WHL-001 — The Wheel

This page explains what the Wheel Lab does, what it assumes, where the numbers come from, and where it should not be trusted. Read it before you draw a conclusion from the results.

What this study found

Over thirty years at 30-delta on SPY, the option programme made money — a median of $1.64 million net of everything paid back at settlement, plus $2.92 million of interest on idle cash. About $4.56 million of real income. And the wheel still finished $6.4 million behind a portfolio that bought the index and did nothing: $10.51 million against $17.22 million, a gap of 1.81 percentage points a year, ahead on only 16.1% of paths.

The premium is not the problem. The default state is. A strategy that starts in cash and returns to cash was in the market 33% of the time, and in an asset compounding near 10% a year, being out of the market is the most expensive thing you can do.

A second, independent engine then answered the same question by market regime and showed when that bill comes due. The wheel gives up 29 points over five years in a strong bull market — and beats buy-and-hold by 16 points in a bear market, on 96.5% of paths.

The strategy, stated exactly

One position, two states, one option written every month.

• In cash — a one-month put is written at the chosen delta, fully cash-secured on the whole balance. No leverage, ever. The cash earns the risk-free rate. If the put finishes in the money the shares are bought at the strike; under §1234(a) the premium reduces the cost basis rather than being taxed as income. Any cash left over is invested at market.
• In shares — a one-month call is written at the chosen delta on 100% of the position, dividends are collected. If the call finishes in the money the shares are sold at the strike and the position returns to cash.

The benchmark buys the index on day one and does nothing. Both portfolios see identical return paths, so every comparison is paired: each result is one path measured against itself.

Where the option prices come from — the part that decides everything

This is the assumption that makes or breaks a wheel study, so it was measured rather than assumed. This study is more sensitive to it than most, because it sells both wings.

The at-the-money level. Implied volatility is generated from trailing realised volatility as IVATM = 0.5529 × RV + 7.23pts. That relationship was fitted to 3,722 daily observations of real implied volatility from 2011 to 2026 (r² 0.645), taken from Interactive Brokers' historical option-implied-volatility series. It is not a guess and it is not derived from VIX — VIX is a variance-swap rate and sits about 20% above at-the-money volatility.

The skew, with separate wings. Volatility varies by strike as IV = IVATM + β × ln(K/S), and the two wings are not the same:

Two independent chains agree to 0.01, with r² of 1.00 on the put wing. The real asymmetry is 2.1× — the put you sell is meaningfully dearer than the call you sell at the same delta. Across the run that shows up directly in the prices received: the average put was sold at 16.83% implied vol against an at-the-money level of 14.68%, a two-point premium in the seller's favour, while the average call was sold at 17.56% against an at-the-money 18.91% — a 1.35-point discount against the seller.

Using one number for both wings is the single most common error in wheel arithmetic, and it is the error that makes the strategy look profitable. It is also the mistake this project had made: the corrected put wing reversed the conclusion of this study before it was published. The lab exposes both wings as controls; set them equal and watch what happens.

Premium is reduced 3% for bid–ask before it is credited.

Validation — does the engine reproduce reality?

Cboe publishes indices that have run these exact strategies with real option prices for decades. Each is one realised path, so the test is whether the real outcome falls inside the distribution the model produces, not whether it equals the median. Benchmarked against SPY's adjusted close — a total-return series, because the Cboe indices are total-return and comparing them to a price-only S&P is the error that flatters every casual buy-write comparison.

All three are consistent with the model, and the strongest of them lands on the leg that drives the conclusion: the put side is close to exact. The two buy-writes land in the lower tail, which hints the call wing may still be slightly generous — but they are the same underlying over the same window, so they are effectively one observation rather than two, and the bootstrap resamples that same period, so they are not independent of the model either. One correlated observation at the 13th–19th percentile cannot be told apart from luck.

Note the direction of that residual error. A call wing that is too generous makes the covered-call row of the bracket below look better than it should — which, if anything, understates how badly the wheel compares to simply holding the shares and writing calls.

You can run the BXMD comparison yourself: set Strategy to "covered calls only" and Call delta to 0.30.

A second engine, and what it adds

Everything above comes from one model, and a model can be wrong in ways its own diagnostics cannot see. The same question was put to the research engine behind the full app, which shares no code with the engine on this page. It does not resample history at all: it generates prices from six synthetic market regimes, each with its own drift, volatility and jump behaviour, and prices options off a full Black–Scholes chain with a term structure and a smile, calibrated to twenty years of Interactive Brokers implied volatility. Different price process, different pricing surface, different portfolio mechanics, same question. 2,000 paths per cell, five-year horizon, 30-delta, paid the measured real-world premium.

It agrees the wheel loses in rising markets. It disagrees about that being the whole story. Two engines sharing no code reaching the same sign is the strongest evidence in this project. What the second one adds is the case-by-case breakdown the bootstrap averages away: the bootstrap on this page draws from 2001–2026, and that quarter-century was mostly a bull market, so it reports the blend. The regime engine shows what the blend is made of.

