Bad Runs, Sample Size and the Win Rate a Price Demands

Is sports betting profitable once the price is counted?

A two-way market priced 1.95 on both sides charges a fee the quote does not itemise. The sharp betting guides explain how prices like these are read; this page asks something narrower, which win rate that fee demands and how far chance pushes a record around it.

Betting is an adult activity (18+) and legal in some countries and not in others, so check yours before you open any account. This page shows how wide luck is and gives no reason to stake more; the responsible gambling page lists where to get help if betting has stopped feeling like a choice.

The win rate a bet needs to break even is one divided by its decimal odds: 52.36 percent at 1.91, 50.00 at 2.00 and 47.62 at 2.10 (a price above 2.00 appears only to show the direction). Say you win exactly half of your bets at 1.95, short of its break-even rate. You lose 2.5 units per 100 staked on average, because 0.5 x 1.95 - 1 = -0.025, and that gap is the cost of playing.

What does that cost look like when someone measures it? In a Data Golf analysis of golf matchup odds with no named author, a bettor who backed every outcome at Pinnacle lost 3.31 percent on average across 11,434 bets, in a database covering 2019 and 2020 (the database itself holds 106,204 matchups and 3-balls across 11 books). Data Golf counted the return from betting everything, so the figure measures what the prices cost, whatever the skill of the bettor. It is one book, one sport and one period. Data Golf's own note puts Pinnacle's typical margin in that sample at about 3.7 percent, and explains the gap by ties that Pinnacle voids but the average still counts. The same post reports about -7 percent at three of the six other books, part of it from bet types Pinnacle does not offer. Data Golf is a golf analytics business, so the post also works as a showcase for its models.

Apply that blind-betting cost to a single price of 1.95 and the win chance of each bet falls from 51.28 to 49.58 percent ((1 - 0.0331) / 1.95). The chart turns this into the chance of standing ahead after a given number of bets, next to a fair price and two assumed edges. It illustrates what a cost does to a record; it does not forecast a market.

Start with the zero-edge line. From about 100 bets on it hovers near half, so more bets widen the range while the chance of finishing ahead stays put. The cost line does the opposite. Each added bet pulls the centre further below zero while the spread grows only with the square root of the count, so the line sinks.

Data Golf ran its own simulation at Pinnacle's margin-free prices, and found 11.6 percent of random bettors ahead after 1,000 bets and 3.4 percent after 2,500. That is close to the curve here; the few points of difference come from Data Golf's varying odds and its 2,000 simulated bettors per sample size.

The assumed-edge lines run up instead. At +3 percent the chance of being ahead stays above 80 percent from about 790 bets on. At +1 percent it takes about 6,550 bets to reach 80, and one series in five is still behind there. The catch sits in the word assumed: an edge is exactly what a record cannot show you.

Can you run your own odds and number of bets?

Three fields, all pre-filled: the decimal odds, the number of bets, and your real edge per unit staked, in percent. The edge starts at zero on purpose, because that shows the width of luck at a fair price before any story about skill is added. A negative edge is the cost of a blind bet, so typing -3.31 at 1.95 reproduces the chart's cost line at a length you choose. The simulator cannot tell you whether you hold a positive one.

The first result line turns your edge into a win chance and sets it against the break-even rate for those odds. The expected result is the edge times the number of bets. The spread is one standard deviation of the final result, and the range below it holds 95 percent of runs, taken from exact binomial probabilities instead of a bell curve. On the behind, level and ahead line, level appears only where a whole-unit result can land on zero, as at odds of 2.00. The last two lines are about streaks: the longest run of losses inside the series on average, and the chance of at least 8, 10 and 12 in a row.

How long can a bad run last?

At its default inputs the simulator reports the longest run of losses inside 1,000 bets, with the price staying fair in every series.

Pinnacle's article on staking systems, by Joseph Buchdahl on 22 February 2016, gives a formula that puts the typical longest streak at about ten at fair odds of 2.00. An exact count at those odds gives 9.3 on average over 1,000 bets, with ten or more in 38.5 percent of series, although ten in a row starting at one given bet has a chance of only 0.098 percent. The formula is an approximation, and the simulator uses the exact count; that is why its figures at 1.95 differ from these.

A streak is one way to sit in a hole; a drop from a peak is another. In a simulation run for this page, with 20,000 series of 1,000 bets, a fixed seed and a true win chance of 50 percent at 1.95 (which costs 2.5 percent per bet), the median worst drop was 47.4 units, and one series in twenty fell by 88.4 or more. Remove the cost and the median worst drop is still 34.8 units, with 68.4 at the 95th percentile. The simulator does not draw drops from a peak, so these figures come from that separate run.

Pinnacle's bankroll article by Jonathon Brycki, dated 15 November 2018, adds a real edge to the picture: at odds of 2.0, an edge of 4 percent and 1,000 bets, 10,000 simulations gave a 17.4 percent chance of touching a 20-unit loss at some point, and 2.8 percent of finishing there. How many of your own bets would it take to tell which of those worlds you are in?

What can a record prove?

Pinnacle's own betting resources say that most of what happens in a betting market is chance, and that once the margin is counted most bettors will likely end up in the red in the long run according to Joseph Buchdahl on 13 September 2017. In the same article 249 bettors with more than 1,000 bets each had their histories cut in half. The first half explained 0.19 percent of the variation in the second (R squared 0.0019), and the author concluded that most of those bettors were merely lucky or unlucky. That does not mean a skilled bettor cannot exist.

Short records deserve the same caution. Buchdahl wrote on 19 January 2023 that a bettor with zero edge at fair odds of 2.00 shows a return of 5 percent or better about 21 percent of the time after 250 bets, and under 1 percent after 2,500 (his curve says 21.5, an exact binomial count says 20.6). Run it the other way and a bettor with a real 5 percent edge at odds of 2 still has about a 6 percent chance of being behind after 1,000 bets, he wrote on 24 April 2023 (exact count 5.3). Closing line value is a separate test that looks at the price paid, and the closing line value calculator covers it.

The share of bettors who finish ahead, counted as people over a defined period, is not backed by any reliable public figure. Round numbers circulate on tipster and betting-site pages without a named study, and this page leaves them out.

What changes when betting is a profession?

Professional describes how someone uses betting, and carries no published rate. This page prints no income, win rate or return for professionals because reliable public sources do not give one, and individual bettors stay unnamed. What the sections above can say is how wide luck is around whatever an edge turns out to be.

The contract is the other half. Pinnacle's terms, version 3.5, have the customer state that they act "solely on your own behalf as a private individual in a personal capacity and not on behalf of another party or for any commercial purposes". BetInAsia (clause 3.9 in a 2024 copy of its terms) and MadMarket (clause 3.10) use almost the same words, and AsianConnect's terms allow use "strictly in his/her personal non-professional capacity" and prohibit use by a corporate player or for commercial purposes. Sportmarket's terms, searched for personal, private, commercial and professional use, contain such a clause nowhere; that describes the text and does not amount to permission for professional use.

Help pages that say professional bettors can use the service (BetInAsia) or that it works with sharp bettors (AsianConnect) sit beside those clauses, and marketing and contract disagree; the contract is the text each operator can point to. How winning accounts are treated in practice is covered on the page on bookmakers that accept winners.

Does arbitrage change the picture?

It swaps the question of edge for a question of execution: a leg that is refused, repriced or cancelled leaves an open position, which the simulator above does not model. What each operator writes about it is collected on the arbitrage betting page.