THE BETTING FOOL
By a Man Who Should Know Better
New season, new thousand dollars, same guy. Last year finished thirty wins and thirty-three losses and eight hundred forty-eight dollars and ninety-six cents, which is a fancy way of saying I gave back a hundred and fifty-one bucks and a summer's worth of dignity. The bankroll resets. The record doesn't get to.
I spent June and July taking the thing apart on the kitchen table, and I'm going to tell you what I found, because that's the deal we have. The engine I published all last season — the one built on efficiency ratings, the one I wrote four hundred words about in September — was not making predictions. Across twenty-six hundred completed games its average disagreement with the closing spread was twelve one-hundredths of a point. Not twelve points. Zero point one two. It was taking the number off the board, adding a rounding error, and handing it back to me like a waiter returning my own wallet. Its totals arm called the over on sixty-nine percent of games in a league that goes over forty-nine percent of the time, which isn't a forecast, it's a thumb on a scale.
So it's retired. What picks now is the box-score model — six terms, passing yards, rushing yards, sack yards, interceptions, net fumbles, points that don't come from touchdowns — and it at least has opinions. It disagrees with Vegas by three and seven-tenths points a game. Over twenty-four hundred games it hits fifty-two point four percent against the spread. You need fifty-two point four to break even at minus one-ten. So: it is exactly, precisely, insultingly break-even. I've built a machine that produces a coin.
And here's the part that made me put the coffee down. That model has no home-field term in it. None. It predicts each team's score from that team's own numbers and never asks where the game is played, so it hands the home team about one and eight-tenths points less than home teams actually earn. Which is why it picks the road team two-thirds of the time and why four of the five plays below are road dogs. I tried adding home field back. The model got worse — fifty-one point six. Because the market already knows about home field, and a model that agrees with the market is a model paying juice for the privilege of having no opinion. The edge, such as it is, is the bias. I don't love it either.
Two things are new on every card from today. First, the price is the real price — not the minus one-ten I'd been assuming all last season, but what's actually posted on each side. Second, next to the cover probability there's now a number called NEEDS TO WIN, which is what that price demands. A total at minus one-twenty needs you to win fifty-four and a half percent of the time, not fifty-two four, and printing the first number without the second is how a losing bet gets dressed up as a good one.
Third thing, and this one is for the pool players, of whom there are more of you than I realized. The card now carries ten positions in two sections. The Five is five sides, ranked one through five, every single week — because a pick-five pool makes you submit five whether or not any of them is a bet worth making, and a fifty-one percent side is a perfectly sensible pool pick even though it's a lousy wager. A pool pick fights the other guys in the office. A bet fights the vig. Those are different games and I'm going to stop pretending they're the same one. The Board is the totals, up to five of them.
Every position carries a tag that tells you which it is. BEST BET means the model's number clears the posted price and I've got fifteen dollars on it. WIND means it's the weather rule and it gets twenty-five, because it's the only thing here with real evidence behind it. LEAN means it does not clear its price — it's on the card for the pool and for the sake of showing you what the model was looking at, and it gets five dollars and a grey box. Seven of thirty-two candidates cleared their price this week. Three of these ten are leans and they say so on their face. The season record splits the two apart, too: the headline number and the bankroll answer to the bets, and The Five keeps its own pool record beside them.
One play here is not the model's, and I want to be plain about that. The single thing this operation has ever found that beats the number is outdoor games in real wind: ten miles an hour or more, take the under, fifty-seven point seven percent over six hundred and three games, eight of the last ten seasons. It's the only finding I have that survived being attacked properly. Monday night at Arrowhead is forecast at thirteen miles an hour at kickoff, easing to nine by the fourth quarter — call it eleven across the game. So the under goes on the card, and it goes on at a bigger number than anything else, and it goes on over my own model's objection. The box-score model wants the over there by nearly three points. It can want. Its totals arm grades at forty-nine and a half percent across nine seasons, which is worse than flipping a coin, and gating a real edge behind a useless one is how you end up with neither.
Ten positions. Seventy-five dollars across the five sides, twenty-five on the weather, twenty-five on two totals that clear, fifteen on three that don't. A hundred and thirty at risk out of a fresh thousand, thirteen percent, and the biggest single number on the page is the one my own model argued against. Stakes go by tier and nothing else — nine seasons say this model cannot tell its own good ideas from its bad ones, so sizing by its enthusiasm would be sizing by noise. Across all sixteen sides this week its cover probability ran from fifty-two point oh four to fifty-two point nine one. Less than a point of daylight between its best idea and its worst. That's not confidence, that's a rounding error, and it isn't getting paid like confidence.
Last thing, and it's a caveat rather than a flourish. That Arrowhead forecast is seven days out. Wind forecasting at seven days is not much better than looking up the September average and shrugging. I'm posting it now because posting before kickoff is the entire deal here, but I'll re-pull the number Saturday, and if the wind has fallen out of it I'll say so in this space rather than letting a dead play sit on the card looking respectable.
There's a PDF of the whole card at the top of this page — one page, everything on it, for the pocket or the printer or the group chat. Same numbers as the page, generated from the same file, so it can't drift.
Five sides, every week, whether or not any of them is a bet worth making — because a pool makes you pick five. Each one is tagged with what is actually behind it.