The premium is not what decides it. Real option sellers are paid more than fair value — that gap is the variance risk premium, and it is the entire reason selling options is a business. Measured over twenty years, at-the-money implied volatility averaged 16.67% against 15.56% of volatility that subsequently arrived: 1.12 volatility points. Re-running every regime while paying the seller nothing, the measured amount, and generous multiples of it moves the strong-bull result from −32.6 to −29.4 — three points on a thirty-point loss. Every regime improves by roughly the same two to three points per volatility point. It is a level shift, not a rescue. One threshold does get crossed: in typical and choppy markets a wheel paid 2 points is roughly break-even, and at 3 points it edges ahead. It is specifically rising markets where no realistic premium saves it.

The sign is delta-insensitive on this engine too. Across 0.16, 0.30 and 0.45 the ordering of regimes never changes. Delta only amplifies the defensive payoff: in a bear market, moving from 0.16 to 0.45 takes the edge from +11.6 to +17.0 points while median drawdown falls from 18.1% to 14.7%.

What is deliberately not modelled — and why that matters here

Every intra-month management rule works through one channel: it changes how much of the time the portfolio holds shares rather than cash. Closing a put early means fewer assignments and more time in cash; closing a call early means fewer call-aways and more time in shares. The two non-wheel settings on this page pin those extremes — never assigned, and never called away — so they bracket what any management rule could achieve, without modelling a single rule.

That bracket is wider than the difference the study is measuring, so a management rule could certainly move the answer. But the rows are perfectly ordered in time in the market, and the ordering is the finding. Every step toward being invested is a step toward buy-and-hold. Nothing about assignment mechanics, strike selection or roll discipline appears in this ranking, because none of it matters next to whether you own the asset.

This is the opposite of what the wheel community argues about. The debate is over the middle step — take assignment, roll out and down, close at 50%. The evidence here says the middle step is close to irrelevant, and that the ordering runs the wrong way for the popular advice: a disciplined roller, someone who never takes assignment, is moving toward the worst row in that table, not fixing the strategy.

Other limitations, in order of how much they could matter:

• The result depends on the cash rate, which is not an options assumption at all. The wheel sits in cash most of the time, and the rate is held constant while returns are bootstrapped — which severs the historical link between the rate environment and the market environment. It is a control; move it.
• The answer is stated at a delta, not in general. The wheel loses at every strike tested — −2.18%/yr at 16-delta, −1.81% at 30-delta, −2.25% at 45-delta. Thirty-delta is the least bad of the three, which is mildly reassuring for the people already running it and rescues nothing. Far out-of-the-money puts collect too little premium to pay for sitting in cash; closer-to-the-money puts collect more and get assigned into a falling market more often.
• European settlement. SPY options are American and a deep in-the-money put can be assigned early. Not modelled.
• One position, no laddering. A real wheel on a large account staggers expiries. That smooths outcomes without changing the expectation.
• The bootstrap cannot invent a crash worse than the 2001–2026 sample contains. It does contain 2008 and 2020, so this is mild — but it caps how much credit put-selling can be given for surviving the downside, and it is exactly the limitation the regime engine's bear and crash rows exist to probe.
• Bid–ask is a proportional haircut on premium. Real spreads are closer to fixed dollars per contract. That is fair at monthly tenors and would understate costs at weekly ones.
• SPY only. This says nothing about wheeling single names, which is how most people actually run it, and where assignment risk is a different animal entirely.
• Average cost basis rather than lot-level FIFO, for tractability. In an appreciating pool it realises slightly less gain than FIFO would, so it understates the tax drag.

What the wheel actually buys you

It would be dishonest to stop at the loss. The wheel's median maximum drawdown was 32%, against 42% for buy-and-hold — and the pure put-writing programme, the worst performer on wealth, had the best drawdown of all at 24%. A 42% decline is where real people abandon their plan and sell at the bottom. A 32% decline is unpleasant and survivable.

That is the same result as every option-selling study in this project: it is a risk-transfer strategy, not an income strategy. You are giving up return to reduce variance. The wheel's version of that trade is unusually expensive — 1.81 percentage points a year, where a plain covered call on the same pricing costs 0.33. The questions worth asking are not about premium. They are how much of a rising market you are willing to give up, and whether you actually believe you will be in the regime where it pays.

Accounts and tax

There is no separate traditional-IRA setting because in accumulation there are no distributions and no required minimum distributions, so a traditional IRA and a Roth behave identically. The difference is entirely on the way out, which is the subject of RET-CC-002. On the taxable setting, the working-age standard deduction applies — not the over-65 one used by the two retirement studies.

Reproducing this

Everything in the lab on this page runs in your browser; nothing is precomputed. The seed is a control, so any run here is reproducible exactly. The market data is real SPY daily adjusted closes. The calibration parameters above were measured from Cboe's public option chains and index history and from Interactive Brokers' historical implied-volatility series. The regime and premium-sweep figures come from the research engine behind the full app and are quoted rather than run here.