Options
Four crypto option strategies: the figures, the sample sizes and the corrections
Option strategies on crypto (BTC, ETH and others): what was tested, what came out of it, and how many other things were tried alongside it. Figures only — no reading of them is offered here.
Highest value produced so far (before costs, before counting the cases tried)
Level the best of 304 cases needs
This page shows 5 cases out of the 304 that this line of work can reach. For the best of 304 to mean anything, it needs a score of 3.77; the highest score anything here has ever produced, before costs and before counting, is 3.45. The cases in this part were fixed in advance; the part below it is switchable, and it counts and shows its own combinations separately.
Where a filter leaves fewer than 30 independent cases, no figure is shown.
Switch the combination and see what it did
Pick what to do, on which currency, with which contract and when to start. Every combination was worked out beforehand from the records we hold, so the switches do not search for anything — the whole set is counted and the position of the one you are looking at is shown with it.
Tested
Four items.
A1 · Selling the weekend pair
Selling side
On Friday, sell one call and one put at the same price level, expiring Sunday, and hold them to settlement.
- How many independent cases
- 51 weekends on bitcoin, 90 on ether — weekends, not rows
- Period looked at
- 2024-01-11 → 2026-06-30
- What came out
- Before costs it kept 28.4% of the money taken in (t = 3.45). After the measured cost of dealing (12.6% one way) it kept 16% (t = 2.05).
- After counting how many things were tried
- It belongs to a family of 42 comparable cells. In that family, chance alone produces a best result of t = 3.03 on average, so 2.05 is below the line: the corrected probability is 0.163.
- How much fits
- The buying interest standing at that price level was $680, and the whole exchange trades about $635 (bitcoin) to $759 (ether) a day at that point on the board — roughly $4,000 to $12,000 a year before costs. A position that size is most of the day's flow at that point, and therefore the most predictable order on the board.
A2 · Buying the Monday pair
Buying side
On Saturday, buy one call and one put at the same price level, expiring Monday, and hold them to settlement.
- How many independent cases
- 113 one-day cases on ether; bitcoin has 30 and is flat
- Period looked at
- 2024-01-11 → 2026-06-30
- What came out
- On ether, before costs it gained 32.0% of the money paid (t = 2.74); at the most generous cost anyone proposed, 27.0% (t = 2.32). On bitcoin, nothing (t = 0.98).
- After counting how many things were tried
- Its own shuffled family has a median best result of t = 2.50. At 2.32 it does not reach the middle of its own noise.
A3 · “The far-out options expiring today are cheap” — closed
Measurement
Buy the options far from the current price on the day they expire, on the theory that they are systematically underpriced.
- How many independent cases
- 9 series × 3 ways of building it, over 56–69 days
- Period looked at
- 2026-06-29 → 2026-09-05
- What came out
- Built from the exchange's own reference figure it repeats on 9 of 9 series. Built from the middle of the two prices on screen it changes sign on 8 of 9, and the sample collapses by 96%. Built at what a buyer actually pays it is positive on 100% of the cells that have data.
- After counting how many things were tried
- No correction is charged: this was a named target checked on purpose, and it failed.
A4 · The time-of-day mispricing — real, large, and out of reach
Measurement
Deal at the hour of day when the price on the board is furthest from what the movement in that stretch of the day justifies.
- How many independent cases
- 1,666 stretches per hour of day, 4.5 years, 2 currencies
- Period looked at
- 2022-01 → 2026-06
- What came out
- The price on the board rises from 0.99 to 1.45 times the level the movement justifies (bitcoin), 1.00 to 1.39 (ether), moving the deal from 08:00 to 04:00. The two currencies agree.
- After counting how many things were tried
- 48 descriptive cells, declared in advance. The figure below is the one that is measured.
The reference line: what the plain version paid
Before any condition is applied, this is the same pair of options bought and sold with no filter at all, six hours before it expires, held to settlement. Every conditioned result above is meant to be read as a difference against this line, never on its own.
Accumulating — under 30 independent cases
Fewer than 30 independent cases so far. Each one carries the date the sample reaches that size at the current collection rate, or a dash where no date can be given.
| What is waiting | What we have | What is needed | When |
|---|---|---|---|
| A1 and A2 at a sample size that could actually settle them | 51 Sunday expiries on bitcoin | 218 at the size of the effect we measured | 2029 Q4 |
| Anything at all about a different kind of market | 56–69 days, all of it a very quiet stretch | one episode of the market moving hard, with the option prices being recorded | 2027 Q1 |
| The seven-day version of the pair | 9 settled Fridays | 15 before a line is drawn at all | 2026-10-16 |
| Two more option exchanges, and the only gold and silver options we hold | 3 days | about 210 days for the gold and silver side | 2027 Q2 |
| The same test on solana | 10 of the 15 needed, because the board is only two-sided on a minority of expiries | 15, and we will not promise a date | — |
Rejected — reason shown
Kept here so nobody starts them again, and so the absence of each one reads as a decision rather than an oversight.
