What Are Teams Paying For A Win In Free Agency? 2026 Edition

What are teams paying for a win in free agency? Earlier this month, I answered a FanGraphs Weekly Mailbag question about that very issue, outlining a rule I’ve been using in formulating my contract predictions. I left my explanation loose and vague because it was one of four questions in a mailbag, but to give you the general gist, I think about free agent salaries on a graduated scale, with role players being paid less per win above replacement than superstars. Today, I’d like to back up my argument with a bit more mathematical rigor.

One of the benefits of writing for FanGraphs is that smart baseball thinkers read the site. I woke up last Monday to a direct message from Tom Tango, MLB’s chief data architect. Tango had a few suggestions for further research, a method for adjusting past years of data for current payroll situations, and even a link to a discussion of the cost of a win with Sean Smith. Smith, better known as Rally Monkey, is the creator of Baseball Reference’s calculation of WAR – when you see rWAR, that actually stands for Rally WAR, not Reference WAR. In other words, I got help from some heavy hitters.

With Smith’s excellent article on free agency as a guide, I built my own methodology for examining the deals that free agents receive and turning them into a mathematical rule. I took every starting pitcher and position player (relievers are weird and should be modeled differently due to leverage concerns) and noted their projected WAR in the subsequent season, as well as the length and terms of their contract. I excluded players who signed minor league deals, were projected for negative WAR, or whose contract details were undisclosed. To give you a sense, applying this approach to the 2025-26 offseason leaves us with 89 players, from Kyle Tucker all the way down to Jorge Mateo.

I then used a formula, lightly modified from the one Smith uses, to handle future years. I assumed a decline of 0.4 WAR per year on all projections due to aging. Smith used a 0.5 WAR decline, but 0.4 is the mean decline across the multi-year ZiPS projections I used most recently, so I went with that. This tends to depress the modeled $/WAR cost of stars, who sign longer contracts with a less precipitous WAR decline, but I think it reflects reality better. I also applied a 5% annual inflation rate to future year guarantees. That’s the rate Smith used, and it makes sense to me. I put everything in terms of 2025 dollars, too, just to put each year on the same scale. For the record, this inflation-and-aging method is what I use in my own contract modeling and in player evaluation for our annual Trade Value Series – I just try to get better measures of aging directly from Dan Szymborski. I think it’s a great theoretical foundation.

With this data in hand, I went about analyzing it in two different ways. First, I just added up all the projected WAR on one hand and all the inflation-adjusted guaranteed money on the other, and divided the second by the first. That’s as simple as it gets, with one dollar-per-WAR level for everyone. Well, almost as simple as it gets: I subtracted the minimum salary for each year, for reasons I’ll explain later.

The second method makes me incredibly thankful that I read Smith’s piece on the cost of a win. If you’re familiar with his work, you know that he has a knack for cutting to the core of the issue and lopping off needless complexity. That’s what happened here. I have a complicated contract model that I use to make player-by-player predictions every year, and I approached the question of how much a win costs in free agency from that angle. But there’s an easier way to do it, and I loved Smith’s framing so much that I used it to guide my analysis.

I divided players into tiers based on their Steamer-projected WAR in the first year of their new deals. I calculated the same dollar-per-WAR amount from up above separately for every tier, with the same minimum salary adjustment from up above. I went with tiers of 0-1, 1-2, and 2-plus WAR for the sake of simplicity, though I also tested other breakpoints with similar results, including a test where I excluded the top free agent from each year to avoid outlier problems. It’s easier than working out different costs for different marginal wins, and the results were telling in every case.

Obviously, each of these methods knows how much total money was spent in free agency, so they both have the total spending level perfectly correct. Given that I used actual contracts as a guide, it couldn’t be any other way. They perform differently in describing individual contracts, however, and the three-tier model does much better. To explain the difference, I’ll first present the results from the 2025-26 offseason, then follow with an expanded method that I used to handle a sample of the last seven offseasons, as far back as the RosterResource database goes.

In the 2025-26 offseason, 19 players projected for 2 or more WAR by Steamer signed contracts in free agency. They received an aggregate $1.866 billion in guarantees. After accounting for the length of their contracts, using Smith’s inflation-and-aging method of valuing future years, they received $12.84 million per projected WAR. Steamer projects another 29 of this winter’s free agents for between 1 and 2 WAR. They got paid $8.51 million per projected WAR. Finally, 41 free agents project for less than 1 WAR in 2026; they received $6.74 million per WAR. That’s the three-tier model. The one-tier model would tell you that there were 89 free agents, and that they received $11.23 million per projected WAR.

Now, both of those things are true. I don’t find them to be equally descriptive, however, and the numbers back me up. For one thing, the three-tier method has a mean absolute error of $8.5 million, as compared to $9.7 million for the simple rule. But to get to more measures of statistical interest, we’ll have to broaden out the sample.

I took every free agent contract in the RosterResource database going back to the 2019-20 offseason. For 2020, I pro-rated the projections up to a full season; the contracts players signed in that year were signed before the COVID interruption, and I’d use the full-season projections even if they weren’t, because a 2020 WAR projection for a 60-game season doesn’t play nice with aging curves unless you convert it into a full-season projection. I took the projected pitching and batting WAR for all players and added them up to account for two-way players – thanks, Ohtani, for the extra complexity.

