95° in Coney Island: Betting NO on Joey Chestnut 70+

It started with a YouTube video. On May 9, 2026, Joey Chestnut broke his own record at a bologna eating contest — 16 pounds of smoked bologna in 8 minutes, up from 15.75 the year before. That video matters because it kills the laziest version of this bet before it starts: Chestnut is not washed. He's 42 and still setting personal bests. If you're going to bet against the greatest eater of all time, the thesis had better not be "he's declining."

Mine wasn't. My thesis was about the weather.

The market

Kalshi listed "Joey Chestnut: 70+ hot dogs" for the 2026 Nathan's contest. When I got involved, YES was trading around 49¢ and I could buy NO at 52¢ — with real size sitting on the book. Over $85K resting at the 52¢ ask alone.

That last part is the actual story. These novelty props have existed at sportsbooks forever, but historically you couldn't get more than $100 down before the book cut you off — they were loss leaders for content, not markets. A prediction market with an open order book changed that. For the first time, doing real research on a hot dog eating contest could pay real money.

I bought $20,000 of NO at 52¢.

The thesis: nobody checks the weather

The National Weather Service hourly forecast for Coney Island (gridpoint OKX 35,36) for July 4, 2026 at 12:30 PM ET: 95°F, mostly sunny, 48% humidity. Pulling official NOAA/NCEI hourly observations from JFK — the nearest long-term station — for every contest back to 2007, the picture is stark: the contest has only been run at 95°F once before, in 2010. Every other year sat between 70°F and 85°F.

Competitive eating at that pace for ten minutes is an athletic event. Heat should matter. The question is how much, so I built the dataset properly instead of vibing it: closest JFK observation within ±45 minutes of the 12:30 start, humidity derived from temp/dew point, NOAA's own heat-index formula. Two years needed manual handling — 2020 was held indoors with AC (81°F outside, plotted at ~75°F), and 2023 had a two-hour thunderstorm delay (plotted, excluded from the trend).

Recent window: r = −0.73, and the linear estimate at 95°F is 59.3 hot dogs. Toggle the 2023 storm-delay point back into the trend and watch the slope flatten — that fragility matters later.

His last three outdoor contests: 63 (2022, 82°F), 62 (2023, storm delay), 70.5 (2025, 83°F). The NO side needs him under 70 — he'd been under it in two of his last three outdoor appearances at temperatures 10+ degrees cooler than the forecast.

Zooming out to his full career weakens the correlation (r = −0.19 — one hot dog record was literally set at 85°F), but even the full-career trend line only estimates 64 at 95°F. And the one prior 95°F contest, 2010? He ate 54, in an era when his ceiling was the high 60s.

Full career: the relationship is much weaker, but the 95°F estimate still lands at 64 — under the line. The lonely 54 at 95°F is 2010; the 74 at 85°F is the revised 2018 count.

The 2018 problem

One thing in the data genuinely worried me: 2018. Joey was originally scored 64 that day, and the count was later revised up to 74 due to an "error." It was also one of the hottest contests on record (~85°F) — so is that point a heat-defying 74, or a soft count? I went looking for the raw footage to recount it myself — there's an ESPN3 dedicated plate-camera feed that exists somewhere — but the videos are too old and I couldn't find it.

I couldn't resolve it, so I kept it in the data and kept the lesson: judge counts are soft, and revisions historically go up, not down. If you're short at the number, that's settlement risk. File that away — it comes back later.

The counter case

Here's the question that should bother you, and it's the one I'd ask anyone showing me this research: if the under is so obvious, why is someone resting $85K+ on the book to give you NO at 52¢? This tweet from BrodysBets is a good artifact of what the other side looked like — it's really obvious what was happening there.

The honest steelman:

• Five data points is curve-fitting territory. My r = −0.73 leans hard on 2022 and 2023. Plot the same eaters against humidity instead of temperature and the story inverts completely — Joey correlates at r = +0.85 with humidity, and that model predicts 69.8 for 2026. Small samples will tell you whatever story you went shopping for.
• He ate 70.5 last year at 83°F and 74 (allegedly) at 85°F in 2018. The heat penalty, if real, might be a few hot dogs — and the line only needs him to find 70.
• Count revisions skew up. The one mechanism I documented (2018, and later Miki) works against the NO side at the margin.

The same data, different axis, opposite conclusion. This chart is in the post to keep me honest.

And who was actually on the other side? July 4th is the one day a year normal people bet a hot dog contest. Kalshi was running ads with Chestnut himself. The YES side was holiday money anchored to two numbers: last year's 70.5 and the champ's record of 76. My read wasn't that the counter-case was empty — it was that it was priced like a certainty by people who hadn't done any work. Low hurdles: something other people would have bought too, if they'd done the work. Nobody does the work on a hot dog contest.

Contest day: the Miki Sudo shock

July 4 came in as forecast — brutally hot. The women's contest runs first, and I was watching it as a live read on the heat thesis. Miki Sudo looked like she was really struggling. She kept dumping water over her head between dogs just to keep going. Everything the model wanted to see.

Then they announced her total: 34. And then they revised it up to 38.75.

That shook me twice in one minute. First, the naive temperature model had projected Miki at 8.6 hot dogs at 95°F — she beat it by thirty. The linear heat penalty was clearly far too steep, and Joey's 59.3 estimate came from the same machinery. Second: there it was again, the upward revision, live, in the exact contest where I was short the count.

The steepest trend line in the dataset — and she beat its 2026 estimate by 30 hot dogs. When your model misses by 4x, you stop trusting the model.

My edge was a model, the model had just failed its only live test, and NO had run up to the low 60s anyway. I sold out at 61¢. On $20K at 52¢, that locked in about +$3,500.

Result

Joey Chestnut ate 66 hot dogs. Under 70. NO settled at $1.00.

Holding to settlement would have paid about $18,500 instead of $3,500, so the scoreboard says the hedge cost me fifteen grand. But I'd make the same exit again. The thing I actually knew — 95°F makes 70 harder, and the market is full of anchored holiday money — was always more robust than the thing the chart said — the slope is 1.7 hot dogs per degree. When Miki demolished the model mid-contest, the honest move was to admit the precise version of my edge was gone and take the imprecise version's profit. The direction was right. The magnitude was a guess dressed up as a number.

The bet was never really "Joey eats 59." It was "the odds of him clearing 70 in that heat are worse than 52% and I'm the only one at this window who checked." That part held: 66.

Then I went back to doing nothing.