What is adverse selection?

Adverse selection occurs in markets where buyers and sellers have different levels of information about the good they are trading.

The concept was famously outlined in Akerlof’s “The Market for Lemons” which focused on the market for second hand cars. Akerlof found that prospective buyers could not distinguish between high-quality cars, known as “peaches”, and poor quality “lemons”.

If buyers are uninformed about the quality level of a given car compared to the car’s current owner, they will be unwilling to pay a fair price for a peach.

This drives the market price down to below the level required for willing sellers of peaches. As a result, only lemons are willing sold by their owners. This can lead to market collapse.

Adverse selection in betting markets

Asymmetric information is at the heart of every betting market. By their very nature the bookmaker generally knows more about the probability of a given event than the bettor.

This is because the bookmaker has access to all the information available on the market from their customers, and can therefore see the full picture. The bettor usually has access to less information of inferior quality.

The bettor faces what is known as an adverse selection problem since he has an information disadvantage compared to the bookmaker. The seller (bookmaker) has the full picture and a very good idea of what the bet is worth. A bet at this stage is no different to buying one of Akerlof’s lemons.

Why do betting markets still exist?

In Akerlof’s scenario he posited that the reluctance of peach sellers would cause buyers to lose trust in the market and could eventually cause complete market collapse. Since the betting market has a similarly unbalanced informational structure, why do bettors simply not stop placing bets?

Some solutions to this suggest that answer is simply that betting is an irrational pursuit, bettors are overconfident or that the enjoyment provided by betting outweighs a bettors losses.

However, we do know that some bettors are profitable, so it is likely that the enticement of finding a peach amongst the lemons keeps bettors coming back.

Application of adverse selection to betting

There are certainly applications here for bettors aware that they are working in a market for lemons. Before placing a bet think “If I were buying a used car what would I consider?”.

Alternatively consider this. Why should a buyer wish to buy a used car if the seller wants so much to sell it at that price? After all the seller knows more about the car. In the same way ask yourself why you are “buying” a bet that a bookmaker, with access to all that information, would want to sell to you.

Do you have enough of an informational edge to overcome this disadvantage? If not, reconsider why you are placing the bet in the first place.

Are tipsters selling lemons or peaches?

In a market of lemons, the spotting a rare peach is a much sought-after skill.

“The demand for accurate predictions is insatiable whilst reliable suppliers are few and far between. The gap between demand and supply creates opportunities for unscrupulous suppliers to fill the void by gulling desperate customers into thinking they are getting something no one else knows how to provide.”

Within the betting market this allows tipsters of varying levels of reliability to offer their services. In a similar situation to Akerlof’s used car market, it could be argued that the bad tipsters are liable to drive out good ones by lowering the overall quality of predictions available within the market, eventually causing all predictors to be perceived as unreliable.

This is why analysis of a tipster’s record is so important. The used car market may not function efficiently without standards checks and consumer protection, often in the form of “Lemon laws”.

A good tipster needs to reduce the asymmetry between themselves and bettors by being open about their profitability in order to allow consumers to identify reliable peaches accordingly. Anything less than this suggests they may well be providing lemons to the market.

“Some of it you learn the hard way

Some of it you read on a page

Some of it comes from heartbreak

Most of it comes with age

And none of it ever comes easy

A bunch of it you maybe can’t use

I know I don’t probably know what I think I do

But there’s something to some of it”

As I drove across Canada a week ago to my mother’s 60th birthday, the new song by Eric Church continuously came on the radio. The song, fittingly about life lessons, got me thinking as I cruised through the mountains. What if I got the chance to start over and go back to the Dominican Republic with $2,000 in my pocket to see how long I could last betting on sports? Where did I go wrong and what would I do differently?

If you are a new bettor, I hope these can help. If you are an experienced bettor, I am sure you can relate.

Sports betting never ends

When I made my first deposit into an online sportsbook at the age of 15, I was obsessed with turning the $100 into $1,000,000. I lived and died with every $10 wager I made and had a goal to win money every day. It took me a number of years to fully understand that sports betting never ends.

There is always a chance that the opportunity tomorrow is better than any opportunity today. The only thing preventing bettors from taking advantage of those future opportunities is forcing wagers and losing money on bad decisions today.

‘Short term success does not matter’ is unfortunately interpreted too frequently as an excuse, instead of sports betting truth.

Nobody controls the outcome

I was fortunate enough to have someone send me an email I sent them analyzing a football game in 2010.

Two sentences in, I wanted to go back in time and smack myself in the face. The entire premise of the analysis was based upon statistics that “I” found and research that “I” did.

It became clear to me that bettors have an edge in passing on bets. Selectivity and patience are two key skills to master

I was completely convinced – and did my best to convince others – that because I was betting on a game it had a better chance to win. It humbles me to say that upon starting out, I believe that I had control on the outcome. As a hyper competitive person, this was a specifically hard lesson to accept as it forced me to put my ego aside. It was not until I did, that I was able to learn and become better.

Saying no is often the biggest edge

This is not a lesson I understood until my second year of setting prices as a bookmaker. Account balances were up, company revenue was down, and bettors were picking apart soft prices in the middle of the summer.

I was overwhelmed by continuously having to open up full game, first half and secondary markets for all games on the board each day.

Fast forward years later and reverse roles, it became clear to me that bettors have an edge in passing on bets. Selectivity and patience are two key skills to master.

Prices not predictions

I got into sports betting mainly due in part to my love of sports. The competitive nature of playing sports growing up fuelled this idea when I started betting that I was smarter than everyone else and knew more than the betting market.

Had I focused more on prices than predictions in the early stages of my career, I have little doubt my arc of success in the industry would have been accelerated dramatically.

Separate bankroll from life

I have a vivid image in my head of a blue stack of $2,000 Dominican Peso bills on my night stand with betting tickets wedged in-between. When I first started out, I would live life and bet out of the same bankroll.

Looking back, the toll that this took on my mental well-being was astonishing.

My lifestyle outside of betting was entirely predicated by the success of my wagers. It is impossible to estimate how much being emotionally connected to every wager cost me early in my lifetime, but it is an enormous amount.

Separating betting resources from life resources is critical for success.

So, what if I started over?

As tempting as it would be to go back to walking around Punta Cana with a wad of pesos in my pocket boasting about how right I was about the football game that just finished, I am glad I made it through those times somewhat unscathed. There is a ton of truth in how enjoyable betting with emotional investment and living for tonight is, but I much prefer what I have settled into now.

I can say with full confidence that the success that comes from separating life from betting, focusing on the long term, buying good prices consistently and detaching myself from the result, is far more enjoyable than the alternative.

None of it ever comes easy, but you will find the same success if you can get there too.

A close look at the Chelsea vs. Arsenal odds

The odds imply that the Europa League trophy is most likely to leave to Baku with the Blues, with Chelsea’s odds of 1.751* giving them around a 57% chance of victory by the end of the night. While Arsenal’s odds of 2.179* currently give them a 45.89% chance of winning.

The Over/Under is set at 2.5, with Over odds of 1.961* and Under odds of 1.917* – suggesting a low scoring game.

Time and date: Wednesday May 29, 20.00 CET kickoff

Venue: Baku Olympic Stadium

Team news and predicted lineups

Chelsea predicted lineup

4-3-3: Kepa; Azpilicueta, Christensen, Luiz, Emerson; Jorginho, Kante, Kovacic; Pedro, Giroud, Hazard

Chelsea team news

Blues manager Maurizio Sarri is optimistic midfielder N’Golo Kante will be fit in time for Wednesday’s final – with the player having returned to training following a hamstring injury suffered on May 5.

Meanwhile defender Antonio Rudiger and forward Callum Hudson-Odoi are certain absentees – alongside midfielder Ruben Loftos-Cheek who ruptured his Achilles’ tendon in a US friendly game against New England Revolution.

Arsenal predicted lineup

3-4-1-2: Cech; Sokratis, Koscielny, Monreal; Maitland-Niles, Torreira, Xhaka, Kolasinac; Ozil; Aubameyang, Lacazette

Arsenal team news

Much of the pre-match build for the final is the news that Arsenal’s Armenian midfielder Henrikh Mkhitaryan will not travel to Baku over fears for his own safety – due to political tensions between Azerbaijan and Armenia.

Elsewhere, it’s suggested injury-plagued striker Danny Welbeck, who hasn’t played since breaking his ankle against Sporting Lisbon in November, could make the bench in what would be a farewell appearance – should he play a part.

Head-to-Head history and results

• Chelsea and Arsenal have met twice before in a UEFA competition, with a two-legged Champions League Quarter final in 2003/04 ending 3-2 to the Blues on aggregate.
• In all time domestic meetings, Chelsea have 52 wins to Arsenal’s 63, with the teams sharing 49 draws.
• In this season’s Premier League, both teams won at home. Arsenal winning 2-0 at the Emirates in January, with Chelsea winning 3-2 at Stamford Bridge in August.
• Across all competitions, recent results between the two sides have favoured Arsenal.
• Since 2016/17, the two teams have met 10 times. The results have been Chelsea 2, Arsenal 5, and Draw 3

Chelsea’s 2018/19 Europa League statistics

Arsenal’s 2018/19 Europa League statistics

Inform your Chelsea vs. Arsenal Prediction

Separated in the end by just two points in the Premier League, Chelsea and Arsenal could both be considered to have had mixed seasons – with new managers Maurizio Sarri and Unai Emery drawing almost equal parts praise and criticism throughout the course of the season.

For Sarri, a third place finish behind the exceptional Manchester City and Liverpool can itself be considered an achievement – particularly given Tottenham’s occupation of third for much of the second half of the season.

Losing to Manchester City on penalties in the League Cup in March, as another achievement of sorts, this was perhaps the defining image of Chelsea’s season came during that game. When in extra time, an injured-looking Kepa Arrizabalaga defied his manager’s orders to be substituted and remained on the pitch.

Symbolic of the rumoured fractured relationship between Sarri and some of his players, victory in the Europa League final would cement this season as being one of outright success for Chelsea – though it may not even be enough to keep Sarri in his job.

In many ways, Wednesday’s final holds much bigger significance for Emery’s Arsenal – given they, unlike Chelsea, failed to qualify for the Champions League via the Premier League.

Finishing just a point behind bitter North London rivals Tottenham in fourth, Arsenal missed a series of chances to go above Spurs in the closing quarters of the season – losing three successive league games to Crystal Palace, Wolves, and Leicester in April.

Going unbeaten for 22 games in all competitions between August and December, Arsenal’s form and the perception of Emery’s tactical prowess cooled post-Christmas – though their return of 70 points in the Premier League was a marked improvement from 2017/18’s tally of 63.

While Chelsea won the Europa League in 2013, and the Champions League in 2012, Arsenal haven’t won a European trophy since 1993/94 when they beat Parma to win the now-defunct European Cup Winners’ Cup.

Should Emery guide Arsenal to victory, the perception of their season would change dramatically – with Champions League qualification also placing the Gunners in a much better position to attract high-calibre players in the summer transfer window.

Looking at both teams’ respective routes to the final, Chelsea breezed past Group L – winning five and drawing once against BATE Borisov, MOL Vidi, and PAOK.

Overturning Malmo FF 5-1 on aggregate in the round of 32, the Blues smashed Dynamo Kyiv 8-0 over two legs in the round of 16 – before edging past Slavia Prague 5-3 on aggregate in the Quarter-Finals.

However, their semi-final opponents Eintracht Frankfurt proved a much more even test – with Chelsea eventually seeing them off 4-3 on penalties, with Eden Hazard scoring the vital spot kick.

With both ties ending 1-1, the combined expected goals for the two-legs was 1.9 – 1.9, proving the lottery of the penalty shootout really was all that separated the two teams.

Looking at the first leg, which had an expected goals tally of Eintracht Frankfurt 0.4 and Chelsea 1.2, the Blues dominated midfield but kept coming short of creating the number of clear chances you’d expect from their overall dominance.

In the return leg, Eintracht Frankfurt had two shots cleared off the line in extra time – against a sloppy Chelsea team. In terms of expected goals, Chelsea recorded 0.7 and Eintracht Frankfurt recorded 1.5.

