Monthly Returns From Selling Options: Why 2–3% of NAV Is the Realistic Ceiling

If you are selling premium on a diversified wheel book you intend to run for years, 1–3% of net asset value per month is the realistic band, and 2–3% is the top of it. Higher figures are easy to quote and the arithmetic below shows exactly how to reach them. What makes 2–3% the practical ceiling is what happens either side of it. Up to about 3% of NAV a month, each extra point of premium costs you almost no extra risk. Past that, the same extra point costs roughly ten times as much. Draw income against risk and the line runs cheaply along, then bends hard. That bend is what engineers call a knee, and 2–3% is where it sits.

The name is as literal as it sounds. A line that runs straight and then turns sharply looks like a leg with a joint in it, and engineers have called that point the knee of the curve for a very long time. Materials science gave us the useful version: stretch a metal and it lengthens neatly in proportion to the pull, right up until the knee, after which it keeps deforming for good. Anyone who has bent a paperclip has felt it. A little pressure and it springs back; past a certain point you own a differently shaped paperclip. Premium selling has the same shape, and the same second half. Below the knee you are being paid for risk the book can absorb and recover from. Above it you are bending the book into a shape it does not come back from.

It is also a much higher bar than it sounds. A sustained 2% a month would have beaten the Nasdaq's last decade.

Every number here comes from the same pricing and stress engines behind levelbox, at 30 days to expiry with a 4.2% risk-free rate, plus implied volatility we recorded across our tracked universe on 13 August 2026 and five years of price history to the same date.

Where the premium comes from

Premium is bounded by implied volatility. So the ceiling can be calculated.

Monthly premium as a share of collateral, for a 30-day cash-secured put:

30-day ATM IV	0.10 delta	0.20 delta	0.30 delta
25%	0.38%	0.88%	1.48%
39% (universe median)	0.64%	1.45%	2.43%
60%	1.08%	2.43%	4.02%
76% (universe 90th pct)	1.48%	3.27%	5.37%
100%	2.17%	4.71%	7.66%

Turned around, the implied volatility you need to hit a target:

Target, % of collateral per month	at 0.10 delta	at 0.20 delta	at 0.30 delta
1.0%	56.3% IV	28.0% IV	17.3% IV
2.0%	94.5% IV	51.1% IV	32.8% IV
3.0%	124.2% IV	71.0% IV	46.8% IV

A low-delta book cannot get near 3% a month: at 0.10 delta you would need 124% implied volatility, which on 13 August described exactly two names in our universe, BLNK at 174% and SOXL at 125%.

But 3% of collateral per month is not hard to quote at all at 0.30 delta on a 47%-IV name, and 128 of our 312 tracked names carried implied volatility at or above 45%. Roughly 40% of the universe qualifies. The arithmetic alone does not cap you anywhere near 3%.

The high-IV route works, and better than most people admit

You can earn a great deal more than 3% a month by concentrating on high-implied-volatility underlyings. This is not a mirage.

A 0.30-delta 30-day put on a 100%-IV name pays 7.66% of collateral per month, more than three times what the identical structure pays on a median-volatility name. Our universe had a 90th percentile of 75.6% and a maximum of 173.6%, so the material genuinely exists.

It also holds up on a measure we expected it to fail. Here is what a uniform fall costs, expressed as months of premium needed to earn it back, with implied volatility expanding 1.5× into the fall:

Book	Premium/month	Cost of a −20% move	Months to earn back
0.10 delta, 39% IV	0.64%	10.8% of collateral	16.8
0.30 delta, 39% IV	2.43%	14.9%	6.1
0.30 delta, 76% IV	5.37%	15.0%	2.8
0.30 delta, 100% IV	7.66%	16.0%	2.1

The mark-to-market hole is only modestly deeper on the high-premium book while the income repairing it is many times larger. On this measure the conservative book is the weak one. We report it because it cuts against our own conclusion.

So neither the premium arithmetic nor the payback speed sets the ceiling. Three other things do.