Cincinnati lays three and a half at home and my number says Tampa Bay is the better football team by nine-tenths of a point, which puts the honest line somewhere around Tampa pick-'em. Four and four-tenths of daylight, the widest gap I've got that isn't the model shouting into a canyon. Bengals at home in September with a fifty and a half hanging over it is the sort of game the market prices off last January's memory. Priced at minus one-oh-five, which means it only needs fifty-one two. It gets fifty-two five. That's the whole margin — one and a fifth points of it — and I'm not going to pretend it's more.
The Jets are getting a point and a half in Nashville and my number has this game dead even, seventeen apiece. Tennessee has been priced as a home favorite in a game between two teams that combined for very little last season, and a thirty-nine and a half total says the market agrees nobody's scoring. When the projected margin is two-tenths of a point, the side you take is whichever one is getting points. That's not insight, that's arithmetic. Minus one-oh-five again, needs fifty-one two, gets fifty-two two.
Here's the biggest disagreement on my board and the one I trust least, so let me put the knife in it myself. Vegas has the Chargers by ten and a half. I have this game at Chargers by one. Nine and three-tenths points apart, and that ought to be a klaxon, not a green light — because my model has no strength-of-schedule term either. It takes Arizona's raw production and Los Angeles's raw production and subtracts, and it does not know or care who those numbers were compiled against. A double-digit Week 1 line usually means the market knows something structural. I'm taking the ten and a half because ten and a half is a lot of points and my number says it's absurd, but if this one comes back a nineteen-point beating, the model's ignorance is the reason and I'll say so here in seven days.
Miami getting three and a half in Las Vegas, and my number has the Dolphins winning by five and a half — a nine-point gap on a forty and a half total, which is the market saying it expects a rock fight. Both of these teams project under twenty-one points, so this is a game decided by which offense is less bad on the day, and in that kind of game you want the points rather than the certainty. The projection is Dolphins twenty, Raiders fourteen. Straight minus one-ten, so it needs the full fifty-two four and clears it by half a point. Half a point.
The Rams lay three and a half against San Francisco on Thursday night at the Melbourne Cricket Ground, which is a sentence I did not expect to type in my lifetime. Eleven thousand miles, a ground built for a different sport entirely, and the NFL has declared Los Angeles the home team. Here's the one time my model's blind spot works in its favor: it has no venue term at all, so it isn't wrongly crediting the Rams for a home field that is on the other side of the planet. It just says Los Angeles is four and a tenth points better and the number is three and a half. Six-tenths of an edge, minus one-oh-eight, clearing its price by two-tenths of a point. It is the thinnest play on the card and it is on the card because it clears, which is the only rule that matters now.
Forty-two and a half in Kansas City on Monday night with a thirteen-mile-an-hour wind at kickoff, easing through the evening — about eleven averaged across the game. This is the one rule I have that has survived every attack I've made on it: outdoor, wind over ten, take the under. Fifty-seven point seven percent across six hundred and three games. The mechanism is right there in the market's own numbers — books drop the posted total from about forty-five to about forty-four when it blows, while actual scoring falls from forty-six to under forty-three. They move a third of the distance the scoreboard does. My own model hates this bet, incidentally; it has the game at forty-five and a third and wants the over. I'm overruling it, on purpose, because its totals arm is a forty-nine and a half percent proposition and this rule is not. Fifty-four and a half after I discount the raw figure for the fact that a seven-day wind forecast is a guess wearing a decimal point. Needs fifty-two four. Twenty-five dollars, the biggest number on the card, and I'll re-check the forecast Saturday.
The second bite at the same game, and a separate question from the side. Fifty and a half is the biggest number on the board this week, and my total has it at forty-three and eight-tenths — six and seven-tenths under. What makes this one worth real money isn't the size of that gap, it's the price: the under is hanging at even money, plus one hundred, while the over is juiced to minus one-twenty. At even money the bet only has to win half the time to break even instead of the usual fifty-two four, and the model says fifty-one two. That's the entire margin, and it exists because of what the book is charging, not because I know something about Cincinnati's defense. Fifteen dollars.
A lean, and I'll tell you exactly why it's a lean. Forty-four and a half in Seattle on a Wednesday night opener, my number says forty-five and nine-tenths, so the model wants the over by a point and change. It gets fifty point eight percent. The over is priced at minus one-oh-five, which needs fifty-one two. It comes up four-tenths of a point short. That is a losing bet by its own arithmetic and it is on this card only because a pick-five pool needs five and a board needs filling, and because I would rather show you the near-misses than pretend the week produced ten good ideas. Five dollars, which is what I think it's worth.
Forty and a half in Las Vegas, the lowest total on the slate, and my number is thirty-four and four-tenths — six points under. You'd think that would be a play. It isn't, because a total that low is already priced for two bad offenses, and my model has both these teams projected in the teens off last season's production without knowing a thing about what either did in the offseason. Fifty point nine percent against a price that needs fifty-two four. A point and a half short. Lean, five dollars, and if it cashes I got lucky rather than smart.
The other half of the Arizona game, and the weakest thing on the card. Forty-six and a half, my number says forty-three, and the model still only gets to forty-nine seven percent — under fifty, which means by its own reckoning this is a coin flip that loses. It's here to round out the board and because I said I'd show you what the model was looking at, not just what it liked. Five dollars. If you're filling in a pool card, ignore this one entirely; it's a totals lean and pools don't take totals.
“Bet less than you brag, learn more than you lose, and never trust a tout who smiles too much.”
— The Betting Fool
Disclaimer: The Betting Fool is for entertainment purposes only. Nothing on this site is financial advice, and past performance does not guarantee future results. Must be 21+ to wager where legal. If you or someone you know has a gambling problem, call 1-800-GAMBLER.