Found money

The cap went from $95.5M to $104.0M this summer, a rise of 8.9 percent, and every contract conversation in July started from the same sum. A player worth seven percent of the old cap is worth seven percent of the new one. Scale the number, sign the deal. It is tidy, it is how the industry prices players, and for the open market it is wrong, in a direction that costs real money.

The percent-of-cap rule assumes the two sides of the market move together. They do not. When the ceiling rises, all 32 teams receive new room on the same morning. Across the league that is 272 million dollars of fresh spending authority in 2026, on top of the 240 million that arrived the summer before. The supply those dollars chase is fixed by contracts signed years earlier: the free-agent class is whoever happens to expire this year, and nothing about the cap makes it bigger or better. More money, same players. Economics has one thing to say about that, and it says it in the price.

This piece tests the claim. We priced every unrestricted free agent who signed in the 12 summers from 2015 to 2026 against what he had just done on the ice, in percent of the cap, and asked whether the price of the same player stayed flat or moved with the size of the jump. It moved. The summer that followed the first big increase did not show it; the summer that followed the second did, and it showed up where the theory says it should: in the middle of the roster, among players who changed teams, not among the stars.

The data are Spotrac’s lists of unrestricted free agents who signed in each summer, 1941 skater contracts with a dollar figure attached. We joined each to the player’s previous NHL season in our own tables, which gave us 1577 with a season to price him on, and kept the 954 who had played at least half the schedule, so that the price is for an established NHL player rather than for a two-way deal signed in the dark.

The price of a contract is its average annual value as a percentage of the cap for the season it begins in, the convention the industry and the public modellers use, Evolving-Hockey’s contract projections among them. A hedonic model then explains that price with the things a general manager can see on July 1: points per game over the previous two seasons, primary assists, individual expected goals, minutes per game, five-on-five expected-goal share, Net Rating, how much of the schedule the player dressed for, whether he is a defenceman, whether he re-signed with his own team, and his age, with a curve. It explains 63 percent of the variation in the log of the price, which is about what you would hope: a player’s contract is mostly his record and his age, and the rest is the market.

The rest is what we are after. The model carries one extra term for each summer, and that term is the price level: how much more or less a skater of identical quality and age cost that July than in the five summers from 2015–19, which we take as the baseline. If prices simply scale with the cap, every summer’s term is zero. If new money bids them up, the term rises in the summers the cap jumped and falls in the summers it did not.

The chart is the whole argument. The grey bars are how much the cap rose each summer. The red dots are the price of the same player relative to 2015–19, with a 90 percent interval from resampling the contracts. Hollow dots are the baseline summers, which sit at zero by construction.

Open-market UFA skaters who had played at least half the previous schedule, 954 contracts, 2015 to 2026. Price is AAV as a percent of the cap for the contract’s first season, net of production, usage, Net Rating, position and age. Bars: cap increase. Dots: price level vs 2015–19; whiskers are 90 percent bootstrap intervals.

Read the flat years first, because they are the control. The 2020 market, held in October after the bubble with the cap frozen at $81.5M and the gate revenue gone, paid −20% for the same player, and the interval does not come near zero. The three summers that followed, with the cap creeping by a million dollars a year while the players repaid their escrow debt, ran between −9% and −6%. No new money, lower prices, in percent of a cap that was itself barely moving. The percent-of-cap rule does not predict that. A market with a fixed supply and a budget constraint does.

Then the money came back. The 2024 summer, the first with a real increase in five years, got prices back to the old level (+0%). The 2025 summer, cap up 8.5 percent, did not move them (−2%), and we will come back to why. The 2026 summer, cap up 8.9 percent, paid +14% for the same player, with an interval from +3% to +28%. A player who would have signed for five percent of the cap in 2018 signed for about 5.7 percent of a cap that was already 31 percent larger. The two effects compound. In dollars, that player’s contract rose by about 49 percent over eight years, against a cap that rose 31.

Across the 12 summers, the correlation between the size of the cap increase and that summer’s price level is 0.74. The pandemic summers do a lot of that work, so drop the 2 in which the cap did not rise at all and it is still 0.68; leave any single summer out and it stays between 0.59 and 0.85. The slope says each extra point of cap growth has come with about 1.5 further points of price, over and above the point the cap itself hands over. That is twelve data points and one of them is a pandemic, so hold it loosely. It is also the sign and the size the theory predicts.

“New room” is the cap increase times the number of teams. “UFA skaters” counts every priced skater on Spotrac’s list for that summer; “contracts priced” is the subset who had played half the previous schedule and enter the model.

If this were a story about stars, the percent-of-cap people would have a point: the top of our sample is priced at roughly the old level, as the table shows. It is not a story about stars. Split the priced contracts into thirds by the quality the model sees, and compare each third with the same third in 2015–19.

Price of the same player vs the same group in 2015–19. Thirds are by the model’s quality index (production, usage, Net Rating, position), 250 contracts in 2024–26. “Re-signed” means the player’s new team is his old one.

In the squeeze the stars took the cut: the top third of open-market players went for −17% in 2020–23 while the bottom third barely moved. When the money came back it did not go back to them. The top third in 2024–26 is priced at −3% against the old level; the middle third is at +9% and the bottom third at +6%. The windfall bought second-pair defencemen and third-line centres. And it bought them at auction: players who changed teams in 2024–26 fetched +5% against the old level, players who re-signed with their own club −4%. A re-signing is a negotiation between two parties. A move is a sale to the highest of many bidders, and the number of bidders is exactly what a cap jump increases.