First: 2–3% a month is already an extraordinary bar

This is the part that gets lost when the discussion stays in percent-per-month. Compounded:

Monthly	Annual, compounded
0.75%	9.4%
1.00%	12.7%
1.50%	19.6%
2.00%	26.8%
3.00%	42.6%
5.00%	79.6%

Now put that beside what has actually been achieved. These are total returns with distributions reinvested, measured to 13 August 2026 over each vehicle's available history:

	Years	CAGR	Per month	Max drawdown
Nasdaq 100 (QQQ)	10.0	20.9%	1.59%	−35.1%
JPM Nasdaq Equity Premium Income (JEPQ)	4.3	16.2%	1.26%	−20.1%
S&P 500 (SPY)	10.0	15.4%	1.20%	−33.7%
Berkshire Hathaway (BRK-B)	10.0	13.1%	1.03%	−29.6%
JPM Equity Premium Income (JEPI)	6.2	11.5%	0.91%	−13.7%
Global X Nasdaq covered call (QYLD)	10.0	9.8%	0.78%	−24.8%
Global X S&P 500 covered call (XYLD)	10.0	8.3%	0.66%	−33.5%
Simplify short vol (SVOL)	5.3	8.2%	0.66%	−33.5%
Global X Russell 2000 covered call (RYLD)	7.3	6.0%	0.49%	−41.5%

Not one professionally managed, permanently staffed premium-selling vehicle has delivered 1.3% a month over its life. The same reading deflates the double-digit coupon on a structured note, which is the identical premium sold in a wrapper. The best of them lands at 1.26%. A sustained 2% a month would have beaten the Nasdaq over a decade that included the largest technology bull market on record. A sustained 3% would have beaten every line on that table by more than double.

Note also the trap in the middle of that group. QYLD currently distributes about 12.5% a year, which reads as 1.04% a month of "income". Its ten-year total return is 9.76% a year. The distribution is partly a return of your own capital, and the gap between the two numbers is exactly the gap this whole article is about.

So when the answer to "how much per month?" comes back as 2–3%, that is not a modest or defensive figure. It describes performance that would place a book at the top of its category.

Second: where a put-selling return can actually come from

Gross premium is not expected return. A short put's expected profit decomposes into three sources, and everything else in the quoted premium is risk transfer with no expected profit attached:

1. Cash yield on the collateral, since you hold the cash while short the put. About 0.34% a month at current rates.
2. The variance risk premium, the amount by which implied volatility exceeds the volatility that subsequently gets realised.
3. Delta exposure to the equity risk premium, since a short put is a long position in the underlying.

Running that decomposition, with the variance risk premium set at a conservative 3 implied-vol points and the equity risk premium at 5.5% a year over cash:

	Gross premium	Cash	VRP	Delta × ERP	Expected
0.10 delta, 39% IV	0.64%	0.34%	0.16%	0.05%	0.56%
0.30 delta, 39% IV	2.43%	0.34%	0.32%	0.14%	0.81%
0.30 delta, 76% IV	5.37%	0.34%	0.34%	0.15%	0.84%
0.30 delta, 100% IV	7.66%	0.34%	0.36%	0.16%	0.86%

Down that table the quoted premium rises twelvefold. Expected return rises by half.

We searched the whole space, every delta from 0.05 to 0.95 against every implied volatility from 20% to 170%, for the best expected monthly return available to a fully cash-secured book. The answer is 1.01% a month, at 170% implied volatility and 0.59 delta, where the quoted gross premium is 41% a month. Almost all of that 41% is compensation for risk you are genuinely bearing.

That single number explains the benchmark table above better than anything else. It is also assumption-dependent in one specific way: if the variance risk premium scales with the volatility level instead of holding at a fixed number of vol points, the ceiling rises to about 2.9% a month. The published work generally finds the variance risk premium is much larger and more reliable on index options than on single names, with much of the index premium attributable to correlation risk that single-name sellers are not paid for. That points toward the conservative column. It is not settled, we have not measured it on our own data, and anyone quoting you a confident number here is describing their priors.