Like Chelsea, Arsenal also had an unbeaten group stage – winning five and drawing one game alongside Sporting CP, Vorskla Poltava and Qarabag in Group E.

Recording a 3-1 aggregate victory over Zurich in the round of 32, Arsenal overturned a 3-1 loss away at Rennes, to win the second leg 3-0 in the round of 14 – sending them through 4-3 on aggregate.

Facing a real test against a good Napoli side in the quarter finals, Arsenal surprised many by keeping a clean sheet over the two legs, winning 3-0 on aggregate.

Drawing with Valencia in the semi-finals, who had dropped down from the Champions League, a 3-1 victory at home and an incredible 4-2 victory gave Arsenal a 7-3 aggregate victory.

Analysing the underlying performances of both teams by looking at their expected goal stats, the first leg was actually a much more even game than suggested – with Arsenal ending the game with 2.6 expected goals compared to Valencia’s 2.1.

While Lacazette (2) and Aubameyang were clinical for Arsenal, Valencia missed two big chances early on,, which could have changed the entire complexion of the game.

Scoring all seven goals of the two legs between them – including an Aubameyang hat trick in the away leg, Arsenal were utterly reliant on Lacazette and Aubameyang – with the attacking duo effectively carrying their teammates to the final.

With a second leg expected goals map reading Valencia 2.0 Arsenal 1.0, Arsenal’s shaky defence can count themselves lucky that they play alongside two top-class strikers.

With combining both legs, the expected goals map reads Arsenal 3.8 Valencia 4.2 – proving poor finishing by the Spanish side and exceptional finishing from Lacazette and Aubameyang was the main difference between the two sides.

Key battle: Eden Hazard vs. Arsenal defence

With Arsenal’s reliance on Lacazette and Aubameyang in the latter stage of the competition being well-documented, Chelsea have actually used their undoubted best player, Eden Hazard, rather sparingly in contrast.

Playing in just seven of Chelsea’s 14 games en route to the final, Hazard spent a total of just 362 minutes on field out of Chelsea’s 1,290 minutes in the competition, or just 28% of the total time.

Making just three starts, the only game Hazard completed was the 120 minute semi-final second leg vs. Frankfurt – in which of course he scored the divisive spot kick.

Often rested by Sarri to keep him fresh for the Premier League, Hazard’s Europa League return of one assist and no goals is not reflective of his actual form or quality – and Chelsea have likely underperformed in the competition in his planned absence.

With the Europa League final almost certainly his last game for Chelsea, amid intense rumours of a summer transfer to Real Madrid, Hazard could provide Chelsea with the perfect parting gift.

Scoring 16 goals and providing 15 assists in his 37 Premier League games – Hazard had an expected goals tally of 12.30, though he has surpassed this number in each of his last five seasons. This is a sign his over-performance is due to genuine ability – rather than luck or poor defending.

Scoring 54 of his past 62 league goals from inside the box, Hazard has netted just five times from outside the box in his past five seasons – a stat which could influence how Arsenal’s defence may choose to combat the Belgium.

With an average Europa League Expected Goals Against figure of 1.6, compared to the actual figure 1.4, Arsenal’s defence has benefitted from poor opposition finishing and moments of luck in the competition – something they can’t retaliate to when coming up against Hazard.

Chelsea vs. Arsenal: Where is the value?

Given the formidable attacking options of both sides – as well as some suspect defending – bettors may be interested in considering the Over odds of 1.961*.

While common wisdom will often state that cup finals, particularly in Europe, are tight affairs, the research actually shows that three of the last five Europa League finals have been decided by more than one goal – with the market often underestimating the probability of more goals than average.

This season, the two games between Chelsea and Arsenal featured an average of 3.5 goals – compared to the league average of 2.82. Interestingly, the Europa League goals per game average is just 2.75 – with one goal scored, on average, every 32 minutes.

Potentially suggesting the London sides are actually more attack-minded than their European counterparts.

What factors should bettors consider?

The competition match-ups are contested over 50 overs, each team must use a minimum of five bowlers, and those bowlers are allowed to bowl a maximum of 10 overs each. The team who scores the most runs in the 50 over period will be deemed the winner.

Cricket has many factors to consider before placing a bet. The different rules for each format will influence betting decisions – something that is explained in Pinnacle’s How to Bet on Cricket guide. This article takes a look at areas bettors should think about before placing a bet on The 2019 Cricket World Cup.

Cricket World Cup team guides

Afghanistan

Outright odds: 95.32*

Afghanistan are a side that comes into the 2019 World Cup with plenty of potential. Their team includes players such as Mohammad Nabi, Rashid Khan, and Mujeeb Ur Rahman, who bettors will be aware of due to the T20 leagues. It is very unlikely they have the team to go all the way, but it wouldn’t be a surprise to see them pull off an upset somewhere in the tournament.

It has been an encouraging build up to the World Cup for the outsiders, with wins over Sri Lanka and Bangladesh whilst narrowly losing by three wickets to Pakistan and a draw with India showing that they cannot be overlooked, even when facing some of the tournament favourites.

Most recent 2019 results: LWLWTLLWW

Key player: Rashid Khan

England

Outright odds: 3.16*

England are the team in form coming into the World Cup and are justified favourites to lift the trophy. The edge in the outright betting market is due to the ease in which they have dispatched their opponents in the lead up to the tournament, and their top order in the batting is extremely powerful, whilst possessing real variation with the ball.

The host nation are positioned in the best possible position to win the trophy and it is likely the best opportunity they have had for many years to be crowned champions. They have made three late changes, bringing in pace bowler Jofra Archer, all-rounder Tom Curran and off-spinner Liam Dawson, which looks a positive move. Their squad for the tournament certainly has the ability go all the way.

Most recent 2019 results: WWWLWLWW

Key player: Jonny Bairstow

Sri Lanka

Outright odds: 30.59*

It is a challenging time for Sri Lankan cricket, and they come into the tournament as one of the main outsiders in the betting. They have just one win in their last ten matches and are a team that is constantly rotated with very little consistency in the squad.

There was no room in their squad selection for Dinesh Chandimal, Niroshan Dickwella and spinner Akila Dananjaya which comes as a surprise. The Sri Lankans would need a complete change of form and fortune to be considered amongst the tournament’s top sides, and eight-straight losses including a rather embarrassing one against Scotland recently is not the kind of form that will attract bettors to their price of 30.59.

Most recent 2019 results: LLLLLLLLWL

Key player: Lasith Malinga

India

Outright odds: 3.89*

India come into the tournament in great form and lead the ICC player rankings with two batsmen and three bowlers among top 10. They also possess the number one batsman in the prolific Virat Kohli, who is the batsman with the highest average in the last two years. They also have number one bowler Jasprit Bumrah in their squad, whose speed and skill will be essential if they are to win the tournament.

Kohli will hope to lead India to a third World Cup title and alongside England they look the most likely side to win the tournament.

Most recent 2019 results: LLLWWWLWWW

Key player: Virat Kohli

Bangladesh

Outright odds: 58.98*

Bangladesh come into the 2019 World Cup with minimal expectations. They are a solid team and enjoyable to watch in some areas, particularly pace bowling, but are missing key ingredients in others to be competitive in the outright betting.

Bangladesh’s current form is mixed, and they are skilled enough to cause a shock on the day but unlikely to be consistent enough to take them any further than the group stages. They possess talent in their squad, with Mustafizur Rahman as one of the best pace bowlers in world cricket, but bettors should expect them to entertain rather than be competitive at the business end of the tournament.

Most recent 2019 results: LLLWLWWWWL

Key player: Mashrafe Mortaza

Pakistan

Outright odds: 11.82*

It has not been a good year for Pakistan in ODI cricket. This side is more than capable of playing some top-class cricket but also capable of self-destructing. Veteran fast bowlers Mohammad Amir and Wahab Riaz have been recalled to Pakistan’s 15-man squad after a whitewash loss to England in the one-day series recently and they are an extremely difficult side to predict.

This is a young side that have been together for the best part of three years now and with the likes of Babar Azam (one of the world’s best batters), Imam-ul-Haq (who is capable of scoring runs at the highest level), and leg spinner Shadab Khan who has been superb, they could prove to be the wildcard of the betting pack.

Most recent 2019 results: LLLLLLLLW

Key player: Babar Azam

New Zealand

New Zealand have a solid-looking team, a commanding and energetic side that is accomplished enough in all areas of the 15-man squad to be considered a serious threat to the tournament favourites.

Kane Williamson is very strong in bat, and Ross Taylor, Colin Munro, and Martin Guptill also provide real quality and a quick and powerful fast bowling player in Trent Boult, means they have a team that looks solid enough to be there at the final stages of the tournament. Bettors should expect the Kiwis to pose problems and be competitive with any side they play.

Outright odds: 10.63*

Most recent 2019 results: WWWLWLLLW

Key player: Kane Williamson

South Africa

Outright odds: 8.88*

This South Africa team could be considered one of their weaker squads recently but there is no doubting that they still possess serious quality in their ranks. Injuries could be an issue and they will need their top men firing if they are to go all the way in the tournament.

Opening batsman Quinton de Kock is a player that can be key for South Africa with his ability to score runs quickly. But it’s likely their progression through the tournament will rely on their bowlers Kagiso Rabada, Lungi Ngidi, and Dale Steyn who possess plenty of pace to go alongside top leg-spinner Imran Tahir, the top wicket-taker in this year’s Indian Premier League.

Most recent 2019 results: WWWWWWLWWL

Key player: Quinton de Kock

West Indies

Outright odds: 17.39*

The West Indies come into the tournament as a rejuvenated side and have a realistic outside chance of liftintg the trophy if their key players can perform to the level they are capable of. The pyramid of West Indies cricket has witnessed considerable change recently and bettors should not read too much into the recent form of this exciting side who are most certainly on the up.

Whilst all the talk is about how well Chris Gayle will perform and his final swansong their key player could be Andre Russell. The 31 year-old is powerful with the ball, a big hitter with the bat and nimble in the field. He is also entering the tournament in fine form having been named the Indian Premier League’s most valuable player this season, partly down to 52 sixes in 14 matches. The Windies will need him firing, and stand a genuine chance of being the live outsider in the outright betting.

Most recent 2019 results: LWLWWLWLL

Key player: Andre Russell

Australia

Outright odds: 5.110*

In a strange turn of events over the past year, Australia head into the World Cup in great form with a serious chance of defending their title. Australia’s captain and vice-captain were suspended for ball-tampering this time last year, but the return of both players, Steve Smith and David Warner, coupled with an upturn in form of the entire team mean they will be a threat. Strong, varied bowling, give the Aussies a very realistic chance of success.

Much like the West Indies, bettors shouldn’t read too much into the Aussies past form from last year, as they are a much more settled, confident side under Aaron Finch’s shrewd guidance and look to have found a way to make things work finally.

Most recent 2019 results: WWWWWWWWLL

Key player: Aaron Finch

Cricket World Cup betting: Where is the value?

Based on recent form it is hard to look past England and India in the outright betting. England will have the help of home advantage and their team that looks the most settled. A talented squad that possesses the likes of Joe Root, Ben Stokes, Eoin Morgan, and Moeen Ali, they don’t seem to have any real major weaknesses.

The team to keep a close eye on are Afghanistan from the outsiders. After a remarkable story, they managed to secure a tie against No.2 seed India recently and they have proven themselves to be a capable team who can compete against anyone on their day. It is likely out of reach that they will go all the way, but a semi-final spot could be potential value to bettors.

For bettors looking to alternative markets, the top batsman and bowler is always a popular one. Virat Kohli is being pinpointed as a hot favourite for top run scorer, but as outstanding a batsman as he is, the stats suggest he’s an unjustified favourite. Elsewhere, Jonny Bairstow may give serious competition to the top field, and comes into the tournament in excellent form. South African pairing Faf du Plessis and Quinton de Kock are both ranked in the world’s top five ODI batsmen, whilst Kane Williamson of New Zealand is also one to keep an eye on.

For bowlers, one-day cricket can be demanding but matches are often won with the ball rather than the bat, and at just 20 years old, Afghanistan spinner Rashid Khan is already one of the world’s best, and ended last year as the leading ODI wicket-taker, so is worth considering. Pat Cummins, has taken 17 wickets in six matches so far this year, and has been the world’s most effective ODI bowler. Whilst Jasprit Bumrah tops the list of ODI bowlers on recent form, and will certainly be in the mix as the tournament progresses.