The 2026 list makes the point in names. These are the eight largest open-market contracts to skaters who had played half the previous season, with the model’s price for the same player at 2015–19 rates, converted into 2026 dollars.

Age at July 1. The last column is what the model says a player with the same record and age fetched in 2015–19, as a share of the cap, times the 2026-27 cap of $104.0M. It is a market price, not a verdict on the player.

Not one of these is a first-line forward or a first-pair defenceman coming off a career year. They are veterans, most of them past thirty, and the premium on them is not an artefact of a few old names. Take the players 35 and over out of the 2026 sample and the price level is +14%; take the five largest contracts out and it is +12%; use the median contract instead of the mean and it is +20%. By age, the 30-to-33 group was at +17% and the 33-and-over group at +18%, while the 27-to-30 group, the one that normally carries the market, was at −5%. The money went to the players a contending team buys when it has room and nothing better to spend it on.

The teams were pricing against next year’s cap. True, and testable. The league published the path in advance: $95.5M, $104.0M, then $113.5M for 2027–28. A six-year contract signed in 2026 is paid mostly against caps larger than $104.0M, so its share of the cap it will actually live under is smaller than its share of the first-year cap. We re-ran the model with the price measured against the average cap over each contract’s term, using the published figures and five percent a year beyond them. The 2026 premium falls from +14% to +11%. Forward pricing explains some of it. It does not explain most of it, and it cannot explain why the premium sits on two-year deals to 36-year-olds, for whom there is no forward to price.

Then why not 2025? The cap rose 8.5 percent that summer and the open market paid −2%. Two things happened. First, the money left the room before the market opened: the class’s best players extended in June or were traded with a contract attached, and none of those deals are in an open-market sample. Second, what reached July 1 was thin, and the industry said so at the time. The Score, citing PuckPedia, counted 121 contracts worth $672.8 million on July 1, 2025, against 185 worth $1.2 billion the year before. Teams that had planned to spend found less to spend on. In 2026 the class was deeper and the money was still there: PuckPedia counted 164 contracts on July 1 and a total value up 35 percent on 2025. One summer each way is not a law. It is consistent with a windfall that has to find a supply before it can show up in a price.

None of this should surprise an economist, and it should not surprise the league, which watched the NBA live through the extreme version. The NBA’s cap rose roughly a third in one summer in 2016, after the players’ union declined to smooth the new television money in over several years. FiveThirtyEight’s accounting at the time found 45 percent of that summer’s contracts overpaid by at least $5 million a year against its own projections, up from 23 percent the summer before, and the next collective agreement capped annual cap growth at ten percent so that it could not happen again. Hockey’s three years of eight to nine percent are the smoothed version of the same thing, and they appear to be producing the smoothed version of the same result.

Three features of cap room make it behave this way, and they are not peculiar to hockey.

It expires. Room that is not used by the end of the season is gone; the cap does not roll over. Jeffrey Liebman and Neale Mahoney studied United States federal procurement, where appropriations expire at the end of the fiscal year, and found spending in the final week running at about five times the weekly average for the rest of the year, with measurably worse project quality. The one agency allowed to roll over its funds showed neither the spike nor the quality drop. Cap space is an appropriation with a hard expiry and no rollover, and every GM’s fiscal year ends in June.

It is earmarked. Cap room can buy one thing. An owner can bank the dollars he does not spend, but the room itself buys players or nothing, and the floor, $76.9 million for 2026-27, obliges even the thriftiest team to spend most of it whether or not the market offers value. Money that can only be spent on one thing gets spent on that thing. Public-finance economists call the public version the flypaper effect: grants that arrive labelled for a purpose get spent on it rather than substituted away, far more than a fungible dollar would. Behavioural economists call the household version mental accounting, and have shown in the laboratory that an unanticipated sum is spent more freely than the same sum anticipated. New cap room is labelled, and until the league published a three-year path in January 2025 its size was not known until June.

It arrives for everyone at once. This is the part the percent-of-cap heuristic misses entirely. A free agent is sold at an auction with up to 32 bidders, and the number who can actually bid is the number with room. A cap jump turns teams that were pressed against the ceiling into bidders. Jeremy Bulow and Paul Klemperer showed that one additional serious bidder is worth more to a seller than any amount of bargaining skill. And when many bidders estimate the same uncertain value, the one who wins is the one who estimated highest, which is why auction winners systematically overpay. Oil companies learned this bidding on offshore leases in the 1960s and called it the winner’s curse. The re-signed column in the table above is the control: the same summer, the same cap, one bidder, no premium.

The cap rises again next summer, to $113.5 million, another nine percent, and this is the right place to say in advance what the argument predicts, so that it can be wrong. If the mechanism is what we have described, the open market in 2027 should again price the same player above the 2015–19 level, the premium should again sit in the middle and bottom thirds rather than the top, and it should again be larger for players who change teams than for players who re-sign. If instead 2026 was a one-off, the premium reverts toward zero and this piece becomes a post-hoc story about one summer, which is the fallacy this newspaper is named after. We will run the same model on the same source next August and publish the number either way.

For now, the practical reading is this. A rising cap does not make everyone richer in proportion. It makes the players who reach the open market in the right summer richer, it makes a mid-roster veteran worth more than his record says, and it makes a long extension signed before the jump look cheap in retrospect. Teams that treated their new room as money that had to go somewhere spent it where there was something to buy. The ones that look clever in two years will be the ones that remembered room is not savings, and that a budget you cannot keep is the most expensive kind to hold.