Football is a fairly random game, no matter how good your model is
A cada quatro anos, o mundo reúne seus profetas — algoritmos, animais, astrólogos e economistas — para tentar decifrar o que o futebol se recusa a revelar com antecedência. Na Copa do Mundo de 2022 no Catar, bilhões de espectadores e dezenas de métodos de previsão convergem para a mesma pergunta sem resposta garantida: quem levantará a taça em 18 de dezembro? O que une todos esses esforços não é a certeza, mas a humildade velada diante de um esporte que reserva 55% de seus resultados ao acaso.
- Cinco bilhões de pessoas assistirão ao torneio, e a pressão por previsões confiáveis mobiliza desde supercomputadores do Alan Turing Institute até polvos alemães e videntes brasileiros.
- Mesmo os modelos mais sofisticados — usados por Goldman Sachs, UBS e ING — erraram os vencedores dos dois últimos mundiais, expondo os limites da ciência de dados diante da imprevisibilidade do futebol.
- A derrota da Argentina para a Arábia Saudita na fase de grupos abalou as previsões logo de início, lembrando que nenhum método isolado é suficiente para capturar a volatilidade do torneio.
- Casas de apostas, padrões históricos e estatísticas de clubes como Bayern de Munique e Inter de Milão oferecem fragmentos de orientação, mas nenhum mapa completo para o resultado final.
Cinco bilhões de pessoas devem assistir à Copa do Mundo no Catar — e quase todas têm um palpite sobre quem vai ganhar. Para dar forma científica a esses palpites, o Alan Turing Institute rodou 100 mil simulações computadorizadas dos 64 jogos. O Brasil emergiu como favorito, aparecendo como campeão em cerca de um a cada quatro simulações, seguido por Bélgica, Argentina e França. Mas o próprio instituto advertiu: ninguém deveria apostar dinheiro nisso. O futebol, disseram, é um jogo em grande parte aleatório.
A advertência tem respaldo histórico. Goldman Sachs, UBS e ING erraram os vencedores dos dois últimos torneios. Até Joachim Klement, estrategista londrino que acertou Alemanha em 2014 e França em 2018, admite que seu modelo explica apenas 45% das chances de vitória de uma equipe. Os outros 55% são pura sorte — uma concessão notável vinda de alguns dos analistas mais sofisticados do mundo.
Os animais também foram consultados. O polvo alemão Paul ficou famoso em 2010 ao acertar 12 de 14 jogos, incluindo o título espanhol. Pandas, alpacas, furões e camelos seguiram seus passos como oráculos improváveis. Cientistas suspeitam que Paul simplesmente preferia listras horizontais a verticais — o que explicaria sua aparente clarividência sem recorrer ao sobrenatural. Ele morreu poucos meses depois de sua hora de glória.
As casas de apostas ocupam um território ambíguo: não são neutras, pois lucram quando os apostadores perdem, mas têm tanto dinheiro em jogo que seus cálculos refletem avaliações de probabilidade genuínas. O Brasil aparece como favorito a 3 para 1 em alguns sites; o Catar, como anfitrião, está cotado a 1.000 para 1.
A história oferece seu próprio tipo de previsão. Apenas oito nações já venceram o Mundial em 21 edições. Vitórias consecutivas são raríssimas — o Brasil de 1962 foi o último a consegui-las. Desde 1982, ao menos um jogador do Bayern de Munique ou da Inter de Milão esteve em todas as finais. Com 17 jogadores no torneio, o Bayern está bem representado; a Inter tem seis, incluindo Lautaro Martínez. Para os torcedores argentinos ainda atordoados com a derrota para a Arábia Saudita, esse pequeno padrão histórico pode oferecer um fio de esperança. Mas esperança, como previsão, é uma coisa frágil no futebol.
Five billion people are expected to watch the World Cup in Qatar. That's five billion chances to be right, or wrong, about who will lift the trophy on December 18th. And so, in the weeks before the tournament, the world's forecasters—the serious ones and the strange ones—have begun their work.
They come from everywhere. The Alan Turing Institute, a major British research center, ran 100,000 computer simulations of all 64 matches, feeding algorithms historical data and statistical patterns. Brazil emerged as the favorite, appearing as champion in roughly one of every four simulations, trailed by Belgium, Argentina, and France. The institute was careful to note that no one should actually bet money on these predictions. "No matter how good your model is," they said, "football is a fairly random game."
