Kelly criterion is a formula developed by John Kelly at Bell Labs in 1956 that tells you the optimal fraction of your capital to risk on a bet, given a known edge and reward/risk ratio. "Optimal" means it maximises the long-run growth rate of your capital — not the expected value of a single trade, but the compounded result over many trades.

For example: if you have a 70% win rate (p = 0.70) and your credit spread pays $1.20 on a $8.80 max loss (b = 0.136), then: f* = (0.70 × 0.136 − 0.30) / 0.136 = −1.45. Negative — no edge at these terms.

The insight is that overbetting — risking more than Kelly — guarantees ruin in the long run, even on a positive-expectation bet. A player who consistently bets 2× Kelly will eventually go broke, regardless of how favorable the bet appears.

High score and NO EDGE measure different things. Mega-Alpha asks: "Are conditions favorable for selling premium on this stock?" Kelly asks: "At the actual premium and max loss of this specific trade, does the math produce positive expected growth?"

The disconnect happens when IV is elevated enough to score well but the specific trade structure has a bad b-ratio. Consider an iron condor on a $200 stock:

The most common causes:

• IV isn't elevated enough yet — the score may reflect improving conditions without IV being high enough to generate meaningful credit
• The structure is too conservative for the current IV level — iron condors recommended at 45–60% alpha often operate in a zone where IV is moderate and premium is thin
• Wing width is too narrow — the auto-computed wing (5% of stock price) can produce a tight spread where the b-ratio is structurally poor

NO EDGE is the tool telling you to either wait for IV to expand further, widen your strikes manually, or look for a higher-scoring name that unlocks a more aggressive structure with a better b-ratio.

The raw Kelly fraction assumes your inputs — win probability p and reward/risk b — are precisely known. In practice, neither is. Win probability comes from a model estimate, not from backtested outcomes on this specific strategy. The b-ratio is computed from Black-Scholes with a flat vol surface, which understates put premiums due to skew.

Betting full Kelly on uncertain inputs amplifies errors catastrophically. If you think you have a 70% win rate but your true win rate is 62%, full Kelly on the estimated 70% will overbet by a material amount and erode capital faster than if you'd bet less.

Many professional options traders use a 25–50% Kelly fraction for exactly this reason. The 0.6× factor puts this tool at the conservative end of that range, which is appropriate given that p is model-derived rather than empirically calibrated.

Win probability (p) is the estimated chance the option expires worthless and you keep the full credit. It is derived from the Mega-Alpha score, not from historical trade outcomes.

The mapping is deliberately conservative: a score of 0.70 (strong) maps to roughly 68–72% win probability. This is grounded in general empirical findings on short put win rates at 20–30 delta, but is not calibrated to actual P&L data from this tool. There is no backtest underlying the number.

A Tier 2 data integration (historical IV data, realized VRP by sector) could eventually allow empirically calibrated win rates. Until then, the 0.6× discount partially compensates for the model uncertainty.

b is the reward/risk ratio: how much you gain on a win divided by how much you lose on a loss. For options premium selling, the gain is the net credit received and the loss is the net capital at risk if the trade goes against you.

The tool computes b automatically per structure:

For defined-risk structures, wing width is auto-computed as 5% of stock price, rounded to the nearest $5 (minimum $5). A narrow wing on a low-IV stock produces a very low b — which is often the direct cause of NO EDGE.

Black-Scholes with ATM IV is used to estimate the credit. Because the vol surface is not flat (puts trade at higher IV than ATM), actual put premium is slightly higher than estimated. This means the tool's b-ratio is slightly conservative — real fills are marginally better than shown.

No. Empirical Kelly is a ceiling, not a target. There are several reasons to size below it:

• Correlation risk. If you're running multiple positions simultaneously, each individual Kelly ignores the fact that a broad market selloff will stress all your short puts at once. With 5 correlated positions each at 5%, you have 25% of capital in highly correlated risk — not 5 independent bets.
• Liquidity and execution. The tool sizes in contracts. On a small portfolio, the Kelly-suggested contract count may round to 1 regardless of the percentage, making the exact allocation academic.
• Conviction and data quality. A trade where you have high conviction on the setup and clean data justifies more of the Kelly allocation than a trade where the data is messy or the signal is borderline.

Think of Empirical Kelly as answering: "what's the most I should risk on this, assuming I'm running it in isolation?" Real portfolio management then layers correlation and concentration limits on top.

A negative Kelly fraction means the trade has negative expected growth rate at these terms. It doesn't necessarily mean the trade is a loser in expected value terms — it means that even if it has a slight positive expected value, the variance is so large relative to the edge that repeated betting will erode your capital over time.

Intuitively: if b is very small (thin credit, large max loss), you need an extremely high win rate to overcome it. A narrow iron condor collecting $0.80 on a $9.20 max loss has b ≈ 0.087. To break even on Kelly you'd need:

The fix is never to override a negative Kelly and take the trade anyway "because the score is good." The score measures setup quality; Kelly measures whether the specific sizing makes mathematical sense. A good setup at bad terms is still a bad bet. Widen the spread, wait for higher IV, or skip.

Kelly sizing is always expressed as a fraction of total capital, not a fixed dollar amount. The Empirical Kelly percentage (e.g., 3.2%) is fixed for a given trade setup — it's determined by p and b, not by your account size.

The dollar allocation is simply empirical_kelly% × portfolio_size. If Kelly says 3.2% and your portfolio is $100,000, that's $3,200. If your portfolio is $50,000, it's $1,600. The fraction is the same; the dollars scale.

This is correct behaviour — it means Kelly automatically sizes down for smaller accounts and up for larger ones, always allocating the same proportional edge relative to your total capital. The contract count then rounds this dollar amount to the nearest whole contract based on the capital at risk per contract.

Classic Kelly assumes sequential, independent bets. Short premium positions are neither — they run concurrently and are correlated (most short puts lose when the market drops broadly). This is the most important practical limitation of applying Kelly to an options portfolio.

The tool applies Kelly per trade in isolation, which overstates the safe size when multiple positions are open simultaneously. The mitigations built in:

• The 5% hard cap per position means a single trade can never exceed 5% of capital regardless of what Kelly computes
• The 0.6× discount further reduces each position below raw Kelly
• The iron rule of 35% maximum total exposure is the portfolio-level guard against concurrent correlation

For a rigorous multi-position Kelly, you'd need a covariance matrix of position outcomes and solve a multi-dimensional Kelly problem. That's beyond the scope of this tool. The per-position caps and total exposure limit are practical substitutes that prevent the worst outcomes.

5% per position is a widely used convention in systematic options strategies. The reasoning:

• Ruin prevention. A 5% position that goes to zero (full wing loss) costs you 5% of capital. Even 3 simultaneous full losses cost 15% — painful but survivable. A 20% position that goes to zero is a portfolio-ending event.
• Diversification math. At 5% max per position with a 35% total exposure limit, you're running 7 positions maximum. Seven uncorrelated positions is a meaningful sample for the law of large numbers to work in your favor. Fewer, larger positions means each individual outcome matters too much.
• Model uncertainty. If the win probability estimate is off by 10 percentage points, a 5% position produces a manageable error. The same estimation error on a 20% position is a serious drawdown.

The 5% cap is conservative for a strategy with a genuine edge. If you have high conviction, a deep understanding of the position, and are running it in true isolation, you might reasonably go to 8–10%. But the tool defaults to 5% because it can't verify those conditions, and the cost of being wrong at 5% is far lower than at 10%.