Here are the per-year results:

Take a look at the overall numbers. One reason I like the three-tier formulation as a description of free agent spending is that using a flat dollars-per-win estimate has directional bias. If you used it to predict all the contracts in each bucket based on their length and the player’s projected WAR, it would miss low by about $10 million on the average contract to a star, miss high by about $4 million on contracts to role players, and miss high by about $2 million on contracts to the 0-1 WAR bench options.

Another way of looking at it is that if I compute confidence intervals for dollar-per-win rates for each tier independently, they’re appreciably different. Players in the 0-1 WAR tier have a $6.3-$8.6 million 95th-percentile confidence interval. The 1-2 WAR tier checks in at $7.8-$9.4 million, maybe a small overlap with the bottom tier. But the stars, the guys in the top bucket? They check in at a $10.6-$12.9 million estimate. They’re very clearly in a different category than the guys at the bottom of the market, even after accounting for variance and the inherent limits of sample size.

I wanted a little more statistical assurance, so I ran two more tests. First, I ran a permutation test. Instead of splitting players up by projected WAR in year one, I split them up completely at random, 1,000 different times, and then checked whether my random splits did better or worse than the by-WAR tiering. If grouping by WAR didn’t add any information, we’d expect to see a null result, but instead, only one of the 1,000 random permutations grouped free agents better, in terms of dollars per projected WAR, than the WAR-based sort. In other words, it’s very unlikely to be random.

Adding moving parts to a model always makes it fit better, even if those moving parts are just modeling noise. To account for that, I ran a likelihood ratio test, which you can think of as comparing how good each of these methods is at describing the contracts in the dataset, while penalizing added complexity in a fancy math way that I had to read about. If you spent some time taking stats in college, it’s like an F-test with a bit more flexibility. I used that added flexibility to test the three-tiered model against the one-tiered model using both a normal and lognormal distribution, and both distribution options reduced the standard error of predictions by about 13% even after the complexity penalty. In other words, the three-tiered model is showing real market differentiation.

Here’s an interesting effect, and one in keeping with my qualitative experience of free agency: The gap between the bottom and top tiers is widening. From 2020 through 2023, players in the 2-plus WAR tier got about 40% more per projected win than players in the 0-1 WAR tier. Even if you toss out the weirdo 2020 season, they got about 55% more. In the past three years, they’ve gotten roughly double the dollars per projected win. In other words, the market is assigning more and more per-WAR value to top players. I can’t tell you whether that’s driven more by the supply side or the demand side of the equation (and make no mistake, both are drivers), but whatever the cause, star players who reach free agency are getting paid more on a relative basis than ever before.

That’s the conclusion. Here are some of my takeaways about it. First, this tracks with how I think about the best teams in the game acquiring talent, and I’m unsurprised to see the general direction of the trend. Observing the trade market in recent years has led me to think of value non-linearly for the Trade Value Series. You can’t just value all wins equally and add them up; teams increasingly value stars at a higher multiple than role players. Contracts signed before the Judge/Ohtani/Soto mega-expansion, a rising tide that lifted all boats, are also starting to look pretty attractive relative to new deals.

That segues nicely into my next point, a fun little mathematical observation. I adjusted these numbers for inflation, and yet a win above replacement cost $9.8 million in 2020-2023 free agent contract negotiations, and $11.7 million from 2024-2026. Where’d the inflation adjustment go?! But I adjusted the numbers based on total league-wide payroll. Take 2021, for example: Teams shell-shocked by COVID offered comparatively small deals in free agency, but existing contracts meant that overall league payrolls fell by much less. In general, though, free agent salaries are increasing faster than the rest of the league, and by a huge margin. In fact, it’s fairer to say that salaries for top free agents are increasing, because the dollar-per-WAR values in the 0-1 WAR and 1-2 WAR buckets are keeping up with league-wide payroll inflation, while the dollar-per-WAR values going to stars are up 30% on a relative basis in recent years. Yes, major league salaries are increasing steadily, but it’s a stratified phenomenon.

Finally, a quick note on adjusting for the league minimum salary. When I ran this analysis the first time, I split it down into 0.5-WAR buckets to look for weird effects. The stars still made the most money per projected win, but the 0.0-0.5 WAR bucket was a close second. I figured that one out pretty quickly, though: If a team pays a guy $800,000 when the league minimum is $750,000, and he projects for 0.1 WAR, that’s $8 million per WAR and also not a reflection of the tradeoff the team faces. To them, he didn’t cost a ton more than $50,000 (plus the loss of a flexible roster spot). Contracts in range of the minimum salary, and in range of replacement level, might quite reasonably behave differently in recognition of that effect, so I included it in my modeling. I don’t think it had a huge effect aside from cleaning up that strange anomaly.

I’m not sure how much bearing this research has on the future of the game, because I think that the upcoming CBA has the potential to change player compensation, perhaps seismically. Whether it’s a cap/floor system (something I find unlikely, to be clear), a revamped arbitration process, or something I haven’t even thought of yet, the economics of the sport could look pretty different the next time pitchers and catchers report to spring training. But for now, the data leads me to a clear conclusion: Teams value stars non-linearly, and they pay more for top players than you’d think from a broad look at the total market.