Alpha After AI

For decades, investors with the greatest perceived edge were often those who could process information, move faster, outwork their competition. Edge can be characterized in many ways but it has frequently accepted the definition of informational asymmetry or analytical length. The best analysts are viewed as the ones who can build models faster with quick shortcuts, scan transcripts, and synthesize information better than those around them.

Markets rarely price change until it becomes undeniable. The same is true here. We expected this shift but what we didn't expect was how quickly it would start becoming reality.

Artificial Intelligence is rapidly commoditizing traditional "advantages." Tasks that used to signal differentiation at the early stage; building a DCF from scratch, scanning company ratios, reviewing earnings transcripts, or even drafting memos can now be executed in seconds by tools such as ChatGPT or Claude. Investing won't be easier, the skills expected of a typical analyst however must change.

Knowing how to build a model from scratch is no longer rare. Understanding what a model misses, where its assumptions break, and how to use it in service of an original view is a very different game. Memorizing technical frameworks or reproducing textbook answers is no longer impressive when AI can do the same faster and often more cleanly. If a machine can replicate your output in a fraction of the time, your value cannot lie in the output alone.

Investors have always generated alpha by seeing what others do not. The future edge is hard to define but it will surely not lie in executional prowess. AI is extremely good at pattern recognition in what it's seen before. The struggles unravel with true structural breaks, when the past stops predicting the future. The investor who can reason well about novel situations, where there isn't easily accessible training data has an edge that cannot be replaced. When old relationships weaken, regimes change or when historical data becomes a less reliable guide. In those moments, the human investor who can reason ambiguity and form a view where no template exists is convincing.

Think equity premium puzzle. Historically, stocks outperform risk-free assets (like treasury bills) by 6-8% annually. While this appears to be compensation for risk, rational economic models struggle to justify such a high premium without assuming absurd levels of risk aversion. Markets do not just reward risk but rather how humans perceive it. The answer is not purely mathematical but behavioral. Our very equity premium puzzle exposes a limitation of purely computational systems. An AI trained on historical data would not "expect" such a persistent premium to even exist if it used historical data. AI can model risk but markets are driven by how humans feel about them.

A recent Harvard study provides a strong illustration of the future. Researchers found that an AI model could predict 71% of fund trades trained on 1990-2023 data. Therefore, most of what managers do is pattern-based and learnable but the striking insight was that the 29% of decisions that AI could not predict were the ones most associated with alpha.

In essence, AI dominates the predictable part of investing but our future alpha will come from recognizing when predictability breaks.

Markets are adaptive and as strategies become understood, they naturally arbitrage away, forcing new dynamics to come to light. If this edge could so easily be learned or repeated, it would too become commoditized but the most valuable decisions clearly are not the ones that can be systematically repeated but those made under dynamic conditions.

Humans cannot compete with AI in processing. Our edge lies in recognizing when what has worked before no longer applies.