Third: the call leg, which changes the arithmetic

Everything above is the put side. A wheel book spends a large share of its life holding shares and selling calls against them, and that leg behaves differently.

A covered call ties up more capital for the same delta. A cash-secured put's denominator is the strike, which sits below spot. A covered call's denominator is the full share value. The gap widens with volatility:

Underlying IV	CSP, %/mo on collateral	Covered call, %/mo on share value	Call as share of put
25%	1.17%	1.07%	0.92
36%	1.75%	1.51%	0.86
51%	2.63%	2.07%	0.79
76%	4.28%	2.96%	0.69

(0.25 delta both sides.)

Roughly half the book ends up in shares. Assignment pushes capital from cash into shares; calls being exercised pushes it back. At the rates implied by these deltas the steady state settles near an even split, which drags the blended rate below the put-only quote:

Underlying IV	Delta	Pure CSP	Blended with the call leg	Drag
36%	0.25	1.75%	1.62%	−7.6%
36%	0.35	2.73%	2.54%	−7.0%
51%	0.25	2.63%	2.32%	−11.5%
51%	0.35	4.06%	3.62%	−10.8%

So a book quoting 2.7% on the put side realistically runs at about 2.5% once the call leg is doing its share of the work, and more of the drag applies the higher the volatility.

Skew ran against the textbook here. Puts are supposed to carry higher implied volatility than calls at matched delta, which is what pays put sellers better. Across 210 matched pairs from 61 live chains cached between 21 July and 14 August 2026, at deltas from 0.10 to 0.25, we found the reverse: median call IV of 57.2% against median put IV of 54.7%, with calls richer in 68% of pairs. That is call skew, and it is characteristic of high-momentum names where speculative upside demand bids the calls. It is a biased sample: those chains are cached for names on covered-call watchlists, so they are not a random draw. Treat it as a caution against assuming put skew is always in your favour, and not as a general result.

The truncation matters most where the premium is fattest. A 0.25-delta call on a 76%-IV name caps your month at +22.3% including the premium. That is the same population where five-year returns run from +3900% to −99%, and where the group's positive mean is produced by a handful of enormous winners. Selling puts caps you at the premium on the way up; selling calls on the shares you were assigned caps you again. For high-volatility names the wheel is doubly mismatched to the return distribution that makes them work as a group.

Where the leaks are, in order

Putting the pieces together, four things sit between a premium quote and a monthly NAV return.

Collateral is not NAV. Cash buffers against margin, capital idle between expiries, and assigned shares each take a slice. This is the yield-on-capital distinction one level up. A portfolio question needs a portfolio denominator.

What you book is not what you keep. Positions get bought back, rolled down and out, and assigned shares get sold below cost. We measure net premium on a monthly clock for exactly this reason.

Assignment stops the quoted rate from repeating. At the universe median 39% implied volatility:

Target delta	P(finishes ITM), per cycle	Of 20 positions, assigned/month	One position surviving 6 months
0.10	12.1%	~2.4	46.1%
0.20	23.3%	~4.7	20.4%
0.30	34.0%	~6.8	8.3%
0.40	44.4%	~8.9	3.0%

These are probabilities of finishing in the money, which run above the delta figure; delta is a useful approximation for that probability rather than the thing itself. For a wheeler this is not a failure state, since assignment is the mechanism. It does mean the quoted rate describes a book you stop having within months.

At the top of the IV distribution you cannot build a book at all. On 13 August, of 312 tracked names: 211 at or above 30% IV, 128 above 45%, 64 above 60%, and 10 above 95%. The ten were BLNK, SOXL, LUNR, INO, CRDO, OUST, MBLY, NBIS, EVGO and AEVA.

