How OddWizard makes its predictions

OddWizard makes its own predictions

For games, OddWizard builds its own chances from results history, not from betting odds. Each league has a model that learns team strength from every game ever played (margin of victory, home advantage, strength carried between seasons), then learns how much rest, back-to-backs, neutral sites, division games and quarterback changes are worth. It is trained walk-forward: every test prediction is made only from games before it, so its accuracy figures are honest out-of-sample numbers.

Every night it adds the day's results and retrains. Every prediction is logged before the game next to the market's price for the same game, and both are graded afterwards. That public head-to-head is the only fair way to say whether OddWizard is more accurate than the market, so it's on the track record page for every league.

Sportsbook prices are used for one thing: to see whether a book is paying more than OddWizard's chance is worth. A pick is labelled market agrees when the sharp market also rates that price as good, or OddWizard only when it doesn't. OddWizard suggests stakes on its own-only picks in a league only after its model has beaten the market there over 300+ graded games.

1. The market's view, used as a cross-check

Some bookmakers take large bets from professional bettors and move their lines whenever those bettors disagree. Their prices end up very close to the true chance of each outcome, and research consistently finds their closing lines are the best single predictor of game results.

Every bookmaker builds a margin (the "vig") into its odds, so the implied chances add up to more than 100%. OddWizard removes it:

When no sharp book lists a game, OddWizard uses the median no-vig price across at least three books instead, and says so.

2. Compare it with every price you could actually get

If a licensed sportsbook you use pays more than the fair price, the bet has positive expected value (EV):

For example, a fair chance of 50.6% at +105 (decimal 2.05) gives EV = 0.506 × 2.05 − 1 = +3.7%. Over many bets of that kind, you'd expect to make about 3.7 cents per dollar staked. OddWizard only shows edges above your minimum (2% by default) and flags anything unusually large, since that usually means a stale line about to move.

Arbitrage is the rare case where the best prices across books add up to less than 100%. Betting both sides in the right proportions then locks in a small profit, until a book limits you.

3. Use its own models where markets are thin

• Stocks, crypto, commodities and currencies: 50 years of daily history, simulated forward from today's level to estimate the chance a price crosses your threshold by your date. Recent decades count double.
• Interest rates and the money market: Treasury bill rates, SOFR and other public-domain series, modeled the same way.
• Games: team ratings built from results history, used as a check when sportsbooks disagree.

4. Correct for known biases

Markets have habits. Cheap prediction-market contracts win less often than their price implies (the favorite-longshot bias), and OddWizard adjusts for that. Its own models are corrected the same way: every forecast is graded when the result is known, and a small model trained only on those grades re-calibrates the next forecast. It only switches on once it has enough examples and actually beats the raw forecasts.

Breaking news, read by a small AI, and learned from

Every 10 minutes a small AI reads the major free sports-news feeds and pulls out the facts that can swing a game: who's out, who's back, who's suspended, who's starting, a coaching change, a game moved for weather. Those facts appear next to the picks for both teams. When several outlets report the same thing, OddWizard counts the confirmations. If major news lands after a game was priced, the pick is flagged and its suggested stake is removed until the line is re-checked.

OddWizard also learns from the news. After every game it compares what the news said beforehand with what actually happened, and learns how much each kind of news (a star out, a starter returning, a coaching change) really moves results beyond what team strength already explains. In each league, news starts changing OddWizard's numbers only once news-adjusted predictions have beaten the plain model on at least 150 graded games. The AI only extracts facts; the predictions come from OddWizard's own models.

5. Size bets conservatively

The Kelly formula gives the stake that grows a bankroll fastest when the edge is known exactly. Edges are never known exactly, so OddWizard suggests a quarter of Kelly, capped at 2% of the bankroll you enter:

Suggested stakes apply only to sportsbook bets, using your own numbers. Stock, rate and prediction-market figures are the same for every member, and you decide what to do with them.

6. Grade itself in public

Every edge is logged as a paper bet the moment it's flagged, before the event, and graded afterwards. The track record shows all of them, losses included. It also tracks closing-line value (CLV): whether the price you got beat the fair price just before the game started.

CLV settles down after dozens of bets, long before profit and loss do. Consistently positive CLV is the best early sign that a method beats the market.

What it can't do

Where the data comes from

Sportsbook odds from a licensed odds feed. Interest rates and economic data from public-domain US government series (for example Treasury and Federal Reserve data via FRED). Copyrighted data series are never shown without the owner's permission. Each forecast lists its sources.