First Principles, Trend Following, and AI — and Why None of Them Wins Alone

Every investment decision rests on an implicit theory of where value comes from and how prices relate to it. Three broad schools dominate practice, and increasingly they’re blending into each other in ways worth understanding clearly.

The three lenses

First principles (fundamental) investing starts from the business itself. It assumes price eventually converges to intrinsic value — the present worth of future cash flows, shaped by competitive advantage and capital allocation. The investor’s edge is supposed to come from understanding the business better than the market does, not from reading the price chart.

Trend following starts from the price. It assumes the price itself carries information — about positioning, flows, and slow information diffusion — that’s more tractable than reverse-engineering “true value.” It doesn’t ask whether something is cheap; it asks whether it’s moving, and whether that movement is likely to persist.

AI and algorithmic approaches start from data at scale. The assumption is that there are statistical patterns — across financials, text, order flow, alternative datasets — too complex or too numerous for a human to find by hand. Crucially, AI isn’t really a fourth worldview; it’s a method that can be pointed at either of the other two. A model predicting earnings surprises from filing language is a turbocharged fundamentalist. A model predicting next week’s price move from order flow is a turbocharged trend follower.

Where each one breaks

First principles fails through reductionism. Markets are reflexive — prices move partly because of what people believe, and those beliefs respond to the price itself. A DCF can feel rigorous while just relocating the guesswork from one number (a multiple) into five harder-to-pin-down ones (growth, margin, discount rate). And dismissing the market price as “noise” ignores that it’s an aggregation of enormous amounts of distributed information and capital at risk.

Trend following fails through correlation mistaken for structure. Trends persist until they don’t, and they tend to reverse sharply exactly when the crowd is most committed to them. The approach has no answer to “what is this actually worth” — so it can ride a bubble up and most of the way down with no fundamental anchor to fall back on.

AI fails in its own distinctive ways:

• Overfitting disguised as discovery — with enough features, a model will find “patterns” that are pure noise and evaporate out of sample.
• Non-stationarity — a model trained on one regime has no causal understanding of why its pattern worked, so it can’t flag when the regime has shifted underneath it.
• The interpretability gap — if you can’t explain why the model holds a position, you can’t size it sensibly or distinguish “working as designed” from “broken” during a drawdown.
• Backtest illusions — look-ahead bias and subtle data leakage routinely make historical performance look far better than anything achievable live.
• Crowding — funds trained on similar datasets converge on similar signals, so quant unwinds can be sharper than fundamentals-driven ones, precisely because the strategies looked uncorrelated until they weren’t.

The synthesis

None of these lenses is sufficient alone, and the strongest practical view treats them as complementary rather than competing:

• First principles tells you what something is worth and why — giving conviction to hold through volatility.
• Trend and price-based signals tell you about timing and crowd behavior — useful for entries, exits, and knowing when not to fight a flow even if you believe it’s “wrong.”
• AI and quantitative methods are best understood as a way to scale the search for either kind of signal across more data than a human can process — but the output still needs a human-understandable causal hypothesis behind it, or it’s just trend-following with extra math and false confidence in the math.

The fallacy common to all three, used in isolation, is the same one underneath each individual critique: mistaking a model of reality for the model of reality — whether that model is a discounted cash flow, a moving average crossover, or the hidden layers of a neural net.

Caution: This post is a theoretical comparison of investment frameworks and the role of AI within them. It is not investment advice, and nothing here should be read as a recommendation to buy, sell, or hold any security or asset. Investment decisions should be made in consultation with a licensed financial advisor who can account for your specific circumstances, goals, and risk tolerance.