A close look at the Tottenham vs. Liverpool odds

The odds indicate that Liverpool are most likely to lift the trophy. The favourites are available at 1.543* to win the competition. The betting market is also favouring a lower-scoring final with Over 2.5 Goals currently priced at 2.150 which could offer value to bettors.

Time: Saturday June 1, 20:00 CET kickoff

Venue: Wanda Metropolitano Stadium, Madrid (neutral).

Team news and predicted lineups

Liverpool predicted lineup

(4-3-3) Alisson; Alexander-Arnold, Matip, Van Dijk, Robertson; Fabinho, Henderson, Milner; Salah, Mane, Firmino.

Liverpool team news

Naby Keita is expected to be Liverpool’s only absentee.

Tottenham predicted lineup

(4-2-3-1) Lloris, Trippier, Alderweireld, Vertonghen, Rose, Wanyama, Sissoko, Eriksen, Alli, Lucas, Son.

Tottenham team news

Tottenham could be without three players due to injury, with Jan Vertonghen and Harry Kane suffering ankle injuries and Harry Winks suffering with a groin problem.

Head-to-Head history and results

• Tottenham and Liverpool have never met in an official UEFA competition before.
• In domestic meetings, Liverpool have 82 wins to Tottenham’s 48, with the teams sharing 42 draws.
• Liverpool have won both League games against Spurs this season – both by a margin of just one goal.
• Liverpool have the superior record over Tottenham in recent seasons, while Spurs have only beaten Liverpool once in their last 14 meetings.

Tottenham’s 2018/19 Champions League statistics

Liverpool’s 2018/19 Champions League statistics

Inform your Tottenham vs. Liverpool prediction

Tottenham come into the fixture with Liverpool after a heroic two-legged victory over Ajax and will be playing in their first ever European Cup/Champions League final. In an unlikely turn of events, Brazilian attacker Lucas Moura managed to grab a second-half hat-trick to secure an extraordinary 3-2 win in Amsterdam and send the London club through to the final on away goals.

When analysing the semi-final using expected goals data, Ajax had an excellent thirty minute period in the first leg in London and it was nearly enough to take them to Madrid, but Tottenham outplayed them overall. Spurs heavily dominated the Dutch side throughout the entire second half in the return leg and their persistence and drive earned them a deserved comeback (xG: Ajax -1.66 – Spurs 3.28).

Tottenham took 7 points from their final three matches in the group stages after a dreadful start – including a draw away to Barcelona, which meant they edged out Inter Milan before a resounding win over high flying Bundesliga side Borussia Dortmund. They then had a seesaw battle with domestic rivals Manchester City in the knock-out rounds before beating Ajax, who had previously knocked out Real Madrid and Juventus, which demonstrates that the North London club will be a live underdog in Madrid on June 1.

Liverpool made it into their second successive Champions League final after dramatically overcoming Barcelona with a stunning second-leg comeback at Anfield. The Reds chances of a second successive Champions League final were around 5% at this stage. With the victory, Jurgen Klopp’s side became the first English side to reach back-to-back Champions League finals since Manchester United in 2008 and 2009.

When analysing the expected goals data, Liverpool were unlucky to lose so convincingly in the first leg, but they also still controlled the match for long periods, as well as having created some good chances. (xG: Barcelona 2.15 – Liverpool 1.58). The second leg saw The Reds run out the convincing winners (xG: Liverpool 2.29 – Barcelona 0.78). They were clinical in front of goal and justifiably qualified for the final despite having two of their most inspirational players in Salah and Firmino injured.

In the group stages, Liverpool did show vulnerability by losing all three away games, and that shows they are a side that can be tamed if you are tactically astute. Domestically, they have been outstanding, only losing one league game and finishing one point behind Manchester City. The Merseyside club have lost just once in their last 23 games in all competitions, including two victories against Tottenham, but only by a single goal margin, ending in 2-1 wins.

A further positive for Liverpool coming into the final is that Tottenham have conceded an average of 1.79 GA per game in the Champions League this season, showing that they have been defensively vulnerable, and with the attacking talent available to Jurgen Klopp, it is hard to see a scenario where Spurs can keep their attacking trio at bay for the full 90 minutes.

Key battle: Kieran Trippier vs. Sadio Mane

Sadio Mane will go into the final as Liverpool’s most in-form attacking player and he could be the difference in a key battle against Kieran Trippier. The Tottenham and England right-back has endured plenty of criticism this season, and with Andy Robertson providing the Senegalese international extra width it will be an area that Jurgen Klopp will feel he can exploit.

While his goalscoring ratio has improved, Mane’s all-round game has stayed at a similar consistency, recording comparable numbers in terms of key offensive stats. This would suggest Mane has become more clinical in front of goal, taking his chances more often, with the 27 year-old  averaging 2.4 shots per game this season, in contrast to 2.8 last campaign.

In terms of dribbles, he averages 1.5 per year this season, down from 1.9 last season and 2.5 the season before. Mane has also amazingly only recorded one assist in the Premier League this season, compared to seven last season, leaving a lot of the key passes to Salah.

Trippier has been accountable in defence for key errors at times for Tottenham this campaign, and whilst he is fully capable of providing a threat when attacking, it will be the defensive side of his game that can potentially be exploited by Liverpool.

Tottenham vs. Liverpool: Where is the value?

Tottenham have just one win in their last 14 fixtures against Liverpool, and The Reds are the superior team, whilst they will also be looking for redemption after last year’s heartbreak against Real Madrid. What Liverpool did against Barcelona gives them the perfect chance to do this against Tottenham and odds of 1.531* to lift the trophy potentially offer value.

With the attacking ability in both sides and the vulnerability Spurs have shown defensively against quality opposition in this competition the final has potential to be a high-scoring affair, although the line is favouring a lower-scoring match. Over 2.5 Goals to be scored in the game is available at 2.150 which could prove to be a value bet.

Bettors looking to alternative markets, the last time a Champions League final failed to see both teams score was back in 2010, so eight finals in a row have witnessed both teams scoring, and the same has happened in seven of Jürgen Klopp’s nine matches against Tottenham, which suggests the probability is high that both sides will find the net at some stage.

Tournament format

Kicking off with the group stage in September, Netherlands, Switzerland, England and hosts Portugal all won their respective groups to make it to the finals in June – with all believing they can become the first ever Nations League champions.

Of the four finalists, Switzerland had the best record from the group stage with nine points from four games, followed by Portugal with eight points and England and Netherlands both finishing with seven points. Portugal are the only finalist to not lose a game during the group stage.

Ahead of the two semi-finals being played on June 5 and 6, before the final on June 9, read on to inform your Nations League betting strategy.

Betting insight ahead of Netherlands vs. England

England come into the match as marginal favourites (around 40% chance). The Over/Under market odds marginally suggest it will be a low-scoring contest, with under 2.5 currently priced at 1.900*.

Date – Thursday June 6

Time – 18:45 UTC

Venue – Estádio D. Afonso Henriques

Netherlands vs. England predicted lineups

Netherlands predicted lineup

4-3-3 – Cillessen (GK), Dumfries, van Dijk, de Ligt, Blind, de Roon, de Jong, Wijnaldum, Babel, Depay, Promes

Stefan de Vrij could move into the team should Netherlands opt for a back three, whilst Donny van de Beek, Tonny Vilhena and Luuk de Jong will all be pushing for a starting spot.

England predicted lineup

4-3-3 – Pickford (GK), Walker, Maguire, Stones, Rose, Henderson, Rice, Alli, Sterling, Kane, Sancho

Southgate will face a formation and selection dilemma ahead of the game, with checks on players from Liverpool and Tottenham Hotspurs who would have competed in the Champions League six days prior.

Should England opt for a five at the back system, Kieron Tripper or Trent Alexander-Arnold could come into the side, whilst Ross Barkely, Eric Dier Marcus Rashford will all be pushing for starts.

Netherlands vs. England predictions

Coming into the semi-finals with identical group stage records, the Netherlands and England will both believe they can progress to the final and lift the inaugural Nations League trophy.

England’s run to the World Cup semi-final last year, combined with their impressive performances in the Nations League group stage, mean they come into the finals at 2.950* to win the tournament outright, only marginally behind favourites, hosts and reigning European Champions Portugal at 2.700*.

The three lions arrive in Portugal in great form having won their last five international matches, scoring 10 goals in their last two games against Czech Republic and Montenegro in their European Qualifying group.

Netherlands meanwhile are third favourites at 4.200*, despite knocking out reigning World Champions France on goal difference, and helping relegate four-time world champions Germany to League B.

Winning two, drawing two, and losing one in their last five fixtures heading into this semi-final, the Dutch are in fair form under Ronald Koeman.

The last time these two sides met in a competitive international fixture was way back during Euro 96, in which the English triumphed 4-1. More recently, England also beat the Netherlands in an international friendly in March last year.

Between those two games however England failed to beat the Netherlands in the next seven international friendlies, losing three and drawing four between 2001-2016.

After kicking off their group stage match with a 2-1 home loss to Spain, England showed resiliency to win the return fixture in Seville 3-2, as well as drawing away and beating Croatia at home.

Scoring six and conceding five England averaged 1.5 GPG whilst conceding 1.2 GAPG, with Marcus Rashford and Raheem Sterling jointly leading their nation’s scoring total with two goals each in the group stage. Jessie Lingard and Harry Kane both have one goal.

The Oranje performed slightly better during the group stage despite amassing the same points as the Three Lions, averaging two GPG and one GAPG. Scoring two more goals than England, Memphis Depay, Georgino Wijnaldum, and Virgil van Dijk all got two goals each with Ryan Babel and Quincy Promes contributing a goal each.

The first meeting in a knockout game for these two sides should create in intriguing matchup between two teams who will think they can eliminate the other.

It will be intriguing to see whether the Champions League Finals six days prior to this game will have any bearing on team selection and the fitness of some of both side’s key players.

Netherlands vs. England: Where is the value?

England have seven players who could all feature in that final; Harry Kane, Danny Rose, Delle AllI, Eric Dier, Joe Gomez, Trent Alexander-Arnold, and Jordan Henderson.

The Dutch duo of Virgil van Dijk and Georgino Wijnaldum are the only players affected by the final, but to lose one or both through fatigue or injury would be a massive blow to Holland’s chances of making the Nations League final.

England have a wealth of attacking options (should they remain fit) and bettors can find value in the Totals market with England over 1.5 goals in 90 minutes currently priced at 2.350*.

Inform your Portugal vs. Switzerland predictions

Portugal come into the match as favourites (around 61% chance). The Over/Under market odds marginally suggest it will be a low-scoring contest, with under 2.5 currently priced at 1.854*.

Date – Wednesday June 5

Time – 18:45 UTC

Venue – Estádio do Dragao

Portugal vs. Switzerland predicted lineups

Portugal predicted lineup

4-2-3-1 – Patricio (GK), Cancelo, Pepe, Dias, Guerreiro, Pereira, Carvalho, B.Silva, Fernandes, Guedes, Ronaldo

Rotating between a 4-4-2, 4-3-3 and 4-2-3-1, Portugal manager Fernando Santos has a wealth of options at his disposal, with his side being the only team unaffected by players playing in the Champions League Final.

Wolves trio Ruben Neves, Joao Moutinho, and Diogo Jota’s fine form at the end of the season will put them in contention to start as well as Benfica’s Rafa Silva and Joao Felix.

Switzerland predicted line-up

3-5-2 – Sommer (GK), Elvedi, Akanji, Rodriguez, Mbabu, Zakira, Xhaka, Freuler, Shaqiri, Embolo, Seferovic

Xherdan Shaqiri may be the only player affected by the Champions League final, but is likely to start from the bench for Liverpool, which theoretically should make him available for selection. The Swiss could opt for Albein Ajeti to partner Embolo up top with Fabian Schar and Stefan Zuber also pushing for a start in defence.

Portugal vs. Switzerland predictions

In what, in theory, could be a more clear-cut contest, favourites Portugal take on outsiders Switzerland in the first Nations League semi-final.

