Among thousands of tipsters, chance always produces a few with spectacular months, and those are the ones advertised. Follow the best of one season and in the next they perform like anyone else: they lose the bookmaker’s fee. Telling skill from luck takes thousands of bets.
Read in
Chance manufactures experts
Picture a thousand people tipping matches with no knowledge of football, tossing a coin. By the end of the season, some will have got well over half right. If one of them opens a tips channel and shows their results, they will look like an expert. The other nine hundred or so open nothing.
That is what the simulation shows. None of these thousand tipsters has any skill, and still the best of one season have spectacular numbers. The question is what happens the next season.
The best of one season, in the next
A thousand tipsters with no skill at all. Ranked on season one, then watched in season two.
The top 10 of 1,000 tipsters who pick at random
Rank
Hits
Season 1
Season 2
1
65/100
+23.5%
-8.8%
2
65/100
+23.5%
-5.0%
3
65/100
+23.5%
-5.0%
4
64/100
+21.6%
+4.5%
5
64/100
+21.6%
-8.8%
6
64/100
+21.6%
+0.7%
7
63/100
+19.7%
-8.8%
8
62/100
+17.8%
+2.6%
9
62/100
+17.8%
+8.3%
10
62/100
+17.8%
+2.6%
Average of the top 10
+20.8%
-1.8%
End season 1 in profit
32%
Average of everyone
-5.1%
Average of the top 10 · Season 2
-1.8%
None of these tipsters knows anything: every tip is a coin toss at odds of 1.90. Even so, the top ten of season one look like experts, with an average return of +20.8%. The next season, those same ten get -1.8%: the bookmaker’s fee. What gets advertised is season one.Every tip wins with p = 0.5 at odds of 1.90 (a two-way market with a 5.3% margin), so the expected ROI is -5.0% for everyone. Ranking on season one selects the right tail of the noise; season two is independent and falls back to the mean: regression to the mean.
What it takes to know
Telling skill from luck takes a huge number of bets. With a real 3% edge at odds around 2, it takes about four thousand bets before the record proves it with reasonable confidence. We go through it with a calculator in Can you make a living from sports betting?
Each tipster’s ROI over a season is noise around −5.3%. Keeping the top ten out of a thousand selects the right tail of that noise. Season two is independent of season one, so its mean falls back to −5.3%. It is the same mechanism that means the best investment fund of one year rarely repeats.
Related questions
How can you tell if a tipster is really good?
You need their complete record, published before each match and verified by a third party, with thousands of tips, not dozens. Look at return on everything staked, not hit rate, and be wary of screenshots.
How do tipsters make money?
Many from subscriptions to paid groups, others as bookmaker affiliates, often paid a share of what the players they refer lose. Either way they earn whether or not their tips win.
What about AI predictions?
A model does not change the bookmaker’s fee. We tested machine learning models on over 200,000 matches: they get close to the accuracy of the odds but do not beat it.