How to Find Positive EV Prices in NBA Markets

A price has positive expected value when the probability you can defend for an outcome is higher than the probability the price implies. At decimal odds d, the implied probability is 1 divided by d, and the expected return per unit staked is your probability times d, minus one.

That definition makes the job concrete. Finding positive EV in NBA markets is not about spotting a team that feels underrated. It is a measurement problem with three parts: estimate a fair probability, find the best price on offer, and check afterwards whether your estimates were any good. Each part can be done carefully or carelessly, and the arithmetic will produce a confident number either way.

The most defensible starting point is the market itself, with the margin removed. Take a two-way NBA market such as a spread priced at 1.91 on both sides (American -110). Each side implies about 52.4 percent, the two sum to about 104.7 percent, and the excess is the margin. Dividing each side by the total gives a fair probability of 50 percent each.

That is the multiplicative, or proportional, method. Other devigging methods exist: additive (subtract an equal share of the margin from each side), power (raise implied probabilities to a common exponent until they sum to one), and Shin's method, which models the margin as protection against informed bettors and tends to shift more of it onto longer prices. On near-even NBA spreads and totals the methods barely differ. On lopsided moneylines and player props they can disagree by a point or more of probability, which is often larger than the edge being claimed, so record which method produced each fair price.

Which market do you devig? The one that is hardest to beat. Books that take large limits and run low margins tend to move fastest on information, so their no-vig price is the usual reference. A fair price built from a slow, wide-margin book mostly measures that book's lag.

Where a model fits

Once you have a fair probability, every available price can be scored the same way: fair probability times that book's decimal price, minus one. Positive numbers are candidates. In practice, most candidates in main NBA markets are small, a percent or two, and come from books that have not yet moved after the reference market did.

Three things decide whether a candidate is real:

• Freshness. If the reference price is a minute old and the soft book's price is ten seconds old, the gap may be the reference catching up rather than the soft book lagging. Timestamp every quote and compare only quotes captured close together.
• Equivalence. The same game can be quoted as a moneyline, a spread at different numbers, or alternative lines. Compare like with like, or convert explicitly, rather than comparing -3.5 at one book with -4 at another as if they were the same bet.
• Limits. A price available only for a trivial stake tells you about that book's risk settings, not about a market inefficiency worth measuring.

Player availability dominates NBA pricing more than in most sports, because one player is a large share of a team's minutes and usage. Prices move sharply when the league's official injury report changes or when rest decisions are announced, and the period between a report and the market's full reaction is where most genuine price gaps appear. Back-to-backs and travel affect rotation decisions too, but those are scheduled and therefore usually priced early.

Player props behave differently from sides and totals. Margins are wider, limits are lower, books price them with different models, and a single minutes projection drives points, rebounds and assists together. That makes props the place where book-to-book disagreement is largest, and also where your own fair price is least certain, because there is no deep, sharp reference market to devig. Treat prop edges with wider error bars than spread edges.

Correlation matters when combining NBA markets. A player's points and the team total are not independent, and neither is a team's spread and a star's minutes. Multiplying independent probabilities across such legs misprices the combination.

Results are a slow and noisy judge. An NBA side at near-even odds with a two percent edge needs a very large sample before profit and loss separates skill from luck. Closing line value is the faster check: compare the price you recorded with the de-margined closing price of the same market. If your recorded prices consistently beat the close, your process is capturing information before the market finishes pricing it. If they do not, positive EV estimates at the time of recording were probably measurement error.

To make that comparison meaningful, define close precisely (last price before tip from a named source), devig both your price and the close the same way, and record the size of the difference rather than only whether it was positive.

Suppose the reference book prices an NBA total at 1.95 over and 1.87 under. Implied probabilities are about 51.3 and 53.5 percent, summing to 104.8 percent. Proportional devigging gives a fair over probability of about 48.9 percent and a fair under of about 51.1 percent.

Another book offers the under at 2.02. Expected value is 0.511 times 2.02, minus one, or about plus 3.1 percent. Before treating that as real, check that both quotes were captured within seconds of each other, that both books use the same total, and how the under closed. These numbers are illustrative, not observed prices.

• Using a book's own implied probability as the fair price and finding zero or negative EV everywhere, by construction.
• Devigging a slow book and calling its lag an edge.
• Ignoring that different devig methods disagree most on the lopsided prices where edges look largest.
• Judging the process on a few weeks of results instead of on calibration and closing line value.
• Sizing stakes as if estimated edges were exact, when they carry an error at least as large as the edge itself.

Keep sessions within limits you have set in advance; this page describes a measurement method and is not betting advice.