Slope Not Intercept
What most people get wrong about trends
Everyone knows trends matter. That’s not an insight. That’s a poster in a high school guidance counselor’s office.
The actual insight is weirder, deeper, and more useful than “pay attention to trajectory.” It starts with a question most people never ask: why do slopes persist?
Once you understand the answer to that question, you start seeing the world differently. You stop being surprised by outcomes that surprise everyone else. And you develop what I think is the single most underleveraged mental model in business, investing, and life.
Slopes are character. Intercepts are circumstance.
Here is the first thing people get wrong. They treat the current state of a system (its intercept) as a fact about reality, and its trajectory (its slope) as a prediction. One is real. The other is speculative.
This is exactly backwards.
The intercept is the noisiest variable. It’s the product of a thousand random inputs. Timing. Luck. Starting conditions. A company’s revenue in any given quarter is shaped by one-time deals, seasonal variation, macro conditions. A person’s net worth at 35 is shaped by which city they happened to live in, whether they graduated into a recession, what their parents could afford. The intercept is mostly circumstance.
The slope is structural. It’s generated by the deep properties of a system. Culture. Physics. Network topology. Feedback loops. Institutional design. These things change slowly. Which means the slope is actually the more durable signal.
Think about that. The thing most people treat as speculative (where it’s going) is generated by more stable forces than the thing they treat as certain (where it is).
A company with a strong slope and a weak intercept is a company with good structural properties that hasn’t had time to express them yet. A company with a strong intercept and a weak slope is a company burning through inherited advantage. Circumstance is being corrected by character.
This is why slope persists. Not because of some mystical momentum. Because the structural generators of slope, the culture, the compounding knowledge, the network effects, the cost curves driven by physics, change on a different timescale than the noise that determines intercept.
The second derivative is where the money is
Most people who do think about slope still think about it wrong. They think linearly. “It grew 20% last year, so it’ll probably grow 20% next year.”
The more important question is whether the slope itself is changing. The second derivative. Is the rate of change accelerating or decelerating?
This is where almost all asymmetric outcomes hide. Every S-curve in history has a period where the second derivative is positive (the slope is steepening) followed by a period where the second derivative is negative (the slope is flattening). The entire game of timing is figuring out where you are on that curve.
But here’s what makes this hard: humans are perceptually wired for linear extrapolation. Kahneman documented this. We anchor to recent rates and project them forward. We are structurally incapable of intuiting exponential curves or inflection points without doing the math.
In 2024, global AI investment crossed $200 billion. Up from around $40 billion five years prior. Most analysts projected continued growth. But the real question was never “will it grow?” It was “is the second derivative positive or negative?” Is each dollar of investment producing more capability than the last (positive second derivative) or less (negative second derivative)? That distinction is the difference between a $10 trillion industry and a $1 trillion one.
Solar energy is the canonical example. For decades, the cost curve had both a persistent negative slope (falling costs) and a positive second derivative (costs falling faster each decade as manufacturing scaled). Around 2020, the second derivative began to flatten. Costs were still falling, but more slowly. The slope was still good, but the slope of the slope had changed. This distinction matters enormously for capital allocation, and almost nobody in public discourse made it.
Slope stacking: where civilizations turn
Here’s the idea I think is genuinely underexplored.
Individual slopes are interesting. But the interaction effects between slopes are where the most consequential outcomes in history come from. I call this slope stacking.
When the cost curve of one technology intersects with the adoption curve of a dependent technology, and both intersect with a demographic shift or a policy change, each slope accelerates the others. These multi-slope systems create outcomes that look like “disruption” or “revolution” in hindsight, but are actually just the predictable result of compounding slopes.
The Industrial Revolution wasn’t one slope. It was the cost curve of coal extraction intersecting with the efficiency curve of steam engines intersecting with the urbanization rate intersecting with the literacy rate. Each slope fed the others. Cheaper coal made steam engines more economic. Steam engines made coal extraction cheaper. Urbanization concentrated labor. Literacy enabled knowledge transfer that steepened every other curve. No single trend caused the Industrial Revolution. The stack did.