It’s the outsiders who come into the game in better form, winning three, losing one, and drawing one of their last five games, including an impressive 5-2 victory against Belgium.

Portugal meanwhile have drawn their last four competitive games in a row, with their last win coming five games ago in a friendly against Scotland.

Both sides have had 2-0 victories over the other in their last two competitive fixtures, which came during World Cup qualification in 2016 and 2017.

During the group stage, Portugal were without their captain and talisman Cristiano Ronaldo through injury and fatigue – relying instead on Andre Silva and Barnardo Silva to get them through qualification.

Andre Silva, who may not even start for the hosts, bagged three goals, Bernardo Silva scored one and the last of their five goals in the group stage was courtesy of Poland centre back Kamel Glik’s own goal.

Despite Portugal averaging the lowest GPG at 1.25, they do boast the best defensive record of the four teams left, averaging 0.8 GAPG after conceding just three goals in qualification.

Switzerland on the other hand were in lethal form during the group stage, scoring almost as many as the other three teams combined with 14 goals – 3.5 GPG.

Haris Seferovic finished joint second top scorer in the Nations League group stage with five goals – the only player to score more than one goal for Switzerland, with nine of his teammates all registering a single goal.

Defensively, the Swiss let in five goals in the group stages, to average 1.3 GAPG.

Home field advantage looks set to play an important role in this fixture, hence Portugal’s position as favourites to win the tournament.

The 6-0 and 5-2 demolitions of Iceland and Belgium by the Swiss will give Portugal cause for concern, but playing on home soil and the return of five-time Balon D’or winner Cristiano Ronaldo should galvanise the Portuguese to reach the final against one of England or the Netherlands.

Portugal vs. Switzerland: Where is the value?

If Switzerland have what it takes to overcome the favourites on home soil, then their current position as 6.250* outsiders will almost certainly change.

Given their record in the group stage, combined with Portugal’s attacking potential at home, bettors could opt to go with over 2.75 goals at 2.320*.

Could you imagine eating 74 hot dogs in 10 minutes? It seems impossible, but Joey Chestnut managed that in 2018 when he won Nathan’s Famous Fourth of July International Hot Dog Eating Contest.

Believe it or not, while there are some people out there looking for value in European soccer leagues or Major League Baseball, others will be analysing the odds ahead of the best-known hot dog eating contest.

Bettors can now bet on whether Joey “Jaws” Chestnut will claim his twelfth Nathan’s Hot Dog Eating Contest title and if he (or anyone else) will eclipse his all-time record of 74 hot dogs.

What is Nathan’s Hot Dog Eating Contest?

Nathan’s Hot Dog Eating Contest started in 1914 and is held every year on July 4. Considered to be the highlight of the competitive eating calendar, millions of viewers tune in to watch the competition. In addition to the kudos that comes with devouring more hot dogs than anyone else, whoever consumes the most wins a hefty chunk of a $40,000 in prize purse.

Nathan’s Hot Dog Eating Contest betting is obviously a bit of fun, but with some basic knowledge of how the competition works and a quick look at previous results, there could be some value on offer. Joey Chestnut is the clear favourite at 1.112* to win, with the odds for the men’s competition currently favouring the Under 73.5 hot dogs at 1.714* (the Over is priced at 2.200*).

Nathan’s Hot Dog Eating Contest format

The International Federation of Competitive Eating (IFOCE) states that the winner of Nathan’s Hot Dog Eating Contest is the contestant that consumes and keeps down (this is an important if rather unpalatable thought) the most hot dogs and buns (HDB) in the allotted time (10 minutes). Drinks are permitted when eating, with water being a popular choice, as are condiments.

A scorekeeper is assigned to each contestant to keep track of each hot dog consumed. Hot dogs that are not consumed in their entirety will count, provided a sufficient amount has been eaten. Hot dogs still in a contestant’s mouth when the time has elapsed will also count if they are then swallowed. Messy eating is punishable by a yellow penalty card, and contestants can be disqualified if they regurgitate.

If the scores are tied after 10 minutes, it goes to a five-hot-dog eat-off to see who can eat the quickest and if the contestants still cannot be split then it will go to a sudden-death eat-off of a single hot dog.

Eat-offs are however a rarity, having only occurred twice in the history of the contest, with the first instance in 1980 and the most recent in 2008 when Joey Chestnut and Takeru Kobayashi tied on 59 hot dogs. Chestnut then defeated Kobayashi by being the first to finish his plate of five to win the mustard-yellow belt.

Nathan’s Hot Dog Eating Contest results

Pinnacle has offered odds on Nathan’s Hot Dog Eating Contest since 2006 and the table below shows the total number of hot dogs eaten by the winner in that time.

*Kobayashi won six straight titles between 2001 and 2006.

As you can see from the table our traders have found it hard to gauge the number of hot dogs eaten, given that the winner has eaten above the Over every year except on two occasions.

The fluctuation in the numbers could be due to the increased expectations of Joey Chestnut over time. It is interesting to note that, despite winning, he massively underperformed in 2010 – the Totals number drastically reduced the following year by 10 hot dogs.

Sharp bettors would have taken into account his prior records all finishing above the 58.5 offered and backed the over at 2.060.

In 2015, Matt Stonie managed to dethrone the eight-year reign of Joey Chestnut by eating 62 hot dogs – well below the 66.5 offered by Pinnacle. Last year, the men’s total hot dogs eaten by the winner market opened at 70.5 with the Over favoured at 1.847 and the Under priced at 1.98.

How to pick your wiener: Is there a value bet?

Joey Chestnut has consumed an average of 65.18 HDB since 2008 (when the 10 minute time period was introduced). He suffered a dip in 2014 and 2015, but has regularly beaten his opponents by 10+ hot dogs.

Given that Chestnut has eaten 70 or more HDB, and bettered his previous effort in each of the last three years, it seems he will be very difficult to beat. When you consider his recent improvements, Over 73.5 HDB also looks achievable.

To see the man himself in action, check out this video of last year’s contest:

Note: Despite a final total showing 64 HDB, Chestnut’s designated counter actually missed a full plate of hot dogs and his official total was a new record of 74 HDB.

Over the years, genuine challengers to Chestnut’s competitive eating crown have been few and far between – Matt Stonie is the only man to come close (his success at Nathan’s Hot Dog Eating Contest in 2015 is the only break in Chestnut’s current 12-year reign).

After Stonie narrowly edged out Chestnut, the MLE (Major League Eating) scene changed. Stonie became the youngest but also the top-ranked competitive eater in MLE, demoting Chestnut to second. However, it didn’t take long for Chestnut to reclaim that position and he managed an impressive competition record of 74 HDB last year.

Weighing just 130 pounds, Stonie is a multiple competitive eating record holder, but after a steady decline in his performance since his title-winning 62 HDB effort in 2015 (57 in 2016, 48 in 2017 and then 42 last year) it’s difficult to see him challenger this year.

Another contender who could be considered a genuine threat is Geoffrey Esper. Esper is ranked number three in the competitive eating standings and recently shocked Joey Chestnut at a doughnut-eating contest – Esper ate 235 in six minutes while Chestnut could “only” manage 200.

The problem for anyone look to challenge Chestnut is that he’s been so consistent and recent years and the jump from 50 to 70 HDB seems astronomical.

One way to keep ahead of the market is to keep an eye on the contestants form in other competitions and check their social media accounts for updates. A good example of this is 2017 when Joey Chestnut posted a video online of his preparations (where he put away an impressive 77 hot dogs). That resulted in a fairly hefty line move on the Total.

Pinnacle had the mark set at 68.5 prior to the video being uploaded and it had moved to 72.5 a day after.

It’s not just about the men: Women’s hot dog eating contest

Women have competed in Nathan’s Hot Dog Eating Contest since 2011 – Sonya Thomas was the first ever winner and currently holds the record with 45 HDB in 2012.

Thomas won the first three instalments of the women’s event before Miki Sudo triumphed with a total of 34 HDB in 2014. Sudo has consistently improved on her total up until last year – she couldn’t beat her 41 hot dogs from 2017 but still claimed her fifth straight title with a total of 37.

In terms of the outright betting for the women’s competition – much like the men’s – there’s a clear market favourite with Miki Sudo currently 1.100* to win. Bettors can also bet on the total the number of hot dogs eaten by the women’s winner with the mark is currently set at 38.5.

Will we see a Nathan’s Hot Dog Eating Contest record?

In competitive eating circles, Joey Chestnut is a superstar. The 6ft, 230-pound California native has dominated Nathan’s Hot Dog Eating Contest for over a decade.

Despite Matt Stonie’s win in 2015, many believe the biggest rivalry in competitive eating is still considered to be Joey Chestnut and Takeru Kobayashi. Unfortunately, fans haven’t got to see many contests featuring both as contractual issues and competition controversy means Kobayashi hasn’t competed at the July 4 event since 2010.

There might be challengers but Chestnut is clearly the man to beat. It will be incredibly difficult for anyone to take the mustard belt away from him come July 4. Anyone looking for slightly longer odds might want to look at the chances of Chestnut beating his own record of 74 HDB.

Can “Jaws” do the unthinkable and beat his record again? Is it humanly possible to eat 75 hot dogs in 10 minutes? We’ll have to wait until July 4 to find out.

What is an outright bet?

An outright bet is a bet placed on the outcome of an entire league or competition as opposed to a single bet on an individual game or event.

Bettors can place an outright bet before the competition commences but markets are often open whilst the competition is taking place.

When does long-term betting offer value?

Long-term outright bets are useful for locking in value on teams whose odds may fluctuate over the competition. As an example a bettor may consider the strength of the Real Madrid to be underrated by betting markets. If this was the case the market would correct fairly quickly on a match-to-match basis.

By betting on Real to win that season’s Champions League outright rather than betting on each individual match the bettor locks in value which may otherwise quickly diminish.

This is not to say that outrights always offer value however. It is not uncommon to see better odds on betting individual events over the outright. One extreme example is an infamous bet placed on Tiger Woods to complete a clean sweep of the golf majors (after winning the 2019 Masters) at 101.0. This price was taken despite the fact that some sportsbooks priced Woods at 11.00 to win the PGA Championship alone.

Assuming this is indicative of the odds Woods will be priced at the further two major events the odds the bettor could achieve by just rolling over winnings into the outright for each event would be in the 1300.00 range, a payout 13 times larger than taking the future on the clean sweep. That’s without considering the chance that Woods may miss an event through injury.

This effect is similar but more difficult to calculate when betting on more conventional outright markets. Could the bettor secure better odds by utilising his bankroll betting on individual events?

Another consideration is opportunity cost. A profitable bettor will likely be able to use his bankroll more effectively over the course of a Premier League season, for example, than simply betting on the outright winner at the start of the season.

Hedging an outright bet

It is usually possible to hedge an outright bet but on some occasions it can be difficult to do so. In competitions involving only two competitors or markets where a “not to win” selection is offered it is as simple as hedging a normal bet, albeit sometimes over a longer time period. Read how to hedge a bet for more information.

Where hedging becomes more difficult is if multiple selections can win the event, or the odds on the selection are so high that large sums are required to hedge the bet. Bettors should consider their exit options prior to placing the bet in this case.

In most cases hedging requires a cost, whether that’s the high margin on a Cashout, commission or a bookmaker’s margin for the second time. It is likely better to allow the bet to run its course in the long run.

Making outright betting predictions

When making a prediction on an outright market a bettor should be looking for a selection that will not just offer value but also offer value above and beyond what that bettor can do with the money in the meantime.

Betting on outrights may actually provide one of the more realistic ways to find value bets. This is because of the opportunity cost associated with placing a long term outright bet. Bettors capable of finding positive expected value bets consistently would be unwise to tie up their bankroll over an extended period of time, possibly leaving the market relatively less efficient due to a delay in signalling.

This can cause a lag in the odds movement that otherwise wouldn’t exist. For example in the summer of 2017 PSG could be bet on as high as 21.00 in the Champions League outright market.

These odds remained available despite the odds on PSG signing two superstar players (Neymar and Kylian Mbappe) shortening significantly. Logically the developments involving their transfers should have followed through to the PSG’s outright odds, since the acquisition of such players would improve their Champions League chances. However it took a while for the market to react accordingly, and by the end of the Summer PSG were available at a highest price of 11.00.

