What the model does,
and where it breaks.
How we estimate a trade’s odds, why a high win rate isn’t enough on its own, how survivors get ranked, and how we check that our stated probabilities hold up out of sample — including an honest list of the assumptions that can fail.
The pipeline
A scan is a funnel, not a ranking. Four stages run in a fixed order, and each is allowed to end the scan on its own.
Stage 1 — portfolio level
Is this a sane day to be short options?
Before any contract is priced, a market-wide regime check reads volatility term structure, short-dated volatility, the level of implied volatility, vol-of-vol, and high-yield credit spreads. In a high-crash-risk regime we surface no new short-premium ideas at all; in the Income Finder the chain stays visible behind a prominent crash-risk warning. Stale or missing feeds are reported rather than assumed benign.
Stage 2 — per underlying
Is premium on this name worth selling?
IV Rank places today's implied volatility inside the name's own recent range. The volatility risk premium compares implied against realized volatility (estimated with a gap-robust, drift-independent estimator, so overnight jumps are not smoothed away). For puts we also require 25-delta skew — the market paying specifically for tail risk. Single-stock names are gated through earnings.
Stage 3 — per contract
Is this specific option tradeable and worth it?
Each contract must clear side, days-to-expiry window, a live mid above zero, bid-ask spread as a fraction of mid, and real open interest or volume. Probability of profit is computed at market implied volatility, including the credit cushion. Then the expected-value test runs at realized volatility — the gate that rejects high-POP, negative-expectation trades.
Stage 4 — ranking
Which survivor is actually best for you?
Survivors are converted to after-tax annualized return on the capital each one commits, applying Section 1256 60/40 treatment where it applies and your own rates from Tax Settings. Covered calls are instead ranked on net-vs-hold: after-tax premium less expected upside forgone and expected assignment tax.
When nothing survives, the scan reports which gate stopped it rather than returning a bare empty list. An empty result is information.
What “probability of profit” means here
For each contract we price the option with Black-Scholes at the market’s implied volatility, then compute the probability that the underlying finishes on the profitable side of your strike including the credit cushion you collect. A 90% POP means the model estimates a roughly 9-in-10 chance the trade expires a winner — under its assumptions of lognormal returns and constant volatility.
Real markets have fatter tails than that, so POP is never the last word. It is the first of three numbers, and the least important of them.
Why win rate alone is misleading
A far-out-of-the-money option can show a 90%+ win rate and still be a losing trade on average if the occasional loss is deep enough. So every candidate must also clear a positive expected-value filter: we integrate the option’s payoff over a lognormal terminal distribution driven by the underlying’s realized volatility rather than its implied volatility.
That deliberate mismatch is the point. The credit is collected at the market’s price for risk; the risk is evaluated at the stock’s own behaviour. A genuinely priced edge shows up as positive expected value, and a fairly-priced lottery ticket shows up as roughly zero. Non-positive expected value is rejected. Alongside it we report expected shortfall — the average loss given that the trade loses — and the modeled worst case.
Every gate a trade must pass
Regime gate
High-crash-risk market conditions suppress new short-premium ideas across the board.Rejects: everything, regardless of trade quality
IV Rank floor
Implied volatility must be rich relative to the name's own recent history — not just high in absolute terms.Rejects: poorly paid risk
Volatility risk premium
Implied must exceed realized volatility. Compressed premium in a green regime may pass with a size-down warning; negative edge hard-fails.Rejects: underpriced insurance
Put skew
Out-of-the-money puts must trade at a volatility premium to at-the-money — evidence the tail is being paid for.Rejects: uncompensated tail risk
Earnings gate
Single-stock premium is not recommended through an earnings report, where the jump risk is unmodellable.Rejects: event gambles
Expected value
The payoff is integrated over a lognormal terminal distribution driven by realized volatility. Non-positive expected value is rejected.Rejects: profitable-looking losing bets
Liquidity
Tight bid-ask relative to mid, plus genuine open interest or volume, so a quoted edge can actually be captured.Rejects: paper edges
DTE floor
No new entries inside roughly 14 days, where gamma can swing a high-POP position to a full loss intraday.Rejects: gamma-zone entries
Survivors are then ranked by after-tax annualized return on committed capital, with Section 1256 60/40 treatment applied automatically to broad-based index options and covered calls scored on net-vs-hold instead.
How we calibrate: the walk-forward back-test
A model that says 90% is only useful if the trades it labels 90% actually win about 90% of the time. We replay history walk-forward: at each past date we run the same gates against only the data available then, record the trades the model would have surfaced, and follow every one to expiry.
Outcomes are bucketed by predicted POP and compared against the realized win rate in each bucket, with a Wilson 95% confidence interval so small samples aren’t over-read:
| Predicted POP bucket | Trades | Realized wins | Wilson 95% CI | Verdict |
|---|---|---|---|---|
| 90–95% | 412 | 91.3% | 88.2 – 93.7% | Calibrated |
| 95–99% | 96 | 92.7% | 85.6 – 96.5% | Overstated |
| 85–90% | 28 | 89.3% | 72.8 – 96.3% | Too few trades |
A bucket is calibrated when the realized rate falls inside its interval, and overstated when realized wins land below what was predicted — a signal to tighten that gate. Wide intervals on small samples mean the honest answer is “we don’t know yet,” and we label them that way rather than reporting a flattering point estimate.
The back-test also runs shadow gates: for each gate, we record what would have happened to the trades it rejected. That is how we can tell a gate that earns its keep from one that is merely expensive.
What calibration is not
Calibration is a property of the model measured over history, not a promise about any single future trade. Volatility regimes shift, the sample includes only the crises that happened to occur in it, and past calibration does not guarantee future results.
Honest limitations
Every item below is a place where the model is known to be wrong or incomplete. We publish them because a risk you can see is a risk you can size for.
- Fat tails. Black-Scholes assumes lognormal returns. Real crashes are larger and faster than the distribution allows, so the modeled worst case is an underestimate of the genuine worst case.
- Trailing realized volatility. The expected-value test uses volatility measured from the recent past. After a long calm stretch this understates forward risk, which is exactly when premium selling looks most attractive.
- No jump process. Gaps and single-name shocks are not modeled as discrete events, only smeared into a continuous distribution.
- Expiry-only payoffs. The tail model scores where the stock finishes, not the path it takes. Path matters most near expiry — which is part of why we refuse late entries.
- No portfolio correlation. Gates are evaluated per trade. Ten short puts across ten names in one market are far more correlated than ten independent bets, and position sizing remains your responsibility.
- Data quality. Sandbox and live feeds may omit greeks; we fall back to model deltas. Where IV history is unavailable we substitute a realized-volatility proxy for IV Rank and label the scan accordingly.
- Your inputs. Estimates use the tax rates and assumptions you provide. Garbage in, garbage out.
- Early assignment. Nothing here accounts for early assignment on American-style equity options beyond the dividend-vs- extrinsic check.
Prefer the plain-language version?
The Learn course walks through the same ideas with worked examples and diagrams — win rate versus expected value, the volatility risk premium, market regime, gamma, covered-call trade-offs and taxes.
Start the courseSee the FAQ for assignment, rolling and downside questions, or the Disclosures. Nest Money is an educational tool — not investment, tax, or legal advice, and not a registered investment adviser or broker-dealer. Probability, yield and after-tax figures are model estimates, not guarantees. Options involve substantial risk of loss.