AI scoring model analysis: How is Score calculated?
Why not simply sort
Many people ask: Why not sort directly by annualized income? The tallest is the best?
Answer: No.
A contract with an annualized return of 50% but an exercise probability of 30% is far inferior to a contract with an annualized return of 20% but an exercise probability of 2%.
Score is to solve this problem - not to find the one with the highest return, but to find the one with the best risk-benefit ratio.
Five Dimensions of Score
Hyperstock Score combines 5 dimensions, each with different weights:
| Dimensions | weight | illustrate | Why is it important |
|---|---|---|---|
| annualized rate of return | 25% | Royalty ÷ Margin × Time Annualization | Directly reflects capital efficiency |
| Assignment probability | 25% | Possibility of exercise at maturity | Life-saving indicator, the lower the safer it is |
| Margin efficiency | 20% | The income generated by unit margin | Fund utilization rate |
| volatility adjustment | 20% | Current IV compared to historical IV | Avoid IV Crush risk |
| time decay | 10% | Number of days until expiration, Theta decay rate | Time value harvesting efficiency |
Total score range: 0 - 10,000
How to calculate each dimension
1. Annualized rate of return (25%)
Annualized income = (premium / SPAN margin) × (365 / remaining days) × 100%
- Scoring method: Score in segments
2. Assignment probability (25%)
Assignment probability ≈ |Delta| + volatility adjustment term
- Scoring method: reverse scoring, the lower the better
Key: The exercise probability weight is the same as the annualized return, which shows that we believe that safety and return are equally important.
3. Margin efficiency (20%)
Margin efficiency = premium / SPAN margin
- For the same principal, contracts that can collect more premiums have higher scores.
- Avoid funds being used inefficiently
4. Volatility adjustment (20%)
IV Percentile = current IV / 52-week IV range
- IV Percentile > 80% → Bonus points (high IV, high royalties, favorable to the seller)
- IV Percentile < 20% → Point reduction (IV is low, the premium is cheap, and the income is small)
- But IV > 90% and the event is close → downgrade (event risk is too high)
5. Time decay (10%)
Theta efficiency = Theta / option price
- Theta High → Time value decays quickly → Good for sellers
- 30-45 days remaining on contract Theta sweetest
Comprehensive calculation example
Assume that the scores of each dimension of a contract are:
| Dimensions | original value | Dimension score (0-100) |
|---|---|---|
| annualized rate of return | 25% | 85 |
| Assignment probability | 4.2% | 88 |
| Margin efficiency | 12% | 80 |
| volatility adjustment | IV 65th percentile | 75 |
| time decay | DTE 35 | 82 |
Score = 85×0.25 + 88×0.25 + 80×0.20 + 75×0.20 + 82×0.10= 21.25 + 22.00 + 16.00 + 15.00 + 8.20= 82.45 (percentage system) = 8,245 (ten thousandths system)
Why not simply sort
Compare the two methods:
| method | Recommended #1 | question |
|---|---|---|
| Sort by annualized | Annualized 50%, exercise probability 28% | There is a high probability that the option will be exercised, and the loss will far exceed the premium. |
| Sort by Score | Annualized 22%, exercise probability 3.5% | The best risk-benefit ratio |
The core philosophy of Score: live and make money.
For a contract with a high exercise probability, no matter how high the annualized rate is, Score will be beaten down. Because we believe that not being exercised is the first priority of Sell Put.
Limitations of Score
Score cannot tell you:
- Will this company suddenly explode (black swan)?
- Will the financial report exceed expectations (event risk)
- Will macroeconomic policies change suddenly (systemic risk)
Score is a sorting based on a mathematical model, not a crystal ball.
Suggestions for use:
- Score > 7000 → Priority
- Score 5000-7000 → Judgment based on fundamentals
- Score < 5000 → Be cautious unless you have special judgment
- Always combine financial warnings and event calendars
We will continue to optimize
Score Models are not static. We will base on:
- User feedback (which recommendations actually work well)
- Market changes (the optimal weight may be different in different cycles)
- Academic research (new option pricing models)
Continuously adjust weights and algorithms.
Have suggestions? Send email:hyperstock.werich@gmail.com