A bookmaker’s odds essentially provide a direct forecast of the probability of sporting outcomes. When they are expressed in decimal notation, you simply invert the figure and you have your implied percentage (although there is a little bit of extra work involved to remove their margin). Odds of 2.50, for example (with the margin already removed) imply a 1/2.50, 0.4 or 40% outcome probability.

Of course, a bookmaker is not always right. Their mistakes provide opportunities for their customers to find expected value and make a long-term profit. Readers, however, will be familiar by now with my message that Pinnacle’s mistakes, for a soccer match betting market at least, are randomly distributed, and that on average their odds offer a very reliable reflection of the chances of things happening. In this context, their closing odds are the best of all, and can be used to predict how much profit a bettor can expect to make.

However, there are other forecast groups who offer predictions about football match outcomes. One of the best known is Fivethirtyeight.com, the politics, economics and sports forecasting blog created by the American statistician, former poker player and author of The Signal and the Noise Nate Silver. Nate made a name for himself by successfully calling the outcomes of 49 of the 50 states in the 2008 US presidential election, and all 50 four years later.

From the perspective of the sports fan or bettor, FiveThirtyEight’s match forecastsare particularly useful because they explicitly offer the probabilities for home, draw and away. By inverting these we can immediately obtain their implied fair odds. The goal of any value bettor is to find fair odds that are more accurate than those of bookmaker. If they can, then it’s simply a matter of betting the bookmaker’s odds when those are longer. If the bettor is making a long-term profit, that’s a sure sign that they really have more accurate odds than the bookmaker. In this article I have collaborated with @PlusEVAnalytics to find out whether FiveThirtyEight can do the job.

The FiveThirtyEight football prediction model

FiveThirtyEight first started publishing football match predictions in January 2017, although their results database extends back to August 2016. Their methodology is based on a “substantially revised version of ESPN’s Soccer Power Index (SPI)” originally devised by Nate Silver, that utilises expected goals (or xG) and Poisson statistical analysis to generate a matrix of possible match scores, from which home, draw and away probabilities are calculated.

FiveThirtyEight think their soccer predictions are pretty good, arguing that they are more valuable than an unskilled guess. In that respect I am sure that they are right, but are they better than Pinnacle’s implied probabilities? Let’s find out.

Can FiveThirtyEight’s prediction make a profit?

Combining an historical Pinnacle closing odds database with the implied probabilities made available by FiveThirtyEight, I put together a sample of 16,635 matches from European soccer leagues played between 12^th August 12, 2016 and March 31, 2019, for a total of 49,905 home/draw/away odds pairs.

On 20,093 occasions, Pinnacle’s closing odds were longer than those implied by FiveThirtyEight’s probability forecasts. The average superiority of those odds (average 4.12) was 16.2%, implying that had we bet those odds to level stakes we should have made a profit of about 16.2% assuming FiveThirtyEight’s odds, on average, to be an accurate or efficient reflection of the ‘true’ odds. In fact, they showed a loss of -6.0%, worse than the -4.3% loss betting all 49,905 prices (although not statistically significantly so).

The first chart below shows how FiveThirtyEight’s implied odds failed to predict actual returns from betting Pinnacle’s closing odds. Dividing Pinnacle’s closing odds by FiveThirtyEight’s implied odds gives us an expected return for that bet, assuming the hypothesis that FiveThirtyEight’s odds are efficient is true.

Grouping bets by incremental expected returns (with a 0.01 resolution) we can see that expected returns fail to correlate with actual betting returns at all. Regardless of the ratio of Pinnacle’s closing odds to FiveThirtyEight’s implied odds, the average return is a loss of about -6%. It would seem FiveThirtyEight’s odds offer no predictive value at all, relative to Pinnacle’s closing odds.

What if we turn things around? This time let’s pretend that FiveThirtyEight is the bookmaker and Pinnacle is the forecast model. Now the ratio of FiveThirtyEight’s odds to Pinnacle’s fair closing odds (with the margin removed) is used as the measure of expected return.

Betting FiveThirtyEight’s ‘odds’ on the 25,557 occasions they exceed Pinnacle’s fair closing odds, the actual return is 15.5%, very close to the average superiority of 15.9% (average odds 4.49). The scatter plot confirms the strong correlation between expected and actual returns for this reverse hypothesis. The slope of the trend line is almost exactly 1 and passing through the origin. (see the y = mx + c equation in the chart), implying that Pinnacle’s closing odds, on average, are highly efficient, not FiveThirtyEight’s.

Apples and Oranges

When I published these findings on my twitter feed in April it was rightly pointed out that we are comparing apples and oranges. FiveThirtyEight’s forecast probabilities are produced in advance of the fixtures, with their final probability estimate published after a team’s penultimate match has finished. This may be many days before the fixture in question. Their predictions can only be as good as the information that was available to them at the time.

Pinnacle’s closing odds, by contrast, will reflect all information available to the market right up to the point the match actually starts. This will include other factors like player injuries, changes to team selection, weather and the state of the pitch, things that FiveThirtyEight’s does not and cannot include.

To make a wholly fair model comparison with Pinnacle’s closing odds we would need FiveThirtyEight to issue probability forecasts at the same time, i.e. kick-off. That’s not going to happen. Alternatively, we could use Pinnacle’s odds at the time FiveThirtyEight issued their final forecast probabilities for a match. Unfortunately, I don’t have time-stamped data for Pinnacle’s odds and even if I did, I would imagine that the publication of their opening odds will often post-date the publication of FiveThirtyEight’s final forecasts.

Nevertheless, using Pinnacle’s opening odds will potentially offer a fairer model comparison than their closing ones; let’s look at the results. For the 18,952 occasions Pinnacle’s opening odds were longer than for FiveThirtyEight (average 3.97) the average superiority of those odds was 14.2%, They showed a loss of -4.1%, marginally better (but not statistically significantly so) than the -4.4% loss betting all 49,905 prices. Again, there was little correlation between expected and actual returns.

As before, when performing the model comparison in reverse (with Pinnacle’s opening odds used as the benchmark of ‘truth’), there was a much better correlation, not as precise as for the closing odds but still close to parity. Betting FiveThirtyEight’s ‘odds’ on the 25,775 occasions they exceed Pinnacle’s fair odds odds, the actual return is 12.8%, again reasonably close to the average superiority of 14.8% (average odds 4.54).

The Signal and the Noise

After posting the opening odds data on twitter, @PlusEVAnalytics, my co-author for this article, remarked that “if you’re asking yourself whether FiveThirtyEight is superior to Pinnacle, you’re going to get the obvious answer.”

Perhaps we are, and what we’ve found here is all rather self-evident. It is still probably true that FiveThirtyEight’s forecast probabilities will be informationally deficient relative to Pinnacle’s, both because of timing and because Pinnacle’s primary business model is to get their odds right, whilst FiveThirtyEight’s is simply to entertain. The latter doesn’t have to make money from its forecasts; well not directly anyway. To be fair, furthermore, FiveThirtyEight admit that their forecasts are not designed with betting in mind.

However, @PlusEVAnalytics proposed an interesting thought experiment. By combing the two forecast models to make a potentially better one than merely Pinnacle’s odds on their own we could uncover whether there was any signal in FiveThirtyEight’s forecasts residual to the signal in Pinnacle’s odds.

The analysis thus far has been framed as an “either/or” proposition – which of the two predictions is a more accurate source of truth? The results are as conclusive as they are unsurprising. Now, let’s add a little twist by rephrasing the question as follows:

“Ultimate” probability of any given outcome = Z * (FiveThirtyEight’s probability of that outcome) + (1 – Z) * (Pinnacle’s probability of that outcome), for any given Z, 0 ≤ Z ≤ 1.

What value of Z maximises the predictive value of these ultimate probabilities?

This model is flexible enough to accommodate the construction of the ultimate probabilities using entirely FiveThirtyEight’s projections (by setting Z = 1), entirely Pinnacle’s projections (by setting Z = 0), as well as everything in between (by setting Z between 0 and 1).

How do we determine the best value of Z? There are several possible ways, but the one we’ll use here is maximum likelihood estimation (MLE). The purpose of MLE is to find the value(s) of one or more unknown parameters that provide the best “fit” to a set of observed data. How do we measure how good the fit is? By looking at the likelihood, conditional on the value of the unknown parameter, of observing exactly what we have observed.

Our model, defined above, has a single parameter – Z. Any given value of Z will allow us to calculate a set of ultimate home/draw/away probabilities, based on that value of Z, for each match in the data set. For each match, the likelihood of observing what we observed is our ultimate home/draw/away probability, if the result of the match was a home win/draw/away win. For example, for H/D/A probabilities of 0.5, 0.3 and 0.2, if the result is a draw, the likelihood of observing what we observed is 0.3.

Because all the matches are independent events, the likelihood of observing the exact set of outcomes that we observed is the product of the likelihoods from each individual match. This product is our objective – it’s what we are trying to maximise by adjusting Z.

Unfortunately, the product of 16,635 probabilities is infinitesimally small – think of it as a 16,635-leg parlay. This creates a computational problem when solving the MLE using a software tool such as Excel that is unable to handle small numbers beyond a certain threshold – anything smaller is rounded to zero.

To circumnavigate this problem, we can instead maximise the logarithm of the likelihood. Since the absolute value of the likelihood is irrelevant to the MLE procedure – what matters is how that likelihood changes as we adjust our parameters – maximizing the so-called “loglikelihood” is mathematically equivalent.

In addition to taking the logarithm of the ultimate match probabilities, we now also take the sum of the loglikelihoods rather than their product. Running this analysis on the data set gives the following results:

• When using Pinnacle’s closing odds, loglikelihood is maximised when Z = 0. In other words, FiveThirtyEight contributes effectively nothing to the “ultimate” forecast model when considering just FiveThirtyEight and Pinnacle forecasts.
• When using Pinnacle’s opening odds, loglikelihood is maximised when Z = 0.04. In other words, FiveThirtyEight contributes about 4% to the “ultimate” model.

The wisdom of models

It is the second of these two results that is interesting. When forced to choose one or the other, Pinnacle’s opening lines provide predictive value that is objectively far superior to FiveThirtyEight’s predictions. But here’s the twist…

A weighted average made up of 4% x FiveThirtyEight’s prediction + 96% x Pinnacle’s opening line provides better predictive value than either of the two predictions does individually!

What value does this provide to bettors? Practically, not much. The 4% is too small to be of much use, and it may not even be statistically significant. But what if Z was larger? And, what if instead of two predictions there were many, with each one getting its own Z? This is a variation of the “wisdom of crowds” theory, which states that a combination of separate predictions put together can be more valuable than even the best of those predictions by itself.

Essentially, this is what makes Pinnacles odds so accurate (on average). They have the most sophisticated and knowledgeable traders setting lines. They also allow other sharp bettors to play rather than refusing their custom, thereby helping adjust their lines to become ever more accurate or wise. Pinnacle’s closing odds effectively represent a “wisdom of forecast models” and their odds reflect the “ultimate” outcome probabilities. And that is probably why FiveThirtyEight never stood a chance against them.

The difficulties of measuring the effects of VAR

It is hard to assess the impact of VAR on soccer matches. Intuitively bettors may feel that the VAR system may weaken the effect of home field advantage, however it is difficult to quantify this. This is due to a few key factors.

There have not been many matches involving the use of VAR so it is difficult to collect a statistically significant set of data.

Equally, soccer is a low-scoring game which is naturally quite variable, so collecting data from a single season may cause issues especially with teams changing style, quality, managers and personnel.

To further complicate isolating the effect of VAR, it is unknown to what extent home advantage is down to refereeing decision making and how much can be put down to other factors VAR does not affect such as the crowd and familiarity with the stadium.

Additionally, since overall results are so dependent upon changes in the quality of different teams season-to-season the effect of VAR is unlikely to be a big contributor to overall outcomes over the course of a campaign. This makes it tricky to assess what effect it might have on a one-off fixture.

These complications make it difficult to find a metric by which to measure the effects of VAR. It is, therefore, best to focus on those match events directly influenced by the use of VAR technology.

