The Formula That Should Work But Doesn’t
Feb 3, 2026
High-conviction investors blow up. Over-diversified investors underperform indexes. Between these two failure modes sits a mathematical solution that promises optimal position sizing: the Kelly Criterion. The formula is elegant. It maximizes long-term compound growth while theoretically protecting against ruin. It has been used by blackjack players, hedge funds, and legendary investors. It should solve the allocation problem permanently.
It does not. Kelly criterion limitations are not mathematical. They are human. The formula requires two inputs: your probability of being right and your payoff ratio. Both inputs must be accurate for the output to be useful. Both inputs are almost impossible to estimate correctly. The formula is perfect. The humans using it are not.
This creates the central paradox. Concentrated investors who ignore Kelly blow up from oversizing. Diversified investors who ignore Kelly underperform from undersizing. Investors who try to use Kelly blow up anyway because their probability estimates are wrong. The tool that should bridge both failure modes often accelerates them instead.
The Input Problem Nobody Solves
Kelly criterion limitations begin with edge estimation. The formula asks: what is your probability of winning this bet? That question sounds simple. Answering it honestly is nearly impossible. Studies consistently show that confidence and accuracy are weakly correlated. Traders who feel 80% certain are not right 80% of the time. They are right at rates that bear little relationship to their stated confidence.
Overconfidence bias infects every probability estimate. When you research a company thoroughly, conviction rises. The filings make sense, the thesis feels obvious, and certainty builds with each confirming data point. That certainty is the trap. Everyone with access to the same information did the same work. Your feeling of edge is not evidence of actual edge. Yet Kelly requires you to quantify that edge precisely.
The formula amplifies errors in both directions. If you overestimate your edge, Kelly tells you to size larger than you should. The position blows up and takes your capital with it. If you underestimate your edge, Kelly tells you to size smaller than optimal. You leave returns on the table across hundreds of trades. Either way, garbage in produces garbage out. Kelly criterion limitations are fundamentally input limitations disguised as a position sizing problem.
The investors who understand Kelly best often describe it as a thinking tool rather than a calculator. It forces you to estimate probability rather than guess, to quantify uncertainty rather than ignore it. But forcing yourself to think about probability does not make your estimates accurate. It just makes you aware of how little you actually know.
Long-Term Capital Management: Nobel Prizes Meet Kelly’s Limits
If Kelly criterion limitations could be overcome by intelligence, Long-Term Capital Management would have thrived. The fund employed two Nobel laureates in economics, former Federal Reserve officials, and some of Wall Street’s most sophisticated quantitative minds. Their models were rigorous. Their early returns were spectacular, compounding at 40% annually in the first years.
LTCM sized positions based on mathematical frameworks that incorporated probability estimates, correlation assumptions, and historical data. By 1998, they had leveraged their positions to 25-to-1. Their models said this was optimal because the positions were hedged and historical relationships were stable. Then Russia defaulted on its debt.
Correlations that had held for years inverted simultaneously. Every hedge became a concentrated bet in the same direction. The fund lost $4.6 billion in weeks and nearly collapsed the global financial system. The Kelly criterion limitations were not in the math. They were in the assumption that probability estimates based on historical data would hold in unprecedented conditions.
LTCM’s team had more data, better models, and deeper expertise than almost any investor alive. They still could not estimate their true edge accurately enough for optimal position sizing. The Kelly formula told them to size aggressively because their backtested win rates looked excellent. Reality disagreed violently. If Nobel laureates cannot overcome Kelly criterion limitations through superior analysis, retail investors should assume they cannot either.
Buffett’s Edge Is Not Your Edge
Warren Buffett runs concentrated positions that would make most advisors faint. Apple alone has represented nearly half of Berkshire’s public equity holdings at times. This appears to validate Kelly-style conviction betting. If concentration works for Buffett, why do Kelly criterion limitations apply to everyone else?
The answer is asymmetric information that retail investors cannot access. When Buffett invested $5 billion in Goldman Sachs during the 2008 crisis, he negotiated 10% preferred dividends and warrants that regular investors could not obtain. His probability of positive outcome was structurally higher than yours would be buying the same stock at the same price. His edge was real and measurable because it was contractual, not analytical.
Buffett also operates without redemption pressure. Berkshire’s permanent capital base allows positions to recover from drawdowns that would force other investors to liquidate at the worst moment. Time horizon transforms the probability calculation. A position that has 60% odds of working over ten years might have only 40% odds of working over one year. Buffett’s Kelly inputs are different because his constraints are different.
Retail investors who copy Buffett’s concentration without his structural advantages are not applying the Kelly criterion correctly. They are plugging in probability estimates based on hope rather than contractual certainty. The visible output is concentrated positions. The invisible input is edge that does not exist. Kelly criterion limitations become fatal when the edge you believe you have is imaginary.
Sizing Without Knowing Your Edge
If Kelly criterion limitations stem from unreliable probability estimates, the solution is not better estimation. It is sizing based on factors you can actually measure. Two stand out: volatility and liquidity.
Volatility-based sizing adjusts position size inversely to price movement. A stock that swings 5% daily gets a smaller allocation than one that moves 1% daily, regardless of your conviction. This approach sidesteps Kelly criterion limitations entirely. It does not ask what your edge is. It asks how much damage you will sustain if you are wrong. High-volatility positions get smaller allocations because the penalty for error is larger, not because the thesis is weaker.
Liquidity-based sizing limits positions based on exit capacity. If you cannot sell a position without moving the market against yourself, the position is too large regardless of what Kelly suggests. Capping allocations at a fraction of daily volume ensures you can exit when circumstances change. This constraint operates independently of probability estimates. It measures something real rather than something imagined.
Neither framework requires accurate self-assessment. Both impose discipline externally. They work precisely because they bypass the input problem that makes Kelly criterion limitations so dangerous. You do not need to know your edge. You need to know your volatility exposure and your liquidity constraints. Those numbers do not lie to you the way your confidence does.
The Structural Problem Beneath the Formula
Kelly criterion limitations are not fixable through better analysis or more rigorous probability estimation. The formula assumes rational self-assessment is possible. Decades of behavioral research demonstrate it is not. Humans systematically overestimate their abilities across every domain studied. Expertise makes this worse, not better. The more you know about investing, the more confident you become, but your accuracy does not increase at the same rate.
This creates an unsolvable problem at the heart of Kelly-based position sizing. Optimal allocation requires knowing your edge. Knowing your edge requires accurate probability estimates. Accurate probability estimates require accurate self-assessment. Accurate self-assessment is psychologically impossible for most people. The chain breaks at the human link, not the mathematical one.
The investors who use Kelly most successfully describe it as a discipline for thinking rather than a formula for sizing. It forces you to confront uncertainty explicitly. It forces you to admit what you do not know. That confrontation is valuable even when the output is unreliable. But confusing the thinking benefit with sizing accuracy is how Kelly criterion limitations produce catastrophic losses.
Build systems that assume your probability estimates are wrong. Hard position limits that trigger regardless of conviction. Volatility-based constraints that ignore your thesis entirely. Liquidity rules that cap exposure based on exit capacity. These systems feel frustrating because they dismiss your research. That is the point. Your research is generating confidence, not accuracy. A system that ignores your confidence will outperform one that trusts it.
