The math behind 70 to 85 percent win rates, stated precisely enough that a statistician would sign off, and plainly enough that you can use it by Friday. The theorem, the fine print, the peer-reviewed edge, and a real SPY bear call spread.

I am a quant, through and through. Almost every decision I make relies on strict adherence to mathematical and statistical methods, and it is the main reason I have navigated the markets successfully for more than two decades: simply sticking to the probabilities. Benjamin Graham drew the famous distinction, one Jason Zweig's Intelligent Investor commentary returns to often: the market is a voting machine in the short term and a weighing machine in the long run. Short-term noise is meaningless. Long-term probabilities are everything. The bridge between those two sentences is a single theorem, the Law of Large Numbers, and this article states it carefully, prices a real trade with it, and names the fine print.

The Law of Large Numbers, in the form proved for exactly our situation, says this: take a sequence of independent trials that each succeed with probability p, and track the running proportion of successes. As the trials accumulate, that proportion converges toward p, with the probability of any meaningful gap shrinking toward zero. Your middle-school coin flip was the first version you met: a fair coin's heads proportion homes in on 50 percent as the flips pile up.

Two precisions separate the theorem from the folklore version. First, the law governs the proportion, not the count: over 10,000 flips the heads percentage sits very near 50 while the raw heads-minus-tails gap can wander into the hundreds. The law works by dilution, not correction. Second, the gambler's-fallacy vaccine: the coin does not remember. After four losses, the fifth trade's odds are unchanged. The proportion recovers not because a win is "due" but because a fixed-rate process, given enough trials, drowns any finite bad stretch in a growing ocean of ordinary ones.

The Central Limit Theorem is this law's companion, and most versions of this argument garble the relationship, so here it is exactly: the Law of Large Numbers tells you the destination; the Central Limit Theorem tells you the size and shape of the wobble along the way. For a strategy with true win rate p observed over n trades, the standard error of your measured win rate is the square root of p times one minus p, divided by n. At p equal to 0.80, that formula produces numbers worth taping to the monitor: after 10 trades your measured rate carries a standard error of 12.6 points, a 95 percent confidence interval spanning roughly 55 to 100 percent: ten trades tell you almost nothing. After 50 trades the interval is 69 to 91. After 100, it is 72 to 88. Only near 300 trades does it tighten to 76 to 84. The theorem is not slow because the math is weak; it is slow because that is what honest uncertainty looks like.

Run the standard-error arithmetic in reverse and it becomes a discipline tool: it says what your own results entitle you to conclude. Winning 7 of 10 is fully consistent with a true 80 percent strategy, and also with a 60 percent one; the sample cannot tell them apart. Winning 74 of 100 sits comfortably inside an honest 80 percent process. And it cuts both ways: a hot 9 of 10 is equally incapable of proving your strategy is better than advertised. The Law of Large Numbers is a contract with a minimum term, and the fifty-trade minimum I set for judging any system is not a folksy rule of thumb; it is the sample size at which the confidence interval first becomes narrow enough to mean anything.

Here this article parts company with the usual treatment, because a statistical case is only unarguable if it survives its own assumptions. The textbook theorem assumes independent, identically distributed trials, and trading violates both, mildly but genuinely, in three ways. Your probabilities are estimates, not facts: the 80 percent on your screen is a model's output, and model error means your true rate might be 77 or 82. Your trades are not identical: every position carries its own probability, so what converges is your results against the average of your stated odds. And trades can be correlated: five bear call spreads on five index ETFs opened the same week are not five independent trials; they are closer to one large trial wearing five ticker symbols, because the same market move settles all of them together.

None of these breaks the argument, and the reasons are themselves standard results: convergence survives non-identical trials (the law holds for averages of differing probabilities) and weak dependence between them. What it cannot survive is heavy, simultaneous correlation, which quietly shrinks your effective sample size: a hundred trades clustered into twenty correlated batches carry the error bars of twenty while feeling like the track record of a hundred. That is a sizing instruction, collected below. The honest summary: the theorem applies to real options selling the way engineering formulas apply to real bridges, with tolerances, and the tolerances are where the discipline goes.

Now the question a sharp reader is already asking: if the probabilities are known to everyone, why does selling an 80 percent trade make money at all? Price it and the puzzle sharpens. The SPY spread below risks $392 to make $108 at an 80.1 percent success rate, so its expected value held to expiration is 0.801 times $108 minus 0.199 times $392: positive $8.50 per spread, before commissions. Gross, thin, and honest. If option prices were perfectly fair, high-probability selling would be a treadmill.