Taking red cards as a measure proves difficult since VAR is not always used for dismissals involving two bookings. Straight red cards are also relatively rare (there were just 44 in the 2017/18 Serie A season).

Likewise, VAR is only used for close offside calls that lead to goals being scored or disallowed. This is both a rare event and one that is affected by the attacking side converting the chance to begin with, adding an extra level of variance to the data.

This leaves penalties as perhaps the most suitable metric to analyse considering the current restrictions on sample size.

Using home/away penalty data

One of the biggest decisions a referee can make in a soccer match is awarding a penalty kick to either team. We know that penalties are scored approximately 75.8%of the time so with one blow of his whistle the official is essentially awarding 0.758 goals to a team.

Comparing the penalties awarded to home and away teams could be a promising way to see if VAR is having an effect on home advantage, particularly as VAR is beginning to be routinely used for all penalty decisions. This ensures penalty decisions provide a useful way to isolate the effect of the Video Assistant Referee.

Historically referees award more penalties to teams playing at home than visiting sides. However, it is hard to tell whether this is due to home advantage having an effect on the referee’s decision-making process or simply that teams are more inclined to attack at home.

If home advantage is not playing a part in a referee’s decision-making process then VAR would be expected to correct errors equally and, therefore, there would be no noticeable difference in the ratio of home to away penalties awarded.

VAR, penalties and home advantage: Serie A

As discussed, due to VAR being a new process in World Soccer, there is little data to analyse. However, Serie A has introduced the system for the 2017/18 season so there is now a full season of data to review.

The German Bundesliga also began utilizing VAR for its 2017/18 season but its implementation has not been seamless so it is difficult to use data from that league. VAR was not available for some games and VAR project manager Helmut Krug was sacked for ‘influencing the decision of the VAR in a manner not befitting his role’ in November.

In the 2017/18 Serie A season one decision was changed by VAR every 3.29 games. A total of 59 penalties, 16 red cards and 42 goals were changed through referral to VAR.

This is the penalty data from the last ten seasons of Serie A split into whether they were awarded to the home or away side:

This provides a reasonable insight into how the introduction of VAR could affect home advantage. The 2017/18 season, which saw the introduction of VAR, had the third-lowest number of penalties awarded to the home team and the most awarded to the away team of any season.

The number of penalties awarded to the home team for every penalty awarded to the away team fell to 1.16; the lowest of any season on record.

Whilst this is a small sample the dramatic change strongly suggests that VAR has weakened the home side’s advantage, at least when it comes to the award of penalties. Home teams were still awarded a greater number of penalties but only by a comparatively narrow margin.

VAR and penalties in the MLS

Interestingly, there is a similar trend over in the MLS; another league that has implemented VAR.

VAR was introduced mid-way through the 2017 season so it is probably best to ignore this data. Equally, it should be noted that the 2018 season has just a small sample since it is currently in progress.

Here are the MLS Home/Away penalty stats since the 2011 season:

It is common knowledge that home advantage has a bigger influence on results in the MLS than European Leagues such as Serie A. This is reflected in the higher proportion of penalties awarded to home teams with an average ratio pre-VAR of 2.15 home penalties to every one away penalty awarded compared to 1.75 in Serie A.

The effect of VAR has been even more drastic in the MLS so far with the ratio of home to away penalties awarded falling to an average of 1.22. This suggests the phenomena is not confined to Serie A and could demonstrate some weakening of home field advantage due to the introduction of VAR across leagues.

Why might VAR have caused the proportion of home penalties to fall?

It is difficult to isolate the exact reason why the introduction of VAR could have had an immediate effect on the bias towards home sides in terms of penalties.

It may be that referees are less inclined to make borderline decisions knowing that they can instead refer them to the VAR. Those difficult split-second judgements may be more influenced by the presence of a home crowd whilst the Video Assistant Referee is able to be more objective.

Equally, away teams may benefit from winning borderline penalties that the referee would not otherwise have considered awarding without the availability of VAR.

Could this have wider effects on home advantage in soccer?

If the imbalance in home to away penalties awarded was due to home field advantage affecting refereeing decisions then it is not unreasonable to suggest the same may happen to big offside and red card decisions in the long run.

This would certainly weaken home advantage to some extent but how big a factor VAR may ultimately prove to be is certainly questionable, especially considering VAR is not used for every decision so some refereeing bias will remain. A bigger sample size will also be needed to draw any real conclusions about VARs potential influence on matches especially since leagues have variable levels of home-field advantage tobegin with.

Still, despite being based on limited data, the potential rebalancing of penalty decisions should give bettors something to consider when backing home sides in matches featuring VAR.

Unlike most professional sports across Europe, there isn’t much disparity between professional and their amateur equivalents in the United States. NCAA football crowds and TV audiences aren’t far behind those of the NFL, while the betting figures are fairly even. The NFL often draws the majority of attention from September to January, but college football markets will actually rival its professional equivalent in terms of betting volume.

People often hold the assumption that NCAA football is the same as NFL and while the format of the game is the same, there are some important differences to consider when making NCAA football predictions.

NCAA vs. NFL: The key betting numbers

NFL betting markets are dominated by key numbers. Point spreads of 3, 7, 10 and 14 have long been choke points on a game by game basis with 6 and 4 becoming more common after recent rule changes. College football scoring follows the same fixed structure, but key numbers are far less powerful than seen in NFL markets.

Below is a table that shows the percentage of games that have fallen on the key betting numbers in the NFL and NCAA football over the last decade.

The explanation for this is straight forward. In college football, there are significantly more games played and the talent level in the country is much wider resulting in a larger range of outcomes.

Despite this clear difference and a reduced emphasis on key numbers, bettors will continue to pay extra and buy points on or off of 3, 7, 10 and 14 in college football – something that should be avoided at all costs.

Comparing NCAA and NFL game averages

The larger gap in team talent means that game averages are significantly higher in college football than the professional level. An average college game over the past ten years had 52 points scored, in comparison to just 43 in the NFL.

The average Over/Under set by bookmakers was slightly above 49 in college football, more than one touchdown higher than the average of 42.5 in the NFL. Point spreads vary too, in college, the average Handicap is 5.5 points while NFL is just under 2.5 points at 2.40.

The most notable average to consider when betting college football instead of the NFL is the home field advantage. The average home margin of victory in college football is 5.75 points, which is significantly different from 2.50 points in the NFL. While the overall number is slightly inflated due to early season non-conference games, the adjustment made to the point spread in college football for home field will often be much higher than the standard three points NFL.

Is there a difference in market movement?

Betting markets move significantly more in college football than NFL. Since 2008, 16.5% of all college football games had the point spread close three points different than at open. In the NFL the same occurred in just 7.5% of games.

Over/Under markets are more erratic in college football with nearly 1 in 4 games (23.68%) moving at least three points between open and close. Of the 2,103 such games, 1,105 (52%) moved three or more points in direction of the under.

There is significantly less movement in NFL markets with 7.1% of games moving three points or more. Bias to the under is much stronger with 165 of the 211 occurrences (65%) moving towards the under.

NCAA football predictions: Do teasers offer more value?

A benefit of betting college football at Pinnacle is the best teaser payout rate in the industry. A six-point two team teaser at Pinnacle will pay out 1 to 1 instead of the industry standard of 10 to 11. This reduces the required break-even rate of six-point teaser legs from 72.38% to 70.72%.

Basic strategy teasers have become extremely popular in the NFL due to their long-term success rate. The premise is to tease favourites of 7.5, 8 and 8.5 points down and underdogs of 1.5, 2 and 2.5 points up to teased point spreads which move across both 3 and 7.

Due to the variance in college football and lower rate of key number outcomes, many bettors have avoided applying the same basic strategy to college football markets.

Playing with a break-even percentage of 70.72% at Pinnacle instead of 72.38% elsewhere in the market does change things – specifically with away underdogs. Looking at away underdogs of 1.5, 2 and 2.5 points, the collective win rate on such teaser legs is 72.25%. All other variants of basic strategy teasers have proven to be break even or worse in college football over the last decade.

A successful professional tennis tipster by the name of @nishikoripicks on Twitter has been marketing his service on the tipster supermarket Pyckio, and has an impressive (and highly statistically significant) record of profits. However, he doesn’t beat the closing price, or at least not sufficiently so to suggest that his long-term expectation will continue to be beneficial for his customers after subscription costs are taken into account.

According to the closing line value hypothesis this implies he has largely been lucky. According to @nishikoripicks, such a hypothesis may not be wholly accurate. For this article, I have attempted to find out which position is more accurate by investigating the efficiency of the tennis match betting market.

At the heart of the closing line value hypothesis is the idea that the closing line or price of an event represents the ‘true’ price or probability of an outcome occurring because it reflects the most information known about it thus far.

In previous articles I have explored how, for soccer at least, the closing price is a very accurate measure of outcome probability that can be used to predict the profit expectation of a bettor, and how beating it might be used to test your betting skill.

The counter argument insists that the closing price cannot be wholly efficient, at least not for smaller samples of matches, if tipsters such as @nishikoripicks sustain high levels of profitability over thousands of bets. Although it is easy to be fooled by the law of small numbers and make erroneous generalisations from smaller patterns of bets, we should at least consider the possibility that @nishikoripicks’ performance is evidence of real forecasting skill.

There are a number of potential explanations for why a closing market might not be completely efficient. I’ve attempted to model how one might work in practice: a market could be efficient on average but inefficient at the individual price level due to heuristic biases, for example price anchoring.

Another explanation might be that the bookmaker intentionally biases their market to increase profitability at the expense of square (or unskilled) bettors. I’ve previoulsy explained how that might work for a point spread betting market.

Now, we can take a closer look at the tennis betting market and test it for evidence of efficiency.

Analysing the data

Miguel Figueres of WinnerOdds, the tennis forecasting specialist, has kindly provided me with a data set of opening and closing Pinnacle match betting odds for ATP, WTA, Challenger and ITF Futures tournaments covering the period May 12, 2015 to July 19, 2019. After walkovers and first set retirements have been removed (which are void by Pinnacle for betting purposes), a total of 68,361 matches and 136,722 individual betting prices were included for analysis.

I first tested the closing prices for evidence of the favourite–longshot bias, the systematic and disproportionate shortening of longer prices relative to shorter ones. Calculating an expected or implied win probability for each player (by inverting the odds), I categorised the data by 1% probability sub samples and compared each sample to the actual win percentage based on the match results.

The analysis is shown in the chart below. The yellow line represents perfect efficiency, i.e. players with a predicted 10%/50%/90% probability win 10%/50%/90% of the time. The blue dots and trend line show the actual data. For a bookmaker’s odds with a betting margin applied, the blue line should always lie below the yellow.

Despite a high degree of price efficiency, the divergence of the blue and yellow lines as odds trend longer is a clear sign of a favourite–longshot bias, albeit one that is fairly weak given the small margin typically applied by Pinnacle. For example, whilst implied outcome probabilities of 90% end up winning about 89% of the time, implied probabilities of 10% actually win only 6% of the time.

The bias is perhaps better illustrated in the next chart that compares the implied outcome probability to the ratio of the actual and implied percentages. 89% divided by 90% is 0.989 whereas 6% divided by 10% is only 0.6.

What happens when you remove the margin?

To better analyse the efficiency of the betting odds we need to remove the betting margin. The presence of the favourite–longshot bias implies that bookmakers, including Pinnacle, don’t spread their margin evenly across all outcomes. Disproportionately more is applied to longer prices.

The favourite–longshot bias is a well-established bias in numerous sports betting markets, and it is easily accounted for when removing the margin. Whist bookmakers don’t reveal precisely how they apply it, I have speculated on a number of methods which they might use.

For this analysis I have used the logarithmic function since that gives the closest match (yield = +0.47%) to an expectation of breaking even (yield = 0%) when betting every player to level stakes. By contrast, assuming the margin is applied equally across prices gives a sample yield of -2.73%, and -7.18% before it is removed.

Opening vs. Closing prices

In July 2016 I showed how the ratio of the opening price (or any pre-closing price for that matter) to the closing price was an excellent predictor of betting performance in the soccer betting markets. Suppose a price opened at 2.1 and closed at 2.0. The data implied that you would show a return of about 105% over the long term betting such prices, since 2.1 divided by 2.0 is 1.05. If your bet price to closing price ratio was 0.9, you’d expect to return about 90%. If it was 1.2, your return expectation would be about 120%.

