback—may no longer apply.

If your model tells you a coal company is undervalued because it has a low P/E ratio, is that the world speaking?
Or is it you—your spreadsheet, your priors—speaking through the model?

A Mirror, Not a Window

Most financial models are mirrors, not windows.

They reflect back our chosen assumptions:

Which variables we include
What timeframe we consider
Which relationships we assume to be stable
What we treat as “noise” and what we treat as “signal”

Take a standard Monte Carlo simulation for portfolio risk. It assumes distributions, volatilities, correlations—all drawn from the past. But in a climate-altered world, a social tipping point, or a policy rupture, those assumptions can break without warning.

In this case, you’re not modeling the future. You’re modeling your comfort zone.

What Modeling Should Be: Exploration, Not Just Estimation

Here’s a radical thought: maybe the purpose of modeling isn’t just to estimate.
Maybe it’s also to explore.

Explore what could happen if the future doesn’t look like the past
Explore how your conclusions change when assumptions shift
Explore alternative worldviews—especially the ones you didn’t think were likely

This is what scenario thinking, systems modeling, and causal inference can offer. They don’t just say “here’s the number.” They say:

“Here’s how it might behave—depending on how the world evolves.”

That’s a different kind of intelligence. Less arrogant. More useful.

A Case in Point: Carbon Risk and Stock Prices

Let’s take a real example. Many analysts today want to estimate how carbon intensity affects stock prices or volatility. That’s a fair question.

But depending on how you model it, you may get wildly different results.

A regression using historical data might show no significant effect.
A causal model might show that carbon only impacts prices under certain policy scenarios.
A scenario simulation might show that investor sentiment around climate risk could suddenly flip in a single year, repricing entire sectors.

Which one is right?

That’s not even the right question. The better question is:
Which of these views helps you think more clearly about your exposure to a non-linear, policy-sensitive future?

How Models Encode Power, Not Just Numbers

There’s a deeper reason to be cautious. Models don’t just represent financial flows.
They encode narratives, and sometimes, power.

A model that ignores climate damage externalities supports the status quo.
A model that assumes labor costs are fixed can reinforce exploitative dynamics.
A model that treats sustainability as immaterial quietly tells the market to ignore it.

Every cell in a model is a political choice. Every “what if” or “let’s ignore that” decision shapes what decisions seem reasonable later on.

We don’t always like to admit this in finance. But if models shape capital allocation—and capital shapes the world—then model design is a moral act.

So What Can We Do Differently?

This isn’t about throwing out financial modeling. It’s about modeling with more humility—and more awareness.

Here are five quiet shifts that can help you move from mirror to window:

1. Make Assumptions Explicit

Instead of hiding assumptions in the code or behind defaults, surface them. Say:

“This model assumes a stable regulatory environment.”
“This assumes climate risk is priced in slowly and linearly.”
“This ignores supply chain fragility beyond Tier 1 suppliers.”

The goal isn’t perfection. It’s transparency.

2. Use Causal Diagrams to Think Structurally

Before you build, draw. Ask:

What causes what?
Where are the feedback loops?
Which mechanisms matter under stress?

A simple DAG can save you from building a model that looks great but rests on sand.

3. Embrace Counterfactuals, Not Just Forecasts

Ask:

“What would this portfolio look like if emissions were taxed at €100/ton tomorrow?”
“What would happen if this company lost its social license to operate?”
“What if this assumption no longer holds?”

These “what-if” questions aren’t distractions. They’re probes into robustness.

4. Model Behavior, Not Just Metrics

Especially in sustainability, behavior matters.

How do regulators respond to public pressure?
How do consumers react to scandals?
How fast do capital markets move when the narrative changes?

Models that ignore social and institutional behavior will miss key drivers of volatility and value.

5. Design for Breakage

Build models that break well—that make it obvious when assumptions no longer apply, or when a scenario falls outside the training distribution.

Sometimes the most honest model isn’t the one that runs smoothly, but the one that flashes warning signs when the world starts to shift.

The Bottom Line: We Model What We Value

At the end of the day, models are cultural tools. They reflect what we care about, what we expect, what we think matters.

So when a model doesn’t include sustainability forces—when it excludes climate risk, or reduces biodiversity to a coefficient, or treats social impacts as intangible—it’s not just a modeling error.

It’s a value judgment. One that says: this isn’t worth including.

But what if it is?

What if the better model isn’t the one with the cleanest R², but the one that helps us navigate complexity with clarity, humility, and imagination?