The Premium You Sold Was a Direction You Took

There is a trade every options course teaches early. Sell a put spread below the market, collect a credit, let time decay do the work. It wins most of the time, which is the part everyone remembers.

I replayed it. Not on a model — on ten years of real option chains, entry and exit priced the way a fill actually happens, commissions deducted. What came out was not what the win rate advertises.

The setup

One trade per week, no overlapping positions. Short leg at a target delta, long leg ten strikes further out, expiry between 30 and 45 days. Held until the position could be bought back for a set fraction of the credit, or until expiry.

Every leg crosses at mid ±0.5% on the way in and again on the way out, so the round trip is charged, not just the entry. Commissions at $0.65 per contract per leg, both directions. The take-profit is booked at the threshold, not at the close of the day it was crossed — see below for why that distinction is not cosmetic.

Two deltas, three widths, three take-profit levels, on SPY, GLD and TLT. About 2,000 trades in total.

The result that matters

Here is the same configuration — ten strikes wide, short leg at 0.30 delta, bought back at 75% of the credit — run on both sides of three underlyings. Return is expressed against the capital at risk, net of costs:

SPY and GLD both rose over the decade. Selling puts on them made money. Selling calls on them lost roughly three times what the puts made.

TLT, which went sideways over the same window, lost money on both sides — modestly, but on both. Every one of the eight configurations I ran on it landed between −1.1% and +0.7%.

That is not a premium being harvested. That is a direction being taken.

A credit vertical sold below a rising market is a long position wearing a different name. The credit received is real, the theta decay is real, and none of it explains the result — the underlying going the right way explains the result. Remove the drift, as TLT does, and what remains is the friction.

The win rate points the wrong way

The second finding is uncomfortable in a different way. On SPY alone I ran eighteen configurations — three take-profit levels, three widths, two deltas. Sorted by net return:

The relationship holds across all eighteen without exception. The two highest win rates in the set — 96.9% and 96.1% — return +0.3% and +0.1%. The best returns come with the lowest win rates in the table.

This is not a paradox. A high win rate on a defined-risk credit trade is bought by taking profit early and small, which caps the winners while leaving the losers untouched. You win more often and you win less.

Six of the eighteen configurations are outright negative.

The exit convention costs a point

This is the part I got wrong first time, and it is worth more attention than a footnote.

A backtest that books the take-profit at the closing price of the day the threshold was crossed is measuring something that no order type delivers. On a position that moves during the session, the close can overshoot the limit by a wide margin, and the backtest quietly banks that overshoot. A resting GTC fills at the threshold. Nothing more.

Measured across four configurations, the difference is 0.6 to 1.6 points of return — and it flips one line from +1.5% to −0.1%.

The effect scales with how far the position typically travels past its target in a single session. On credit verticals with a 50 to 75% take-profit, the crossing is a slow drift and the overshoot is modest. On a long strangle with a 10% target, where the position routinely moves more in a day than the distance it needs to cover, the overshoot can be several times the target itself. Any backtest of that shape which books the observed close rather than the threshold is reporting a number no resting order will ever produce.

If you run your own backtests, this is the first thing I would check.

Commissions decide which structures are tradable

$0.65 per contract per leg, round trip, is $2.60 on a one-lot vertical. That sounds like nothing.

On a three-strike SPY spread risking about $250, it is a full percentage point. On a twenty-strike spread risking about $1,770, the same $2.60 is 0.14 of a point and changes nothing. Five of the six narrow configurations I tested are negative once it is deducted.

The cost is fixed. The denominator is not. Narrow spreads are where a small edge gets eaten, and narrow spreads are what most screeners surface, because they look cheap.

One thing I expected and got wrong: widening does not improve the credit you collect relative to the width. It gets worse. At 0.20 delta the credit went from 11.3% of the width at three strikes down to 6.3% at twenty. Pushing the long leg further out adds risk against a leg whose value is already near zero. A twenty-wide spread collecting 6% of its width is a naked put with a lottery ticket attached.

The only lever that genuinely improved the ratio was the short delta: 0.30 collected 60 to 80% more premium than 0.20 at the same width.

What this does not say

It does not say credit verticals do not work. It says that on these three underlyings, over this particular decade, the result was governed by the direction of the move rather than by the premium collected.

Ten years is one sample. SPY and GLD rose; a different decade would flip the table. TLT's result is arguably the most informative line here precisely because it removes the drift and leaves nothing behind.

Exits are evaluated on daily marks, so a position that touched its target intraday and closed back below is counted as a loss. For this family the effect is nil — the win rate computed with and without an intraday-crossing proxy was identical across 2,060 trades. Take-profits at 50 to 75% of a credit are reached by drift, not by a spike. That is emphatically not true of long strangles at a 10% target.

And the fills are modelled at mid ±0.5%, not at the far side of the spread. Real execution is worse, and I am measuring that separately on live positions.

What I take from it

Three things I now check before any premium-selling structure.

Whether the same trade works on the other side. If selling puts on something pays and selling calls on it loses, I am looking at a directional position, and I should size it and hedge it as one.

Whether the take-profit is worth reaching. A 25% target on a thin credit is a high win rate and almost no money, and the commission is waiting at the end of it either way.

And what price the backtest assumed I got. That one cost me a point, and I only found it because someone asked.

The win rate is the number that sells the trade. It is almost never the number that decides it.

Methodology

AlphaVantage historical option chains, September 2016 to August 2026. Weekly non-overlapping entries, one canonical structure per configuration rather than an optimised selection. Entry at mid ±0.5% per leg; the take-profit is booked at the threshold, as a resting limit order fills. Commissions at $0.65 per contract per leg round trip. Returns expressed against capital at risk.