Actual returns from hypothetically betting all soccer teams in the analysis sample correlated almost perfectly with expectation. The implication was that the closing price for those soccer matches was highly efficient (after removing the influence of the margin). Is this also true for tennis?

The final chart in this article compares expected yield with actual yields. For the blue dots and trend line, the expected yield is implied by the opening to closing odds ratio, after the margin has been removed from the closing odds. The actual yield is calculated from betting opening prices

A perfect correlation between expected and actual yields would show a trend line gradient of 1. In fact, it is less (0.825). The implication is that the opening to fair closing price ratio is not a perfect predictor of profitability and that closing prices therefore cannot be wholly efficiency or accurate.

Since the gradient is less than 1 this also implies that opening prices to do not fully converge to price equilibrium. In other words, prices that are too long do not shorten enough by closing; prices that are too short do not lengthen enough. I have attempted to explain such a phenomenon by price anchoring although there may well be others, including bookmaker price manipulation as I have mentioned.

Such a finding is consistent with @nishikoripicks failing to beat the closing price by the predicted amount. However, rather than this implying his profitability is more through luck than design, it could simply mean that Pinnacle’s tennis match betting prices he advises don’t fully reach their ‘true’ values by closing.

For the red dots and their trend line, the expected yield is implied by the closing price to fair (margin removed) opening price whilst the actual yield is calculated from betting closing prices. If closing prices were perfectly efficient this line would be horizontal with a gradient of zero.

The fact that is it weakly positive (ideally the blue and red line gradients should sum to 1) implies again that they are not, and further that there remains some residual information in the opening price that is predictive of actual outcomes.

Is @nishikoripicks a skilled tipster?

The data shows that @nishikoripicks beats the closing price on average by about 3%. If the closing line value hypothesis is correct this implies that after the influence of Pinnacle’s margin is removed, he has roughly a break-even expectation. Yet he actually has a 9% yield from 3,000 tips.

We now know that the tennis betting market may not be wholly efficient, at least not as efficient as soccer. We could speculate on the reasons for this. In addition to price anchoring or price manipulation, others might include the lower market liquidity and, for this analysis, the inclusion of lower ranked Challenger and Futures tournaments which have far less available information about competing players than the major ATP and WTA competitions.

Whatever the reasons, it might be reasonable to conclude that using closing prices in tennis might not offer a perfect means of predicting tipster or bettor performance.

The difference between break-even and 9%, nevertheless, is large, too large to be explained solely by the rather minor deviation from price efficiency observed in this analysis. Other explanations will be needed.

One may simply be that Pinnacle is unaware of @nishikoripicks’ activity. Another could be that both he and Pinnacle have very similar pricing models that exploit the biases and ignorance of square bettors.

From Pinnacle’s perspective, it’s more profitable to resist the temptation to move to ‘true’ closing prices so long as squares will continue to bet the wrong side of inefficient lines. From @nishikoripicks’ perspective, it will misleadingly look to followers of the closing line value hypothesis that he has simply been lucky.

Luck is not something we can ever completely rule out in betting. However, the longer that @nishikoripicks manages to sustain his performance well in excess of that predicted by the closing line value hypothesis, the more we will have to accept that it’s time to re-evaluate the use of the closing line as a predictor of ‘truth’, for smaller sports markets like tennis at least.

Seek to validate

After the games are complete on Sunday afternoon, I revisit each box score to validate performance. The majority of the validation occurs at a team level basis, but in some situations, it can get granular on a player by player basis.

My intention of doing this is to understand how teams performed and why the result ended as it did. For many bettors, it is easier to accept the final result than it is to understand the final result. Knowing how teams play is more important than knowing how teams played. In the example below, I will walk through my process, step-by-step, of reading an NFL box score using the NFL Week 9 game between the Arizona Cardinals and San Francisco 49ers.

Establish a baseline

I like to establish a baseline before beginning. What did the market open at, and where did it close? Movement in odds during the week are a reflection of opinions within the open market. I want to validate the difference of opinion and see if it holds in the market.

At Pinnacle, San Francisco opened as a -7 point favourite and were bet up to a closing price of -10.5. In this game, there was a clear shift of opinion supporting the San Francisco 49ers. My job as I work my way through the box score is to validate the support of San Francisco.

Look at the high-level numbers

I start with the basic high-level statistics to look at the game in a glance. The five numbers I focus on are total plays, total yards, yards per play, yards per pass and yards per rush. Total plays show who controlled the football, total yards shows the output, yards per play shows the overall efficiency of those plays, and yards per pass/rush show specific play output.

The numbers for San Francisco vs. Arizona were:

I like to take the sum of the total yards (768) and divide by 15 which is the league median for yards per point. This calculation puts the game into perspective from a pace and production standpoint. In this game, there were 53 points scored which were 1.8 points higher than the expected output from the league median number.

I then like to take the differential in yards per play (Arizona +1.1) and divide by 0.2, which comes from a regression run on scoreline compared to yards per play differential. In this game, the yards per play differential suggests Arizona should have won by 5.5 points which is eight points different from the result.

At this point, I am looking at this game as near equal to the scoreline and total output, but potentially misleading showing a 49ers victory. Anytime there is a discrepancy, I want to seek to validate and determine why. Often it is several fortunate circumstances working in favour of one team or yardage gained late in the game to pad the numbers.

Throwing the football can be a good indicator of a box score being misleading – or not – as passing yards are significantly more efficient than rushing. In this box score, Arizona were ahead by 1.1 yards per play, but San Francisco were 0.6 yards per pass better.

Considering game state

After evaluating the high-level stats, I begin to look at a drive-by-drive breakdown. The first thing I look for is how many drives occurred in a non-neutral game state. A neutral game state is any play or drive that occurs when a lead for either team is seven points or less.

Non-neutral game states – especially late in the game – often result in drastically different play-calling both for the team leading and trailing. For a team with a lead of eight points or more late in the game, it is more beneficial to exchange time for yardage to increase the probability of winning.

Box score performances can be misleading if a team gained a large portion of their yardage in a negative game state (trailing by eight points or more) in the second half as they frequently will see more conservative defensive schemes and coverages.

In this game, there were nine drives in the second half. Each drive except for the last of the game started with San Francisco leading by eight points or more. Arizona gained 245 (68%) of their 357 total yards in a negative game state.

Great teams earn a big lead early in the game and hold on to the lead for the duration. Despite losing by just three points, my take away is that Arizona gained just 32% of their yardage in a neutral game state and never earned a positive game state in comparison to 100% for San Francisco.

The influence of big plays

Any big plays gaining a large number of yardage at one time will skew the per-play numbers. Explosive offensive play is an asset, but parsing out large plays can be an indication of luck within games.

On the second to last drive of the game when trailing by 11 points, Arizona gained 88 yards and seven points on one pass. The other plays in the game that gained more than 20 yards were 36, 21, 21 and 20. In order to see how much of a difference that play made in the final numbers, I like to remove it and recalculate the five main stats. The below table is the same as the above but with Arizona’s 88-yard pass to Andy Isabell removed.

By parsing out the one big play, the game takes on a very different look that reflects the state of the other 118 plays. The total yards in the game is equal to the actual points scored (45 vs 45.2), the projected scoreline dividing the yardage total for each by 15 is San Francisco 26 – 17 Arizona, compared to 28-17.

The yards per play differential of San Francisco +0.6 suggests a 49ers victory by three points, reflective of the actual score, but not of the box score above. As mentioned above, the game state dictated a lot of the play calling within this game.

The San Francisco 49ers were in a neutral game state for 34 of their 69 total plays. In those 34 plays, San Francisco gained 7.4 yards per play. Substituting the 7.4 for the 6.0 makes a differential of 2.0 or a 10-point expected winning margin, which is equal to the winning margin produced by the yardage gained.

Turnovers, short fields and defensive scores

If the game state stayed neutral for the full game and there were no big plays to parse out, chances are any discrepancies can be explained by turnovers, short fields or defensive scores.

Teams that win the turnover battle win 80% of games in the NFL. Identifying the turnover differential is something that can explain a lot of any discrepancies in the box scores. A turnover will often result in a short field, or reduced number of yardage to gain for a touchdown or a defensive score.

Often teams will be opportunistic taking advantage of turnovers and a short field to score at a higher rate. Arizona and San Francisco did not have any turnovers, so evaluation of the box score is clean in this respect.

Red zone inefficiency

Although the media marks wasted red zone chances as an inefficiency, continuous drives into the red zone can often suggest a good performance. I like to look at the length of each drive and see if a team was lucky or unlucky near the opponents’ goal line.

Each box score in the high-level stats will have a red zone conversion rate which expresses the number of drives inside the opponents 20-yard line compared to the number of touchdowns scored. The media focuses on the conversion percentage, but I prefer to focus on the attempts.

With more data available, it has become evident that red zone conversion is quite random, but red zone attempts are reflective of skill. From an expected points basis, the more snaps you take close to the opponents’ goal line, the more will you score.

I like to focus on the yard line of where first down snaps occurred. San Francisco were 2-2 on red zone trips, but Arizona were 2-3. They had a 1st and ten from the San Francisco 15-yard line but settled for a field goal from the San Francisco 18. This outcome is a minor example of a team failing to capitalise on an opportunity leaving just shy of four expected points on the field.

Why these considerations matter in NFL betting

The goal of working through the box score was to validate the move from San Francisco -7 to San Francisco -10.5 through the performance within the game. By settling for looking at the final score and high-level stats, bettors can make a strong case that the move was unwarranted and the Cardinals should have won the game by multiple points.

However, by working through the box score, it becomes quite clear that San Francisco dominated the game from start to finish and Arizona was the benefactor of one big play, and much of their output was from the most favourable situation in a negative game state.

In total, the above work took me between 15 and 30 minutes. I can work through an entire Sunday in two to three hours. I highly recommend it to all betting on the NFL.

To summarise

Winning a bet is a nice feeling. Losing, however, is considered to be twice as painful, psychologically speaking, as winning is enjoyed. Frequently, a bettor’s response to losing, particularly a run of losses, can lead to reckless behaviour, either betting more often or betting more money, in an attempt to recover their losses.

Even for potentially skilled bettors with positive expectation on their side, losing runs could lead them to irrationally question the efficacy of their betting system. It’s much easier to doubt yourself after 10 from 10 losing bets than it is after 10 from 10 winning ones, even though in statistical terms they will have very similar outcome probabilities. Hardly anyone would bother to question an over-performing system.

I’ve previously written about betting drawdowns and how they can be managed. In this article, I want to complement that work with a simple modelling of losing runs, and more specifically how long we might expect one to last.

I have limited myself here in order to keep things simple. I will use a consecutive run of k losing bets in a sample of n bets with the same odds, although there is no reason why one couldn’t extend this to cover more complex losing periods and more complex betting records with variable betting odds using a Monte Carlo simulation. At least with the simplest losing run, however, we can offer some basic mathematical formulae to describe them, difficult with more complex scenarios.

The probability of losing

Let’s consider a bettor who is skilled enough to break even over the long term. In other words, they are achieving fair odds. As such the odds will then reflect the ‘true’ probability of a result. In fact, what follows doesn’t really change very much for either unskilled bettors betting to the margin, or skilled bettors managing to find marginal profitable expectation, the reason being that the vast majority of what happens in betting is a consequence of randomness. Odds of 2.0 imply a 50% probability, odds of 4.0, 25% and so on. The probability of winning k consecutive bets each with odds o is then given by:

For example, the probability of winning five consecutive fair even-money bets with odds of 2.0 is 1/32.

But here we are interested in losing bets. For odds of 2.0, the odds of losing would be the same, also 2.0, since the probability for winning or losing is both 50%. More generally, the odds of winning and losing are not the same. Given that the probability of losing is 1 minus the probability of winning, it follows that the odds of losing are given by the expression:

Thus, the probability of losing k consecutive bets each with odds o is given by:

The expectation of losing sequences

What is the probability of having k (or more) consecutive losses in a sample of n wagers of odds o? It turns out that the mathematics for this is not trivial, and well beyond my pay grade. But we can ask the question in a slightly different way, and for that the mathematics is much easier. Instead, let’s ask how many times we can expect to lose k consecutive times in a sample of n bets of odds o?