The same pattern is playing out right now. AI capability curves are steepening. Energy demand curves are rising. The cost curve of inference is falling. The labor productivity curve in knowledge work is inflecting. The cost of launching satellites is dropping. The amount of data generated per capita is rising. Each of these is interesting individually. But they don’t exist individually. They interact. Cheaper inference means more AI agents. More AI agents mean more energy demand. More energy demand accelerates investment in generation. More generation investment steepens the solar and nuclear cost curves. Cheaper energy makes inference cheaper. The loop tightens.
People who analyze these trends in isolation will get the next decade wrong. The right mental model isn’t “AI is getting better” or “energy demand is rising.” It’s “what happens when these slopes multiply?”
History’s biggest mispricings occur when slope stacks are underway and the consensus is still evaluating each slope independently.
Why slope perception is asymmetric (and exploitable)
There’s a well-documented asymmetry in how humans perceive positive and negative slopes. Loss aversion means we notice decline faster than growth. A company losing 10% of its customers per quarter triggers alarm. A company gaining 10% per quarter gets a polite nod.
But there’s a subtler asymmetry that’s less discussed. We’re better at perceiving slopes in things we can count than in things we can’t.
Revenue slope? You can see it in a chart. Customer count slope? Same. But what about the slope of institutional trust? The slope of employee morale? The slope of technical debt? The slope of a founder’s judgment?
These are all real slopes generated by real structural forces. They compound. They persist. And they are nearly invisible in standard reporting.
The best investors I know spend most of their time trying to measure invisible slopes. They’re not looking at the dashboard. They’re looking at the rate of change in things that don’t have dashboards. How fast is this founder learning? Is this company’s engineering culture getting better or worse each quarter? Is the regulatory environment tightening at an accelerating or decelerating rate?
These unmeasured slopes are where the market is most inefficient. Because if a slope can’t be charted, most people act as if it doesn’t exist.
The slope trap
I’d be dishonest if I didn’t flag the failure mode.
Not all slopes persist. Some slopes are temporary expressions of a one-time force. A company growing 50% because it just got featured in the press. A country’s GDP spiking because of a commodity price surge. A person’s career accelerating because of one lucky break.
The question is always: is the slope generated by structure or by event?
Structure-driven slopes persist because their generators are durable. Physics-driven cost curves (solar, compute, genomics) tend to be structural. Culture-driven performance curves tend to be structural. Network-effect-driven adoption curves tend to be structural.
Event-driven slopes revert because their generators are temporary. Stimulus-driven economic growth reverts when the stimulus ends. Hype-driven adoption reverts when attention moves. Charismatic-founder-driven culture reverts when the founder burns out.
The discipline is in distinguishing the two. And the honest answer is that it’s hard. The best heuristic I’ve found: look at the slope over multiple timescales. If it persists across different macro conditions, different leaders, different market environments, it’s probably structural. If it only appears in one context, it’s probably event-driven.
The punchline is about probability, not direction
Here’s what I actually want you to take from this.
The claim isn’t “trends always continue.” They don’t. The claim is that the base rate of trend persistence is significantly higher than most people’s intuitive estimate.
Jegadeesh and Titman’s foundational work on momentum showed that stocks with positive 12-month returns continue to outperform over the next 3-12 months at statistically significant rates. This has been replicated across equities, bonds, commodities, currencies, and real estate in virtually every major market. Not because of magic. Because the structural forces that generated the trend (earnings growth, sector rotation, institutional allocation) operate on longer timescales than the trend itself.
The same principle operates outside of markets. Countries that are growing tend to keep growing. Companies that are improving their margins tend to keep improving them. People who are getting better at their craft tend to keep getting better.
Not always. Not forever. But more often and for longer than your gut tells you.
This means that when you evaluate anything, a business, a hire, an investment, a relationship, you should be giving more weight to the slope than you currently are. Not infinite weight. But more.
Most people’s mental model allocates maybe 30% to trajectory and 70% to current state. The correct allocation is probably closer to the inverse.
The intercept tells you where something is. The slope tells you what it is.
Bet on slope.

I agree. The question is how? Very thought provoking.