Computational Irreducibility & The Return to Heuristics

Successful long-term investors rely on heuristics — Value investors must “see the light” of these methods — Conviction in a heuristic is as valuable as a logically provable, deductive solution.

VIEW AS PDF

COMPUTATIONAL IRREDUCIBILITY & THE RETURN TO HEURISTICS

I have spent over a decade searching for a proven investment philosophy. Armed with Karl Popper’s “The Logic of Scientific Discovery,” I relentlessly searched for a deductive approach that would “prove” an edge or alpha opportunity. However, there is growing evidence that this deductive approach, while scientifically valid, is far from decoding our fluid, reflexive, biased, and ultimately human markets.

Computational Irreducibility

I keep running into the term “computational irreducibility.” Coined by Stephen Wolfram, the term denotes when a problem cannot be simplified further; an answer to a computationally irreducible problem requires performing the calculation step-by-step.

Taking an example from “The End of Theory” by Richard Bookstaber, predicting the exact flight of a baseball requires inputting numerous variables and updating these variables as the ball travels. This problem is computationally irreducible as we must calculate each step of the flight path with updated variables, such as a change in windspeed, to arrive at the correct conclusion.

For these problems, no matter how fast our computers are, or how “smart” AI becomes, the problem is only solved by proceeding step-by-step through the calculation.

Is Stock Picking Computationally Irreducible?

Quantitative investors and academic researchers have, for decades, poured over financial data, looking for a way to “solve” the market. Notably, most success has come by exploiting very short-term inefficiencies, driven by the plumbing of financial markets, largely agnostic to the business underlying the security. Latency, liquidity, and information propagation arbitrage, say by having an AI read and react to an earnings report faster than a human, have all generated impressive returns.

Research into long-term, fundamental investment styles has proven less fruitful. Although many return-driving factors have been discovered, they are often competed away over time. Statistical support for most long-term investment philosophies is lacking.

Needles In The Haystack

Can any algorithm or AI-agent find long-term fundamental winners a priori? Is there a combination of a million data points that accurately identifies these businesses? Or is this a computationally irreducible problem requiring the algorithm to follow the business every step of the way, calculation by calculation, ultimately only predicting success a second before it arrives, like predicting the next millimeter a baseball may move, before requiring a new calculation to take account of a shifting wind?

A Return to Induction

Continuing with the baseball comparison, professional outfielders perform no calculations to catch a baseball. Instead, they employ an inductive approach using a heuristic described in The End of Theory: “Fix your gaze on the ball, start running, and adjust your running speed so that the angle of gaze remains constant.“

Inductive reasoning is not as strong as deductive reasoning. Concluding that “all swans are white” because an individual has never seen a black swan results in an incorrect answer. Induction exposes thinkers to mistakes.

But for computationally irreducible problems, the inductive answer can approach the deductive. The deductive solution is at its limit; as our inductive heuristics improve, we can approach this limit – calculation-free.

Why Investors Follow Buffett

Until our models of the markets get better, a mathematical approach is unlikely to predict the future long-term performance of individual stocks. Mathematically, we simply do not have the tools to model human behavior or the complexity of the markets themselves.

But, like an outfielder, there is a proven inductive method that leads us to the ball. Warren Buffett and other Superinvestors trained by Benjamin Graham have shared their value investing heuristics.

Conviction in Heuristics?

The battle for investment survival in active management is the application of these known heuristics. This is a perpetual challenge specifically because they are heuristics, not logically proven axioms.

The application of the heuristics, particularly in times of crisis, requires conviction in an unprovable approach. This is why Buffett has said that investors either have the “ah-ha” moment for value investing…or they don’t. The “ah-ha” is the conviction in the heuristic. The faith to believe in any market environment.

Lacking Conviction

This conviction is deeply lacking in the active management industry, almost by definition. Value investing legends have seen the light, committed to these heuristics, and stuck to their philosophy through crisis after crisis.

But investors trained in the industry, simply following the direction of their bosses, often fail to have the “ah-ha” moment. Conviction slips and the performance chasing begins. Heuristic following-founders are deemed outdated, annual returns are prioritized, the definition of a “moat” is stretched, new valuation techniques are employed, and styles devolve.

The sum total of this philosophy degradation is the poor aggregate performance of active management.

Conviction Over Logic

The most important attribute of an active manager is the manager’s conviction in their style, whatever that style may be. The most detrimental act of an active manager is to change styles, ultimately underperforming when they were asked to outperform. This fatal flaw, typically only identified in hindsight, is the direct result of a lack of conviction in the philosophy.

In a computationally irreducible situation, conviction in a heuristic is equally as valuable as a logically provable, deductive solution.