Risk Models Ch. 1: The Kelly Criterion — Why Bet Size Is the Whole Game

Investment Background

Every book in this library that discusses an "edge" — a positive expectancy, a factor premium, a mispricing — eventually runs into the same unanswered question: having found the edge, how much do you bet on it?

Trading in the Zone teaches you to think probabilistically and trust the process over any single outcome. Market Wizards profiles traders who all say some version of "position sizing matters more than entry." Neither book gives you the number.

The Kelly Criterion, developed by Bell Labs scientist John L. Kelly Jr. in 1956, is the number. It is a formula — not a heuristic, not a rule of thumb — that computes the exact fraction of your capital to risk on a favorable bet in order to maximize the long-run growth rate of your wealth.

It was built for a completely different problem — Kelly worked on noisy telephone lines and was solving how to transmit information at maximum rate through a channel with errors. Edward Thorp, the mathematician who beat blackjack, is the one who recognized in the 1960s that the same math answers "how much of my bankroll do I bet on a card-counting edge?" The formula migrated from telecommunications to gambling to, eventually, trading and portfolio management.

The single fact that makes this chapter necessary: an investor with a genuine, real, positive edge can still be mathematically guaranteed to go broke if they size positions wrong. This is not a risk-management footnote. It is the central result.

The Wall Street Translation

The Formula

For a simple win/lose bet, the Kelly fraction is:

f* = W − (1 − W) / R

Where: - f* = fraction of capital to wager - W = probability of winning - R = ratio of win amount to loss amount (the payoff odds)

A concrete example makes this real. Suppose you have a trading system that wins 55% of the time (W = 0.55), and your average win is the same size as your average loss (R = 1). Then:

f* = 0.55 − (0.45 / 1) = 0.10

Kelly says: risk 10% of your capital on this bet. Not 2%, not 50%. A specific, derived number.

Why Betting Too Big Destroys You — Even With a Real Edge

This is the counterintuitive result that justifies the whole chapter.

Consider a trader with a genuinely favorable coin flip: 55% chance to win, even-money payoff. Every single bet has positive expected value. And yet:

Bet size (fraction of capital per flip)	Long-run outcome over many flips
10% (≈ Kelly)	Wealth grows at the maximum achievable long-run rate
20% (2× Kelly)	Wealth grows, but slower, with far larger swings
40%+ (4× Kelly)	Growth rate goes negative — wealth trends to zero with certainty over time, despite every bet still having positive expected value

Read that middle row again. A bet with positive expected value, repeated with too large a fraction of capital, still drives your wealth to zero almost surely as the number of bets grows. This is because wealth compounds multiplicatively, not additively — a single large enough loss erases gains that took many wins to build, and the "average" outcome (which is what expected value describes) is not the same as the "typical" compounded path.

This single fact explains most catastrophic blowups by traders who were, in a narrow sense, "right": they had genuine edges and still went to zero because they sized as if only the mean outcome mattered, when in a compounding, sequential game, the full distribution of paths is what determines survival.

Why Betting Too Small Also Costs You

The other side is less dramatic but still real. Betting a fraction well below Kelly (say, 2% when Kelly says 10%) does not risk ruin, but it leaves most of the achievable growth rate unclaimed. Growth rate as a function of bet size is a curve that rises to a peak at f* and then falls — sub-Kelly sizing sacrifices growth for a smoother ride; over-Kelly sizing sacrifices both growth and safety.

Fractional Kelly: The Practitioner's Answer to Estimation Error

Full Kelly assumes you know W and R exactly. You never do. Your win rate and payoff ratio are estimated from a finite history, and any estimation error pushes you toward the dangerous, right-hand side of the growth curve more often than the safe side — because the curve is asymmetric: overbetting is punished far more severely than underbetting.

This is why every serious practitioner — Thorp included — uses "fractional Kelly": betting some fraction (commonly 1/4 to 1/2) of the formula's output.

Kelly fraction used	Growth rate vs. full Kelly	Volatility of the path vs. full Kelly
Full Kelly (100%)	Maximum	Highest — large, gut-wrenching drawdowns are normal, not exceptional
Half Kelly (50%)	~75% of maximum growth rate	Roughly half the variance
Quarter Kelly (25%)	~44% of maximum growth rate	Roughly a quarter the variance

Half Kelly is the practitioner's classic compromise: you give up a modest amount of theoretical growth in exchange for a dramatically smoother, more survivable path — and it buys insurance against the near-certainty that your W and R estimates are wrong.

Executable Trading Rules

1. Never size a position off expected value alone. A positive edge tells you the direction to bet, not the size. Size is a separate calculation, and skipping it is how a real edge produces a real blowup.

2. If you use a systematic strategy, estimate W and R from as long and representative a history as you can get, then compute f*. Treat this number as an upper bound, not a target.

3. Bet a fraction of Kelly, not full Kelly. Half Kelly is the standard starting point for anyone with real uncertainty about their inputs — which is everyone.

4. Understand that large drawdowns are not evidence you did something wrong. Even correctly-sized Kelly betting produces significant volatility; the formula optimizes long-run growth, not the smoothness of any particular stretch.

5. Recognize the warning sign of over-betting: position sizes that make you unable to think clearly about the position. Kelly-sized bets should never be large enough that a single loss is catastrophic — if a loss would be catastrophic, you are already past the peak of the growth curve, on the side where more risk means less return.

Relevance to a Retirement Portfolio

Here the honest caveat matters more than the formula. The Kelly Criterion was derived for a sequence of discrete, known-odds bets — a card-counting hand, a coin flip. A diversified retirement portfolio held over decades is a different animal: returns are continuous, correlated across time and across holdings, and the "true" win probability and payoff ratio of "the stock market" are not knowable the way a blackjack edge is.

Kelly's real value to a retirement investor is not a formula to plug numbers into. It is the mental model it installs:

Kelly's lesson	How it applies to your retirement portfolio
Position size determines whether an edge helps or destroys you	This is the core argument against concentrating a retirement account in a single stock or sector "conviction bet," however well-researched
Overbetting a real edge can still lead to ruin	Applies directly to leverage — margin, leveraged ETFs — which multiplies both the edge and the chance of a ruinous path
Estimation error should push you toward smaller, not larger, bets	Applies to any tactical tilt: if you are not certain of your edge (and you should not be), size the tilt small relative to your core
Full Kelly is not the goal — survivable growth is	The entire argument for a diversified, low-cost core over concentrated, maximally-aggressive positioning

Our standard position stands: this chapter is not a license to leverage your retirement account or run a concentrated "high-conviction" sleeve at full size. It is a rigorous, mathematical explanation for why your core holding should remain a low-cost, globally diversified portfolio, and why any tactical or satellite position — even one you believe in — should be sized as a small fraction of the whole, never as a bet-the-account position.

If you take one number from this chapter, take this: professional bettors with real, measurable edges cut the formula's own output in half or more before using it. An amateur investor with a much fuzzier "edge" should be at least that conservative, and in most cases, that argues for no concentrated bet at all — just the diversified core, with any tactical conviction expressed at a size small enough that being wrong does not matter.

Chapter 2 continues with the practical mechanics: how to estimate the inputs Kelly needs when you don't have a casino's clean odds, and how position sizing interacts with the rest of a real portfolio.