We pulled five years of daily closes for the 308 names with sufficient history (16 August 2021 to 13 August 2026). NBIS drops out of the top bucket, leaving nine, having not been listed long enough:

30-day ATM IV	Names	Mean 5y return	Median 5y return	Share negative	Share that lost >50%
<30%	101	+90%	+66%	8.9%	2.0%
30–45%	83	+112%	+58%	22.9%	7.2%
45–60%	64	+139%	+25%	39.1%	20.3%
60–95%	51	+452%	+283%	25.5%	13.7%
≥95%	9	+230%	−50%	66.7%	44.4%

The 60–95% row will surprise people and we are not explaining it away. Over this window those names did enormously well: CLS +3900%, STRL +2343%, MU +1273%. High implied volatility recently has often meant a name in the middle of an AI or semiconductor re-rating.

The bottom row is different in kind. Its +230% mean is produced entirely by CRDO at +2183%. Remove that one name and the remaining eight average −14%; remove the top three and it is −74%. The median is −50% and two-thirds are down.

Three caveats, all load-bearing. This is our tracked universe as it stands today, so delisted and dropped names are absent and survivorship bias flatters every row, most of all the high-IV ones. Five years from August 2021 is a single window. And nine names is too small a sample to assert a rule from on its own; we report it because it is the group the premium arithmetic points you toward, and because the direction is consistent with the 45–60% row, which has 64 names and a median of +25%.

So what does constraining yourself to 2–3% actually buy?

We found this the most useful question of the three, and it has a sharp answer.

We built 1,080 candidate books from the real universe, varying the volatility-bucket mix, target delta from 0.06 to 0.44, and deployment from 1.0× to 2.0× of NAV, then priced each through the stress engine. Taking the lowest modelled crash loss available at each level of gross premium gives an efficient frontier, which is just the best deal on offer at each level of income:

Gross premium, %/mo of NAV	Min modelled crash loss	Extra crash loss per +1% of premium
0.50–0.75%	34.8% of NAV	—
1.00–1.25%	35.9%	2.1 pts
1.50–1.75%	36.8%	1.5 pts
2.00–2.25%	37.4%	1.2 pts
2.25–2.50%	37.5%	0.6 pts
2.50–2.75%	37.8%	1.0 pts
2.75–3.00%	43.3%	22.1 pts
3.00–3.25%	46.1%	11.3 pts
3.50–3.75%	53.0%	15.7 pts
4.25–4.50%	62.6%	20.3 pts
5.00–5.25%	73.1%	25.4 pts

Read the right-hand column. From 0.5% up to 2.75% of NAV a month, each additional percentage point of premium costs between 0.6 and 2.5 points of NAV in modelled crash loss. Premium in that region is close to free. At 2.75% the marginal cost jumps to 22 points and never comes back down.

That bend is the knee, and it is what "2–3%" is really describing. Below it you are collecting premium the risk model barely charges you for. Above it every additional point costs about ten times as much. Staying inside the band buys you all of the cheap income and stops right where it stops being cheap.

It buys three concrete things:

• You stay unlevered. Every efficient book at or below the knee runs 1.0× collateral to NAV. Leverage only starts appearing in the efficient set above it.
• You stay in names that fall less. Our stress engine scales each name's move by its implied volatility relative to the index. At the −25% index rung that means roughly −40% for a 26%-IV name, −57% at 36%, −79% at 51%, and −97% at 76%. The band keeps you in the first two.
• You keep choice. 101 names below 30% IV and 83 between 30% and 45%, against 10 above 95%.

This also cuts against how we had been framing our own optimizer. Allocating a book across several budgets is a knapsack problem, and a knapsack maximises value subject to a capacity, which assumes the capacity is worth consuming. The knee says otherwise. Given budget with room left, that allocator keeps buying premium at ten times the price and returns a feasible, constraint-satisfying plan, because by its own definition that is what it is. The objective never asks whether the last of the budget was worth spending. We have since made it report where the knee falls on your own frontier.

A necessary caveat on those crash figures: they are deliberately conservative, assume correlation of 1.0 to the index, and expand volatility into the drawdown. They rank books against each other and they do not forecast your account. The ordering is the signal. The level is not a prediction.

The composition that holds 2–3% together

The same 1,080-book sweep answers the constructive question, and the answer is counterintuitive.