They have not, historically, been perfectly fair, and this is the one place the academic literature does the heavy lifting for us. Peer-reviewed research has documented a persistent variance risk premium in index options: implied volatility has tended to run above the volatility subsequently realized, meaning sellers of index option premium have historically been paid slightly more than the risk delivered. Carr and Wu's 2009 study in the Review of Financial Studies quantified the premium across markets; Bakshi and Kapadia found the same signature in delta-hedged option returns in 2003; Bondarenko's 2014 work asked directly why index puts have been so persistently expensive. And a study commissioned through the Options Industry Council found the same effect operating at the strategy level. The premium is compensation, not charity: sellers are paid to absorb the market's demand for insurance, and the payment arrives with occasional insurance-sized losses attached. It is the documented, replicated reason disciplined premium selling has historically sat above the treadmill, and why the sixth rule below, sell when implied volatility is elevated relative to its own range, is an edge amplifier rather than a superstition. Add the management layer, taking profits at 50 to 75 percent of maximum rather than riding every trade to expiration, and the realized distribution improves further still.

Suppose SPY trades at $589.49 in a calm-volatility stretch, and my 30 to 60 day outlook runs bearish to neutral. A bear call spread fits: sell a call above the market, buy another at a higher strike in the same expiration, risk defined before entry. I sell the 612 call and buy the 617, about 43 days out, and collect $1.08 at the mid, a fill the world's most liquid option chain routinely allows.

The ledger, every line checkable: premium collected, $1.08, which is $108 per spread. Maximum loss, $3.92 per share, always the $5 width minus the credit, $392 per spread. Potential return, 27.6 percent on risk over the life of the trade. Probability of success, 80.1 percent, with the short strike sitting at roughly a 21 delta. Breakeven, $613.08, the short strike plus the credit. Margin of error, $22.51 of rally room, meaning SPY must climb 3.8 percent against the position before it begins to lose at expiration. Sideways, down, and a modest rally all win; the trade needs one specific thing not to happen.

One bear call spread proves nothing, and here the vague reassurance that "losing streaks happen" gets replaced with exact numbers, computed rather than estimated. For a true 80 percent process, the probability of suffering at least one streak of three consecutive losses somewhere in your next 100 trades is 47 percent, nearly a coin flip. A four-loss streak appears in about 12 percent of hundred-trade samples, and across a 300-trade career, one trader in three will endure one. Five in a row shows up for roughly one trader in fifteen over 300 trades. These are not signs of a broken strategy; they are the arithmetic signature of a working one, because an honest 20 percent loss rate must occasionally cluster.

This is where most traders fail, and Graham said why with a line Zweig's commentary keeps front and center: the investor's chief problem, and even his worst enemy, is likely to be himself. The emotional response to a statistically ordinary losing streak destroys more accounts than any market crash. The trader who abandons a sound process after four losses is not escaping variance; he is locking in its worst stretch and forfeiting the convergence he paid for. The Law of Large Numbers pays its coupon only to holders who do not sell the bond in week two.

The theorem has one demand, and it is not intelligence: survival. Convergence happens across many trades, so anything that removes you from the game before the sample matures, financially or psychologically, forfeits the edge entirely. Position sizing is where the earlier fine print comes home. Keeping each position at 2 to 3 percent of capital, 5 percent as an absolute ceiling, does two jobs at once: it makes a maximum loss a recoverable event rather than a setback measured in months, and it limits the correlated-cluster problem, because small positions across staggered entries keep your hundred trades behaving like closer to a hundred independent draws instead of twenty batches. Sizing is not separate from the statistics. Sizing is what makes the statistics apply to you.

Choose strategies that let you select your probability before entry: bear call spreads, bull put spreads, iron condors, cash-secured puts, targeted in the 70 to 85 percent band. Define the risk with spreads that cap the maximum loss; no surprises. Keep positions at 2 to 3 percent of capital, 5 at the ceiling, staggered rather than clustered. Withhold judgment until 50 trades; the error bars say anything earlier is noise wearing a costume. Hold the line through streaks the math told you to expect. And sell when implied volatility sits high in its own range, when the documented premium is fattest.

The Law of Large Numbers rewards patience mechanically, not morally. Traders who chase home runs, switch strategies monthly, or quit after a normal streak never hold the position long enough for convergence to pay. I have traded this exact framework for more than two decades: probability selected before entry, risk defined, positions small, enough trades for the proportion to find its destination. It is not glamorous. It is a weighing machine, and I feed it. The votes are noise. The weight is the edge.

Probabilities over predictions,

Andy Crowder

📩 Want to see how a 24+ year professional options trader approaches the market?

Subscribe to The Option Premium, my free weekly newsletter where I share:

No hype. No predictions. Just the frameworks I've used to trade options for over two decades.

📺 Want more? Follow me on YouTube for in-depth tutorials, live trade analysis, and the kind of education you won't find anywhere else.

Connect with me:

This newsletter is for educational purposes only and should not be considered investment advice. Options trading involves significant risk and is not suitable for all investors. Past performance does not guarantee future results. Always consult with a qualified financial professional before making investment decisions.