| Idea | Why it is closed |
|---|---|
| Buying far-out options that expire the same day | The gap between the buying and the selling price is 54.5% (bitcoin) to 60.9% (ether) of the money involved, and the fee adds up to 12.5% on top. Roughly 40% is gone on the way in. |
| Selling the middle and buying the sides to cap the loss | The money taken in is $132 and getting in costs $108 of it. There is no price at which the difference being tested is visible above the cost. |
| The Saturday structure built from four contracts at once | Costed on all four legs, getting in takes $189 of a $558 credit against an expected $78 — the cost is 2.4 times the whole expected gain. |
| Anything that has to be closed before it expires | At the longer dates, both a buying and a selling price exist only 47–58% of the time. You cannot close a position on a board that carries a price half the time. |
| Only selling when the price of movement looks high | The bar set in advance was 1.4 points; getting in alone costs 1.55. And splitting the sample into thirds raises the statistical floor to 8.4–10.9 points — above the largest effect this repository has ever measured. The filter has to add more than the whole unfiltered effect just to pay for the sample it destroys. |
| Buying when the price of movement looks low | Measured, the cheap state is fair, not cheap: what the seller kept in the cheapest tenth is still +2.9% to +3.5% (bitcoin) and about zero (ether). The cheap state destroys the seller's edge; it does not create a buyer's one. The honest product of that finding is a rule to stand aside, and we label it as one. |
| Any rule that stops out, takes profit, or offsets the position while it is open | Every rule for getting out early that we tested lost to simply holding to expiry, and 22–35% of same-day contracts never print a price again, so getting out cannot even be reconstructed afterwards. There is no switch for this on the page, and there will not be. |
| Using the direction of the market as a condition | Direction is not learnable on this surface. Every result on this page is about how much things move, never about which way. |
Questions
Why is the line going down?
Because you pay the difference between the buying and the selling price on the way in, and then the contract turns into cash at expiry. It is not a forecast that it will keep going down; it is a record of what that difference cost over the cases shown.
Can I actually do this?
One of the two exchanges here has the better record and is not available to residents of Japan; we show it as a yardstick only. The one you could reach is also the one with the wider gap between the buying and the selling price. This site does not introduce accounts and does not give advice.
The switches let me try many combinations. Does that not make the numbers easier to fool myself with?
It would, if the set were open-ended. It is not: every combination the switches can reach is worked out in advance, the whole set is counted, and the position of the one on screen is printed with it. Trying them all changes nothing about that count. What the switches cannot do is add a condition that is not already in the count.
There are two and a half years of records. Why not use all of them?
That file records deals that happened, not prices anyone could have taken, and the two halves of the pair printed a median of two hours apart. When we require them to be at the same moment, the sample falls to 16 to 23 cases and the answer changes sign. A number whose sign depends on how strict you are is not a number, so we do not draw a line from it.
Where these numbers come from
The figures above are frozen on this date and not recalculated when you load the page: 2026-09-05.
The one thing that is rebuilt daily is the reference line and its table of cases. Everything before the first date on each line was never recorded and cannot be bought afterwards, so no line here starts earlier than its own first case.
Questions we are asked
What decides whether a figure here counts?
By how many variants were tried alongside it. Each row lists its family's size, the level that family requires, and the value that came out.
When you test many variants and report the best one, the best one looks good even when nothing works — so the bar has to rise with the number of variants tried. The strongest result here reaches a t-statistic of 3.449 before costs; after the measured cost of getting in and out it falls to 2.05, and the same family of 42 variants produces a best-of-42 value of about 3.03 from noise alone. It does not clear its own bar. This page shows that arithmetic rather than the headline, because the headline on its own would be the misleading number.
What is listed here?
Four tested items, each with what it does, how many independent cases, over what period, what came out, and the size of the family it belongs to.
Two are strategies with figures. Two are measurements of a stated idea: one is that very short-dated wings look cheap on mid prices across every series held, and reverse once the prices a buyer would actually pay are used; the other is that a difference between quoted and fair value grows from 1.00 to 1.45, while the gap between the buying and selling price grows faster over the same hours.
What period do these results cover?
One quiet stretch. Measured against five and a half years of history, the window these figures come from sits in the calmest tenth for volatility.
Bitcoin's implied volatility over this window sits at the 6th percentile of its own five-and-a-half-year record, and Ether's at the 12th. Selling volatility tends to look better in a calm stretch and worse when one ends, so a result measured here is not a result measured across regimes. That is the main reason the figures on this page are values measured over one period, and why each of them is shown with its family size and the level that family requires.