The lowest-risk books that gross 2–3% of NAV are unlevered, hold roughly 25 names, sit in the calmer half of the universe, and get there on delta rather than on volatility. The efficient frontier at every level up to the knee is built from names in the sub-30% IV bucket carried at 0.30 to 0.44 delta, not from high-IV names carried at low delta.

That is worth stating plainly, because the instinct runs the other way. Selling 0.10-delta puts on 76%-IV names feels conservative, since the strike is far away. It is not: it produces the same premium as a 0.35-delta put on a 26%-IV name while the stress model marks its crash loss far higher, because the strike distance is bought with volatility that also shows up in the drawdown.

Concretely, a book in the band looks roughly like:

• Deployment: 1.0× of NAV in collateral. No margin.
• Names: about 25, so no single assignment is more than 4% of the book.
• Volatility: predominantly the sub-30% and 30–45% buckets, which is 184 of our 312 names.
• Delta: 0.30–0.40 on the put side, which is where the premium comes from.
• Assignment budget: expect 8 to 11 of 25 positions assigned per month at that delta, so the wheel's share leg needs planning before it arrives.
• Call leg: on the assigned half, at a delta you would genuinely accept being called away at, remembering the 7–13% blended drag and the upside cap.

Expected net return on that book, under the conservative decomposition, is about 0.9% a month. Under the proportional variance-risk-premium assumption it is about 1.1%. The 2–3% is what you book in premium; the range above is what the model says you should expect to keep before the market does anything in particular.

For calibration in dollars, on a round $2m book the 2–3% band is $40,000 to $60,000 a month in gross premium. A real book of roughly that size that we track closely, already deployed past 100% of its cash, realised about 1.1% of NAV in net premium in July 2026, a little over $20,000, in a market that was not stressed. Against the benchmark table above, 1.1% a month is 14.2% a year, which would have beaten every covered-call fund on that list and Berkshire besides. Getting from there toward 2% is a capital-efficiency problem, and finding fatter premiums will not solve it. That is most of what our premium optimizer exists to work on.

What to do with this

Quote yourself a NAV number, monthly, net, including the call leg. Anything else flatters. An annualised yield on collateral before losses cannot go down in a bad month, which makes it useless in exactly the months you need it.

Ask what your next point of premium costs, not just what you are earning. The useful question is not "am I at 3%?" but "am I still on the cheap side of the knee?" Once each extra point costs ten times what the last one did, you have changed strategy whether or not you meant to.

Treat a high quoted yield as a description of the risk. The premium is the market's price for a distribution it can see as well as you can.

Reach for delta before reaching for volatility. They produce the same premium and the stress model charges very differently for them.

If you do run the high-IV route, run it deliberately. It works, the payback table is real, and concentration is the price. Size it as the concentrated equity position it becomes within six months.

Separate the two ways of raising NAV yield. Selling closer to the money raises premium and raises assignment. Portfolio margin raises yield by shrinking the denominator, and it is leverage. They fail differently and should be budgeted separately.

---

Analytical and educational tooling, not investment advice. Premium and probability figures are Black-Scholes model outputs at a 4.2% risk-free rate with no dividend yield; probabilities of finishing in the money are risk-neutral, a price-implied quantity rather than a forecast. Crash figures come from our own conservative TIMS-lite margin model with correlation of 1.0 and volatility expanding into the drawdown, built for ranking rather than forecasting, and are not your broker's numbers. Expected-return decompositions depend on an assumed variance risk premium and equity risk premium, both stated inline and neither measured on our own data. Implied volatility figures are our 30-day ATM snapshot for 13 August 2026 across 312 tracked names, which is not a market-wide sample; the skew measurement uses 61 cached chains selected for covered-call coverage. Return figures are five years of adjusted daily closes to 13 August 2026 and carry the survivorship bias described above. Fund returns are total returns with distributions reinvested over each vehicle's available history, which differs by fund. Options involve the risk of assignment and of loss, and past distributions do not constrain future ones. Work the numbers on your own book and decide for yourself.