Football Picks - My Review (Part Two of Two)

This is the second part of the post I began below:

https://sportspicks.locals.com/post/7973588/football-picks-my-review-part-one-of-two-now-that-the-2025-26-club-football-season-ended

The biggest change in the official picks from last season to this season was a reversion to moneyline betting from spread betting. Moneyline betting was the main format for the first three seasons in Sportspicks before it became common in the second half of 2023/24 and then the dominant mode in 2024/25. My observations from last season had been that spread betting creates a blind spot whereby it is impossible to directly benefit from an edge on the draw in circumstances where both teams are separately overvalued, particularly when said teams are evenly matched and the median spread is a draw no bet. This aligns with football not having key numbers per se, since every moneyline is a key line. When I have thought about it more, this does not necessarily invalidate value on the draw no bet, but it may not be the most efficient way to bet a game. I have not benchmarked this seasons picks against their spread equivalents because there is no one to one mapping between three way moneylines and two way spreads but I do have reason to believe that spreads have their place. The moneyline for a team to win is by definition the -0.5 spread for that team and a team not to win is by definition the +0.5 for the other team. Indeed, the pick'em spread is the same as the draw no bet, directly constructable through a combination of either win and draw or win and same team not to lose to form a synthetic spread. Since this is in effect a two way moneyline, the only circumstance in which it holds no value is if the line is accurate to within the overround. On prediction markets, this is possible to eliminate entirely via limit order. When I wrote last year's review, spreads were still very new to prediction markets they had poor liquidity. That has changed and now you can access a series of game lines that previously you would have had to pay a premium for on standard bookmakers. The only caveat is that due to the nature of prediction market contracts in their current form, contract value must be constant at resolution, thus making a true push currently unavailable and therefore the only spreads available are half integer spreads. It is still possible to form synthetic integer spread from combinations of the half integers either side but this is a secondary concept and is not foundational to finding value.

I have also been deploying a system of alt lines on the official picks in order to optimise price entry, take advantage of liquidity flow and reduce variance while maintaining the underlying edge. My premise was that when one team is sufficiently clear a favourite (the threshold for this is trading at around 50c or so but it can vary), the draw price and underdog to win price move covariantly so that when the favourite becomes more expensive, the draw and underdog win become cheaper and vice versa. When there is no clear favourite a change in one price does not distribute consistently between the other two, that is to say, they may move covariantly or contravariantly, which is why I tend not to do this for more evenly matched teams. So when the official pick is for such a matchup to end in a draw or an underdog win, I buy no on the favourite via a limit order. Since public betting action slants towards favourites and visible order flow on Polymarket supports this, you have a better chance of being filled on a limit order as their is generally sufficient demand to buy yes on favourites, which translates into having a reciprocal order of no on favourites filled. This not only gets you a price one cent cheaper on the order book, you also avoid paying extra fees. On Kalshi this means you will pay less fees as a market maker than as a market taker and on Polymarket you will get paid a maker rebate, which gives a slight net reduction to your entry price. It also reduces the component of accurately identifying which of draw and underdog win holds the greater edge if indeed both holds the edge. Long shot bets from a sheer binomial probabilistic point of view have lower variance but from bet to bet they can be very feast and famine. A couple of last minute goals on last minute bets can obfuscate the edge even over moderate sample sizes.

It's tempting to say well, let's increase out sample size because that will reduce variance but football is a very complex, every changing sport and no two games are truly the same. So there is a compromise between bets that cluster around the 50c mark and those that drift towards the margins in order to balance minimising probabilistic variance while at the same time not overexposing oneself to feast or famine. This is why I have increasingly adopted a hybrid approach of taking moneyline when there is not a clear disparity between the teams, the no on favourite when there is a clear gap as explained above and when an underdog is somewhere in between, I may take a synthetic draw no bet (I don't have hesitations about mixing draw and win there as at that stage I see it as a variance reduction mechanism). Under certain circumstances, the spread may be the higher value play. By this I mean the median spread. The moneyline is very often an alt line and alt lines are generally long volatility. The median spread is more useful when there is an edge but it is lower volatility. Where this might be useful I haven't fully sketched out but there is a difference both qualitative and quantitative between the middle of a distribution and its tails. Some teams will be heavier in certain parts of that distribution than others. This is observable in the Champions League knockout stages when the first leg scoreline becomes a baseline for the second leg, thus creating key numbers that originally did not exist.

For benchmarking the official picks against the alt lines, I took every price as quoted in the official picks as a base line. When applying an alt line I took the closing price of the favourite on Polymarket and subtracted it from 100 to get the no price and rounded down if reported as a half integer number as this would correspond with a limit order. Over the course of the season I noticed that it yielded a substantial increase in returns. Some of this may be explainable by variance as sometimes a swing of one goal makes a difference of over half a thousand to the bankroll, sometimes closer to a thousand. I can say with confidence that at the very least the alt lines give a smoother yield curve over time.

One more subtlety is timing of entries. Towards the end of the season official picks were given out early in the week as lookahead picks. Although I do like this, it was not as simple as get your entries in early and you get a good price. Although the sample size was small, I did notice a fair bit of drift towards favourites. I am sceptical that closing line value is as important as many bettors make it out to be and so I have no qualms about failing to beat closing line value if I think the underlying edge is good. Nonetheless, I am still concerned about entry optimisation as better entries add up over time. Drift towards favourites is a very observable phenomenon and in football there is not such a clear distinction between sharp and public action as there is in American sports. Due to the global nature of the sport bookmakers often take on very lopsided action which can mean that line movement towards favourites is not necessarily predictive signal from sharp action. It may possibly even have a component of time decay as the closer you get to a game, the less time there is for an asymmetric probabilistic event, for example an injury to a star player to level the playing field. Whatever myriad factors influence line movement, I have noticed that favourite lines are the most elastic while draw lines are the least elastic. Therefore, I do very much see value in the lookahead picks for securing good draw entries, as draw markets are generally not very liquid and it can take several days to get filled on a limit order to buy the draw. It is still useful to have lookahead picks that are a more straightforward favourite fade as you can still place a limit order deeper into the order book or wait until closer to the time or even just place it now if you think the favourite has met a resistance point anyway. This does require discretion though but timing of entries remains an open question that I will hopefully have more data on next season.

Finally, on international tournaments. These are difficult to model because international football is a very different beast to club football. I don't have much meaningful international football data from this season that may be informative for the World Cup this month and the time between tournaments means a lot can change in that time so I won't extrapolate anything from the official picks. I will be doing a separate write up on own betting strategy on the World Cup but based on my studies, international football has far less settled identities, far less predictability of iterations and far more emotional volatility, especially in the World Cup, which is a level of pressure above the Champions League. What I will be looking for will be signs that a team is able optimise for tactical coherence without undue complexity, clarity of roles on the pitch and calm game state
management in high pressure situations. Despite a rough start to the football season the early October adjustments saw a major bounceback and I would say that this season's picks have found a lot of value. We still have one more tournament to go and it's the biggest in the world so hopefully we will finish this season on a high!