From “I just back my team” to thinking like a trading desk

Imagine a fair coin flip. Each side has a true 50% chance, so fair odds are 2.00 on both. A bookmaker doesn't offer 2.00. They offer around 1.91 on both sides.

Bet blindly on a 50/50 shot at 1.91 and you lose about 4.5p of every £1 in the long run, even with perfect luck. Football match-winner markets typically carry 3–8% margin, and each-way markets can be worse. This is the tax every guesser pays on every bet.

Picking the team you “feel good about” (or the horse with a nice name or the right colours) has nothing to do with the price. Bookmakers already build in everything the public knows, so a team being good doesn't make it a good bet: a great team at too-short odds loses money long-term, and a poor team at long-enough odds can make money. Value comes from the gap between price and true probability, not from who wins.

Think of each outcome as a share. Its true probability is its fair value. If a slow high-street bookmaker is still paying more than fair value, you are buying a share below its worth. Do that consistently, size every position carefully, and the maths works in your favour over many bets. We call that gap +EV (positive expected value).

Pinnacle is the world's sharpest bookmaker: low margin, and it welcomes winning players, so its prices are the best public estimate of reality. But even Pinnacle's odds include a small margin, so we remove it. Instead of shaving every outcome equally, we use power devigging, which also corrects the well-known favourite-longshot bias (longshots are over-priced by the public, favourites under-priced).

Synthetic hold (Miller-Davidow) measures the whole market, not one bookmaker: take the best price available anywhere on each outcome and add up their implied chances.

Our own model predicts the score. For each team we estimate expected goals from how many it scores and concedes versus the league average, add a home-advantage boost (×1.15) and a small form adjustment. We then build a grid of every scoreline from 0-0 to 10-10 (11×11 = 121 possibilities).

Plain Poisson maths slightly misprices low scores like 0-0 and 1-1, so Dixon-Coles applies a correction (ρ = −0.10). Add up the right cells of the grid and every market falls out:

• 1X2: home win = all cells where home goals beat away goals; draw = the diagonal.
• Over/Under: add the cells where total goals are above or below the line (1.5, 2.5, 3.5).
• BTTS (both teams to score): add every cell where both teams score at least 1.

Bill Benter's idea: don't trust one model, blend your fundamentals with the market. We blend the sharp market and our goals model, with weights that depend on how efficient the market is:

Each pick stores snapshots of the offered price and the fair (sharp) price, drawn as a 24-hour line chart. If the sharp line is falling toward your price, the market is agreeing with you.

CLV (closing line value) compares the price you took with the final sharp price just before kickoff. Consistently beating the closing price is the strongest evidence of a real edge, stronger than short-term wins. We record it automatically and settle results after the match.

An each-way bet is two bets: half your money on the horse to win and half on it to place (finish in the first few). The place half pays a fraction of the win odds, for example 1/5 odds for the first 4 places. Stake £10 E/W and you spend £20.

From the devigged Betfair Exchange win chances for every runner, the Harville formula works out the chance of each horse finishing 1st, 2nd, 3rd and so on, given the other horses. Plain Harville tends to overstate longshots' place chances, so we apply Henery / Lo-Bacon-Shone discounting (λ = 0.88) which tempers it. We then compare the true place chance with what the bookmaker's terms imply.

E/W Extra Place Steal: bookmakers love advertising extra places (5 instead of 4 in a big handicap). But they keep the fraction at 1/5, so the place half becomes much more likely to land without the price adjusting enough. When the model's place probability beats the terms by a clear margin, the place leg is +EV even if the win half isn't.

Dead-Eight Bad Favourite Steal: eight runners is the smallest field paying 3 places. If the favourite is priced too short, its probability really belongs to the rest of the field, so rivals reach the places more often than the terms assume.

Five coloured boxes show whether today's conditions suit the horse:

BOG: if the starting price (SP) ends up bigger than the price you took, the bookmaker pays the bigger one; if it shortens, you keep yours. It's a free option, worth roughly +1.5% on the win leg in our maths. Look for the BOG ✓ tag.

Rule 4: if a horse is withdrawn before the race, bookmakers deduct a slice from all winnings, bigger when the withdrawn horse was a short price. Our stress test shows the shortest rival price whose withdrawal would wipe out your edge. “Rule 4 safe > @2.80” means you're fine unless a rival priced 2.80 or shorter is withdrawn.

A bet with a 25% chance loses 3 times out of 4. With a real 5% edge you will still hit runs of ten or more losing bets by pure chance. That is not a broken system; it is maths. Panicking, doubling stakes to “get it back” or abandoning the plan after a bad week is how a positive-edge bettor turns negative.

The Kelly criterion gives the stake that grows a bankroll fastest for a given edge, but full Kelly swings wildly and punishes any error in the edge estimate. We recommend Quarter-Kelly: a quarter of the full amount. You give up a little growth for far smaller drawdowns and a much better chance of staying in the game.

In your Portfolio we simulate 1,500 possible futures for your bets (using their real stakes and odds and a modest assumed +3% edge). The shaded band is where 95% of those futures fall. Your actual profit line is drawn on top.

Small samples are noisy: judge the system on hundreds of bets and on CLV, not on this week's result.

Your Portfolio tracks pending stakes as a percentage of your bankroll. The default cap is 10%; go above it and the bar turns red. It stops a bad day wiping out weeks of progress, however good each individual bet looks.

Take a Break pauses every Execute and Commit button for 24 hours or 7 days, and you can end it early. If betting ever stops being fun, free confidential help is available at BeGambleAware.org.