They have a point. Goldman Sachs, UBS, and ING—some of the world's largest financial institutions—all got the winners wrong in the last two tournaments. Even Joachim Klement, a strategist at London-based Liberium Capital whose algorithm correctly predicted both 2014 (Germany) and 2018 (France), admits that his model accounts for only 45 percent of a team's chances of winning. The other 55 percent is pure luck. The irony is that some of the most sophisticated minds in data science and finance have essentially conceded that they're guessing.
Then there are the animals. A German octopus named Paul became famous in 2010 by correctly predicting 12 of 14 matches, including Spain's tournament victory. He would be placed before two boxes decorated with the flags of opposing nations, and he would choose the "correct" box with remarkable consistency. Since then, pandas, alpacas, ferrets, and camels have all been consulted as oracles. Scientists, however, are skeptical. They suspect Paul may have simply been more attracted to horizontal stripes than vertical ones—a preference that would explain his apparent clairvoyance without invoking the paranormal. Paul died a few months after his moment of glory.
Brazil's psychic, Athos Salomé, claims to have predicted the COVID-19 pandemic and Russia's invasion of Ukraine. Getting World Cup matches right, he suggests, would validate his powers. There are also the round-table discussions that media outlets worldwide host with players and coaches before the tournament begins. They are frequently wrong.
Betting houses occupy a strange middle ground. They are not neutral forecasters—they profit when bettors lose. Yet Robert Simmons, an economics professor at Lancaster University, argues they remain "well-positioned" sources of information precisely because they have so much money at stake. If they get the odds wrong, they pay dearly. Brazil appears as a 3-to-1 favorite on some sites, while Qatar, the host nation, is listed at 1,000-to-1. These odds are calibrated to ensure the house makes money, but they reflect genuine probability assessments nonetheless.
History offers its own kind of prediction. Only eight nations have ever won the World Cup across 21 tournaments: Brazil five times; Italy and Germany four times each; France, Uruguay, and Argentina twice; England and Spain once. Back-to-back victories are extraordinarily rare. Brazil managed it in 1962 and has been the only nation to do so since the tournament began in 1930. France, the current champion, would need to defy a century of precedent to repeat. Since 1978, only three World Cup finals have featured a team winning their first title. No African nation has ever reached the quarterfinals. South Korea remains the only Asian team to reach the semifinals, which they did in 2002. Spain is the only team to have won the tournament after losing their opening match—they did it in 2010.
There is one more pattern worth noting. Since 1982, at least one player from Bayern Munich or Inter Milan has appeared in every World Cup final. Bayern has 17 players at this tournament across various national teams. Inter has six, including Argentina's Lautaro Martínez, one of the world's best strikers. For Argentine fans still reeling from their shock loss to Saudi Arabia in the group stage, this small historical pattern might offer a sliver of hope. But hope, like prediction, is a fragile thing in football.
Citações Notáveis
No matter how good your model is, football is a fairly random game.— Alan Turing Institute
My model determines only 45% of a team's chance to win the tournament; the remaining 55% is pure luck.— Joachim Klement, Liberium Capital
A Conversa do Hearth Outra perspectiva sobre a história
Why do we keep trying to predict something that's fundamentally unpredictable?
Because we're drawn to the illusion of control. A betting house makes money whether you win or lose, but you play anyway. A computer runs a million simulations and admits luck matters more than the algorithm—and you still want to know what it says.
But some predictions work. That octopus got it right. That one algorithm predicted two winners in a row.
Paul the octopus probably just liked certain stripe patterns. And Klement's algorithm? He's honest that it might have been luck. We remember the hits and forget the misses. That's how prediction feels true even when it isn't.
So what's the point of any of this?
The point isn't accuracy. It's the conversation itself. A banker wants to prove their model works. A psychic wants to validate their powers. A friend wants an excuse to talk about the game. The World Cup is so big, so watched, that everyone gets a turn at being right or wrong.
Is there anything that actually predicts winners?
History does, a little. Only eight nations have ever won. Back-to-back champions are almost unheard of. But then Argentina loses to Saudi Arabia, and Germany loses to Japan, and you realize that history is just a pattern waiting to be broken.
What about the betting houses?
They're the closest thing to honest forecasters because they have money on the line. But they're not trying to be right—they're trying to make sure they win either way. That's actually more reliable than someone who desperately wants to be correct.