Four Firms Have Now Forecast This NFL Season. Their Numbers for Last Season Are $2.3 Billion Apart.

Editor's note: TrueEdge builds odds tools and earns affiliate commissions from licensed sportsbooks. This piece argues a position on how prediction-market competition is being measured, and our commercial interest runs toward the sportsbooks. Weigh it accordingly.

Eilers & Krejcik Gaming told its subscribers on September 17 that American bettors will put $31.7 billion through licensed sportsbooks on the NFL this season, up about 8% year over year. Eight days earlier, H2 Gambling Capital told its subscribers $31.4 billion, down 0.8% — the first decline in the legal era.

Those two numbers are three hundred million dollars apart, which is nothing. The growth rates attached to them are 8.8 percentage points apart, which is everything. Two professional research firms, working the same season with the same public data, produced forecasts that are almost identical in level and flatly opposed in direction.

The only way both can be true is if they disagree about where the season started. They do.

The arithmetic nobody ran

Divide each forecast by one plus its own growth rate and you recover the baseline the firm is working from.

EKG: $31.7 billion at +8% implies a 2025 NFL season of roughly $29.35 billion.

H2: $31.4 billion at −0.8% implies $31.65 billion.

That is a $2.3 billion gap on a season that finished in February, has been reported by every state that publishes handle, and appears in the quarterly filings of two public companies.

Now put the other two forecasters beside them. The American Gaming Association's September 4 release used $29.4 billion for 2025 and projected $29.5 billion for 2026. RotoWire, working from state reports and operator filings, put 2025 at $31.76 billion and projected $32.3 billion.

EKG's implied baseline sits within $50 million of the AGA's stated one. H2's sits within $110 million of RotoWire's. There are not four estimates of the 2025 NFL season in circulation. There are two, and they are two billion dollars apart, and which one a firm uses is the single best predictor of whether it thinks this season is growing.

There is a cleaner way to say it. EKG's optimistic forecast for 2026 — $31.7 billion — is within $50 million of what H2 believes 2025 already was. One firm's bull case for this season is the other firm's record of last season.

Why the baselines differ, as far as we can tell

Most states do not break handle out by sport. Everyone estimating NFL handle is allocating a football share out of a total, and the allocation method is the whole ballgame.

The AGA's own note describes applying national year-to-date growth projections to the prior season's football total, then using football-specific reporting from selected states to estimate the pro-football share. RotoWire's describes working up from state handle reports, by-sport breakdowns where states publish them, operator earnings calls and SEC filings. One is a top-down allocation; the other is a bottom-up assembly. They produce different answers because they are different exercises.

We do not know EKG's method for the NFL line. The EKG Line note is a summary of subscription research, and the underlying report is not public. We also cannot say with confidence whether EKG's $31.7 billion covers online only or online plus retail, which alone could move a number by a few percent. That is a real gap in this piece and we would rather flag it than paper it.

One more thing we cannot reconcile: the same note has been reported with two different totals. Next.io's write-up on September 18 gives EKG's all-channel figure as $40.1 billion; casino.org's, the same day, gives $40.5 billion. The components — $31.7 billion of sportsbook handle and $8.4 billion of prediction-market handle analog — sum to $40.1 billion. We are using $40.1 billion and noting the discrepancy, because one of those two write-ups has a typo in it and we cannot tell which from outside.

The measurement error is bigger than the thing being measured

Here is why this is not pedantry.

The argument the whole industry has been having since August is whether prediction markets are taking handle from sportsbooks. H2 has published the only explicit number: it estimates underlying NFL betting demand is still growing around 4%, and that prediction markets will remove 4% to 6% of sportsbook handle in competitive legal states this season. Subtract, and you get its small negative.

That is a testable claim, and we said so on September 9 when we wrote about it. But test it against what? The effect H2 is claiming is 4 to 6 points. The disagreement between EKG and H2 about this same season is 8.8 points. The disagreement about the baseline is nearly 8%.

A substitution effect of five points cannot be detected inside a measurement whose practitioners are eight points apart on the starting value. When the season ends and the state filings land, both firms will be able to declare themselves right, because both will be comparing the outcome to a different 2025. Nothing will have been settled, and the same four organisations will publish four more numbers next August.

The strongest case against us

The obvious rebuttal: forecast dispersion is normal and healthy, scope definitions differ legitimately, and nobody serious treats a September projection as a measurement. Analysts disagree about everything. Two research firms landing $300 million apart on a $31 billion market is close agreement, not a scandal.

Most of that is right, and if these numbers stayed inside subscription notes we would not have written this.

They do not stay there. The AGA's figure was built into a lobbying release that used the word "backdoor" in its headline. Missouri's attorney general sent six cease-and-desist letters yesterday. Ohio's iGaming sponsor spent Friday explaining a promotional-credit ban to the online-gaming industry's own trade association. New Jersey has a certiorari petition pending and the Supreme Court's long conference is September 28. The substitution question is currently being used to justify tax rates, licensing regimes and criminal referrals, and the people using it are citing whichever of these four numbers supports the thing they already wanted.

There is a second rebuttal worth taking seriously, and it comes from EKG itself. Chris Grove, the firm's partner emeritus, is quoted saying the quiet part: "Prediction markets are growing the overall market — at least for now." That is the whole reason EKG can forecast +8% while H2 forecasts −0.8%. EKG is not disputing that the exchanges are large; it is disputing that they are cannibalising. If prediction markets are pulling new people into wagering who then also open a sportsbook account, handle rises on both sides, and every substitution model built this summer is measuring the wrong thing.

We think Grove is probably right about this season and probably wrong about next. His own sentence contains the hedge — "at least for now" — and he follows it with the timeline: "We'll start to see more direct competition between sportsbooks and prediction markets by the time we get to the Super Bowl." Expansion and substitution are not alternatives. They are sequential.

What would settle it

Pennsylvania's August sports betting revenue fell 20% year over year. New Jersey's fell 9%. Both landed on September 18 and both will be quoted all week as evidence of substitution.

Neither is. New Jersey's report attaches its own explanation — college football started later this year and pushed handle into September. Pennsylvania's has not been disaggregated at all, and a revenue figure is hold multiplied by handle, so a 20% revenue decline is consistent with handle being flat and the books running cold.

The number that would actually settle the argument is the one nobody publishes: NFL handle by state, by market type, month over month, from states that legalised at different times. If the exchanges are removing 4 to 6 points in competitive states, mature states should underperform recently-launched ones by roughly that margin, and sides and totals — where exchange liquidity is deepest — should underperform parlays, where it is thinnest.

Those filings exist in fragments. Assembling them is a month of work that four firms with the resources to do it have chosen to replace with an allocation assumption. We would rather read that than another projection, and if nobody publishes it by the new year we will try to build it ourselves and show the workings.