The Best and Worst Trades, Graded Twice

In October 2012, a few days before the season, Oklahoma City traded James Harden to Houston after the two sides couldn’t agree on an extension. Our Trade Lab, rerun with only what was known that day, gives Houston’s side +$115M: a decent deal. It returned +$823M, seven times as much, the fourth-best result of any trade old enough to grade. Nobody grading it that week could have known which number was coming.

Grading trades on the day is a whole industry. Kevin Pelton has handed out letter grades within hours of every big deal for years, and the internet hands out its own within minutes. Grading them years later is what fans do, and it has a name. The poker player Annie Duke calls judging a decision by how it turned out “resulting”: a good bet can lose and a bad one can win, and the outcome alone can’t tell you which you made.¹ So every trade here gets both grades, side by side.

The rules

The first grade is the process: what the Trade Lab says each side was worth on the day, using only what was known then, every player’s rating as it stood that morning, his contract, the picks, and the chance he re-signs. The second is the result: what each side actually got, every season it held what it traded for, in today’s money.²

How often do the two agree? More often than not. The sides the model liked on the day lost value one time in six; the ones it disliked came out ahead one time in five. Sort the trades into five groups by the grade on the day and the best fifth returned +$67M a side on average, the worst fifth −$117M.³

The best and worst on the day

The best process on record is my team’s. In July 2007 Boston traded five players and two first-round picks for Kevin Garnett, and the model has Boston’s side at +$802M on the day. It returned +$398M and a banner.⁴ Ten years later Boston traded Isaiah Thomas, Jae Crowder, Ante Zizic and Brooklyn’s first-round pick for Kyrie Irving, and that side is among the worst on both counts. I was thrilled at the time. (Sigh.)

The best and worst results

The best result belongs to Oklahoma City’s trade of Paul George to the Clippers for Shai Gilgeous-Alexander, Danilo Gallinari and a pile of picks: +$1.4B, on a deal the model graded slightly negative on the day. The model valued George, a star, above Shai, a promising guard with one season behind him.

Good bets that lost, bad bets that won

Graded positive on the day, lost value.

Graded negative or even on the day, came out ahead.

My team misses the second table by a hair. Boston’s February 2022 trade for Derrick White graded +$1.8M on the day, about even, and has returned +$628M, which would put it fourth on that list. I’d love to tell you the model saw it coming. It didn’t; it saw a fair trade, and a fair trade can still work out like that.

That’s the lesson of the whole page. Most of the time a good bet pays and a bad one doesn’t, but one time in six the good one loses, and nobody grading it later can tell the difference without the first grade. Every one of these trades is on the trades page with both. Each season I’ll add the trades that turn four years old, and we’ll see which bets paid.

Figures: every trade on the trades page, 1999-00 to 2025-26, graded by the Trade Lab; in today’s money, as of this build. Wins are on this model’s own scale and dollars at the open-market price of a win; the method explains both.

1. Annie Duke, Thinking in Bets: Making Smarter Decisions When You Don’t Have All the Facts (Portfolio, 2018). ↩
2. Results count only trades at least four seasons old, so most of the contracts involved have run their course: 2,042 team-sides. Sign-and-trades are left out, 125 of them: the player picked his club, and the trade was paperwork. The Trade Lab’s method has the rest. ↩
3. Across all 2,471 team-sides, the two grades are correlated at 0.62. Part of that is built in: both charge a club the same thing for what it sent, the forecast on the day, so they differ only in how what it got turned out. ↩
4. The deal’s other side, Minnesota’s, is the worst process on record and the worst result. ↩