Consider a simple example. How many times can we expect to lose three bets in a row, each with odds of 2.0, in a series of 10 bets. We already know that the chance of singularly losing three in a row is 1/8. However, in a series of 10 bets there are numerous opportunities to build a three-bet losing sequence. It could happen on bets one, two and three, or on bets two, three and four, and so on up to bets eight, nine and 10.

In this example there would be a total of eight possible sequences, so the expected number of time we would have such a sequence in 10 bets is 8/8 or one. In other words, on average we would expect to have one three-bet losing sequence every 10 bets. Sometimes there will be more, and sometimes there will be none, but the average will be one.

More generically, the number of possible sequence positions in a series of n bets is n – (k – 1) or n – k + 1.

Hence the expected number of losing k consecutive bets, let’s call it e_k, in a sample of n bets will be given by:

As the number of bets, n, increases, for smaller values of k (and k will always be much smaller than n for losing sequences that are realistically possible and which we are interested in), e_k will tend towards:

For a sample of 1,000 bets with odds 2.0, for example, the expected number of five-bet losing sequences will be 31.25 (31.125 using the more precise formula), rounded to 31 as the nearest integer number.

Since the number of bets, n, is roughly proportional to the expected number of losing sequences of length k, we would expect about 62 five-bet losing streaks in 2,000 bets, and around 93 in 3,000 bets.

When e_k = 1, we can describe k as the length of the longest losing sequence that one will typically expect to see in a sample of n bets. Why? Well, less than 1 and we don’t see it and more than one, and there is opportunity for a longer sequence of losses to occur fewer times.

Thus, when n >> k and e_k = 1:

And rewriting:

where  represent the base of the logarithm.

For 1,000 bets with odds of 2.0 the expected longest sequence of losses will be log₂1000 = 9.97 or 10 to the nearest integer. In other words, in a sample of 1,000 bets we can typically expect the longest losing sequence of losses to comprise 10 bets.

For odds of 3.0, the expected longest losing sequence would be 17, and for odds of 5.0, 31. Odds of 5.0 are quite typical for racing bettors. Do you think you could handle a sequence of 31 consecutive losers without getting cold feet about what you were doing?

I’ve run a 10,000 iteration Monte Carlo simulation to test the mathematics for e_k. The table below compares the results for different values of k. There is an almost exact match between the predicted occurrences of losing sequences based on the mathematical formula above and the Monte Carlo simulation.

In the chart below I’ve plotted the relationship between k and e_k for different win odds. The y-axis, e_k, is logarithmic. The straight line confirms that k is inversely proportional to the logarithm of e_k which is exactly as would be expected given the mathematics. The point at which each line crosses the x-axis (at e_k = 1) is the expected longest losing sequence.

Given the approximation for k above, the expected longest losing sequence in a sample of n bets is thus also proportional to the logarithm of n, as depicted in the next chart. Thus, k doubles with each squaring of n.

The probability of a losing sequence

Knowing expected numbers of losing sequences is one thing, but we still don’t know the probability of them happening. As mentioned previously, the mathematics for this is non-trivial because the frequency (probability) distribution for the number of losing sequences of length k in n bets is not at all obvious and will be different for each vale of k.

For example, we might know that on average we will see about one losing sequence of 10 bets in a sample of 1,000 even-money bets, but that’s just the average. Often, we will see none, sometimes two, occasionally five or more. Instead it’s much easier to resort to our trusty Monte Carlo simulation.

With a 10,000 iteration Monte Carlo simulation I counted the number of times each losing streak of length k was not observed. For example, for k = 10 in a sample of 1,000 even-money bets the maximum losing sequence was shorter on 6,086 occasions, and 10 or more on the remaining model iterations.

Relying on the law of large numbers this implies that seeing a losing sequence of 10 or more bets has a chance of roughly 39%. This seems intuitively about right when we remember that the expectation is to see a 10-bet losing streak about once in such a sample. The chart below shows how the probability of seeing a losing streak of length k or more varies with k.

Evidently the larger the sample of bets, the more likely it is that something bad is going to happen at some point. We know that the probability of 10 consecutive losses in 1,000 even-money bets is 39%. What would it be for smaller or larger samples? I ran another Monte Carlo simulation to find out. The chart below is for k = 10.

We can re-run our model for any value of k or any set of odds. Another example output is shown below for odds of 3.0 and a losing streak of 17 or more.

Analysing losing runs with real world betting histories

Thus far, this analysis has been rather theoretical in that it considers only samples of bets where the odds are all the same. That might be a reasonable assumption for point spread and Asian handicap bettors, but less likely for money line and fixed odds bettors who may be betting a much wider variation of odds. For example, my Wisdom of the Crowd betting system has matches with odds as short as 1.11 and as long as 67.0, an average of 3.9 and a standard deviation of over 4.

Certainly, we could use a Monte Carlo simulation to determine our expected loss sequences, but is there a way to use the mathematics? Yes, we simply need to be careful we use the appropriate value for o, the odds. We cannot use the average odds for our sample because this will be disproportionately weighted towards the longer odds.

Instead we should use the inverse of the average of the implied probabilities for all odds. For example, if our sample has five bets with odds of 2.0, 3.0, 5.0, 10.0 and 20.0, calculate the implied probabilities (0.5, 0.333, 0.2, 0.1 and 0.05), take their average (0.237) and invert (o = 4.23).

I performed this for my Wisdom of Crowd sample history, a sample of 9,436 bets. With a calculated value of o = 2.66 using the method above, there was an excellent match between the expected values for k and actual losing sequences.

The maths predicted 898 losing sequences of at least five bets, but there were actually 889 of them. Likewise, for k = 10, the maths predicted a figure of 85; there were exactly 85. For k = 9, prediction was eight, actual was nine. The expected longest losing sequence (e_k = 1) was predicted to be 19. The longest losing sequence was indeed 19, and there was just one.

What about winning sequences?

We can use the same mathematics to analyse the expectation of winning sequences. Indeed, it is even simpler since we can use the win bet odds directly in the formulae rather than adapt them to reflect the odds of a loss. Hence:

And for e_k = 1:

However, for samples of variable odds, don’t forget to use the appropriate figure for o, not the average odds, but the inverse of the average of the implied probabilities.

What have we learnt about losing runs in betting?

Given enough time, bad things happen in betting. If nothing else, hopefully this rather theoretical analysis of betting sequences will serve as a reminder that the longer you bet for, the more likely it is you will see longer and longer losing streaks.

In addition to finding expected value, the task for any successful bettor is to manage their expectations sensibly and learn to cope with the inevitable bad runs that have the capacity to significantly influence bettor psychology. Knowing what to expect and how to estimate it at least gives you some preparation in that respect.

Ruiz Jr. vs. Joshua: Tale of the Tape

Joshua vs. Ruiz Jr. first fight statistics

Ruiz Jr’s greater output and accuracy is clearly evident in the post-fight statistics. The Mexican threw double the amount of power punches than Joshua, and while Joshua threw 26% more jabs than Ruiz, he landed 12% less, which suggests his accuracy was widely off in the first fight.

Why did Joshua lose the first fight?

The first fight saw Andy Ruiz Jr. reach odds of 17.65 at Pinnacle, making it one of the biggest heavyweight title upsets over the past 30 years. Previous bouts in boxing history across the betting industry have seen Buster Douglas knock out Mike Tyson (43.0), Corrie Sanders destroy Wladimir Klitshckho (21.0), and Hasim Rahman knock Lennox Lewis out back in 2001 (21.0), which highlights just how unexpected Ruiz Jr’s victory was.

Despite this, Joshua was beaten by the better fighter on the night, and exposed in certain areas of his game. The former Olympic champion didn’t underestimate Ruiz’s power, but did underestimate his powers of recovery. When the Brit dropped Ruiz in round three he hurriedly tried to finish the fight, which is one of the biggest mistakes a fighter can make when facing someone who possesses much faster hands than himself.

When facing a boxer who has faster hands than you, and when that fighter is also a very good counter puncher on the front foot, the last tactic they should be considering is to go into exchanges up close. Joshua looked extremely nervous and boxed in a completely reckless manner, which suggests he had doubts about the fight prior to the opening bell.

Evident flaws were also exposed in Joshua’s game and there now hangs a big question mark over the former champions punch resistance. He did show plenty of heart, but his survival skills and powers of recovery were extremely poor for a unified heavyweight champion. It seems the Londoner is a fighter who when he gets hurt he stays hurt, and Andy Ruiz Jr. realised this and jumped on him to finish the fight in spectacular fashion.

How good is Andy Ruiz Jr.?

Ruiz Jr. is not the second coming of Mike Tyson all of a sudden – but he was a quality fighter long before beating Joshua. His physical appearance isn’t the well-proportioned, chiselled figure that boxing bettors are accustomed to seeing, and until you have observed him fighting in the ring, it is easy to dismiss him as a fighter who isn’t devoted to his craft.

As an amateur Ruiz compiled a record of 105-5, his 105 wins include two Mexican National Junior Olympics gold medals, whilst representing Mexico in the 2008 Beijing Olympic Games qualification tournaments.

In the professional ranks, the 30 year-old has climbed up the levels slowly, but outside of dedicated boxing fans had largely gone unnoticed. A WBO world title eliminator was pencilled in against Hughie Fury in 2016 only to be cancelled, so the Mexican went on to challenge WBO champion Joseph Parker, a fight he lost by an extremely close majority decision in New Zealand (Parker’s home country).

Against Joshua, new environments, non-stop media work, and continuous questions regarding Tyson Fury and Deontay Wilder in the pre-fight build up gave Ruiz.Jr an edge, but improved training, and being the challenger this time around, will show whether the former Olympic champion was not at his best, or if Ruiz Jr. simply has the perfect style to beat him.

Top fighters in the past have found ways to make adjustments to opponents’ styles to avenge previous losses. The issue with Joshua remains how well he can come back from humiliation, and how quickly can he adapt his style in such a short period of time to make this effective in the rematch.

The 29 year-old has been in plenty of big fights, but they were fights where he was in his own comfort zone, in familiar surroundings, buoyed on by a pro-Joshua crowd. Combine these changes with the huge pressure that he carries on his shoulders in just 22 fights, and there was always a fair probability that an upset was possible in the last fight against a fighter who is very good at fighting inside – and very well-schooled as a boxer.

Andy Ruiz Jr. vs. Anthony Joshua betting: Where is the value?

For Joshua to retain his belts he simply has to keep the fight at a distance he is comfortable. The jab and applying his physical advantages is the key to victory, and the Brit will need to produce a similar performance to how he boxed Joseph Parker – keep it long, don’t engage, and spoil Ruiz up close when he tries to get into range to let his hands go – a tactic that enhances the chance of the fight going the distance.

Joshua has the natural assets (height and reach) to be able outbox Ruiz, but with that you need the concentration, discipline, and the ability to not get drawn into a fight, especially when hurt. If the former champion hasn’t learnt and adjusted his style from the first fight, then the result will be the same, and possibly even more devastating, so it is important that bettors keep a close eye on both fighters’ pre-fight preparations.

Wladimir Klitschko was stopped in a very similar fashion to Joshua by Corrie Sanders back in 2003, but then managed to go 20 fights at championship level without being dropped, largely thanks to the safety first approach that Emmanuel Steward installed in the Ukrainian. Joshua, like Klitsckho, needs an adjustment to his fighting style to nullify Ruiz Jr’s strengths, as it was clearly evident in the first fight how poor his survival skills were when he was hurt. It might make the Brit less entertaining to watch, and it might mean that the rematch is far less exciting than the first fight, but ultimately it is the strategy he will need to adopt if he wants to be victorious.

Ruiz Jr. will have the mental edge, after dropping Joshua four times and climbing off the canvas himself to do so, but bettors should expect Joshua to be much better prepared this time around. Providing he has the challenger mentality back, then his speed, power, and explosiveness should be enough to gain his titles back, in a fight that should go longer than expected in the betting odds.