Getting to grips with betting in PhotoFinish.Live
As the presence of betting increases in PFL, it's time to dig a little deeper into how betting works, and how you can take a sensible approach to it in game, whether you own horses or not.
ai_paddock
11/26/202415 min read
Horse racing is synonymous with gambling, but given the depth of breeding, racing and shit posting on Discord available, gambling seems to be the forgotten second cousin of PFL to some players. Maybe one of the reasons is people don't trust the algorithm. Or they think racing horses for real money prizes is all the gambling they need.
There are a number of reasons gambling, at an affordable level, without feeling the pressure to bet big bucks, can add more to your enjoyment of the game. It's even fun without spending money on breeding and racing, and maybe a way for long time lurkers to get their teeth stuck into the ecosystem without any longer term commitment. It's even an enjoyable experience for people who didn't know what PhotoFinish.Live was 10 minutes ago and will forget again in 10 minutes time.
Gambling is not easy. Anyone can get lucky, but the goal of any gambler is to make profit for the long term in a reliable fashion. Gambling in PFL is parimutuel, which means the odds you get are based on the weight of money. Everyone bets on the same horse, the odds will be very bad. PFL juices pots to give punters extra incentive to bet. That added value, whilst much appreciated by players, does distort the prices that are available on the horses. Given the money is added evenly to horses it limits the appearance of long-shots. It's rare to see any horse achieve long odds because of the way the juice flows. This affects how we approach betting strategies. This is not a post on how to gamble on horses, there are many great sources of this information on the internet. One of the most obvious, and still one of my favourites is this documentary on Bill Benter and his success in Hong Kong, which has many parallels to how gambling works in PFL.
From my own study of data in this game, I think predicting race outcomes is a fair and robust challenge for any player. Many of the variables of real life, the draw, or jockeys, form and fitness are not present in PFL. However there is enough randomness to keep it challenging. What's more, many of the races available to bet on are the hardest to predict. The majors contain upto 20 horses. More horses, more variance, or to put it another way, you are more likely to get tripped up by sheer luck. There is much less variance in a four horse race, but betting opportunities on these races are fewer and far between. Given the way juice is distributed, you can't access fair prices for outsiders in majors, you are kind of limited to 22-1 shots, where some horses at fair odds would be priced up at 250-1. So I would say a strategy based on picking outsiders is difficult to pull off in this game.
Like most of these infuriating posts, rather than give you the answer (as if I had it) I am going to provide insights from my experience betting on PFL and you can do what you like with some, none or all of it. Remember gambling should be fun, never bet more than you can afford to gamble. You want to avoid long draw downs, which is difficult if you single bet 20 horse races, and don't let gambling losses undermine you breeding and racing strategies. So here are some things to consider (don't say tips) when gambling on PFL.
Before we dive in, a short intro to pricing of bets in PFL. In this race, which at the time of writing is due to run today, we have four horses, in theory the easiest type of race to predict. Below on the right we can see that the pot has been juiced at 1.6k Derby per horse and their is currently 400 Derby bet by players on post 3, AbSOLutely For Sale, brining its total to 2k Derby. As a consequence her odds are 2.89 versus 3.61 for the other three horses. So the favourite has an implied 1/2.89 chance of winning, or 34.6%. Well, not quite. These figures have an edge for the house contained in them. If we add up the four implied chances of winning (34.6% + 3 x 27.7%) we get to 117.7%. The 17.7% is variously called the house edge, overround, vig or whatever around the world. That is the margin the house keeps for running the event. As gamblers we need to overcome this edge to win consistently. The figure of 17.7% is calculated by dividing the true odds by 0.85 for each horse to create the price. Remember the odds themselves are dictated by the weight of money. Number 3 has more money placed on it, so it has lower odds, but that could change. It's important to understand mechanically how the system works before we start experimenting with our hard earned cash. Betting is suspended 10 minutes before the start of any race. Regardless of when you bet, you get the odds when betting is suspended 10 minutes before the off.
A list of somewhat useful insights into gambling on PFL
Check out https://pfldrf.com/ - the best resource for information about today's bettable races
This is the resource for studying the form of today's runners that you can bet on. There is a lot of great information here brought together into one screen, just like traditional racecards. Remember, some of the things you might look for are just not part of the alogrithm in the game, so you need to unlearn your IRL experience to get better at betting in PFL.
Use the Morning Line prices to your advantage
I have written about ML prices at least twice on this blog. My personal opinion is that they are pretty good at explaining the underlying quality or skill level of a horse. There are obviously events that are run or rolled that contribute to how a horse runs on the day. Maybe this is where temper and heart come in as modifiers on a horse's performance. Maybe each horse get a luck roll, and skill + luck combine to make the race less predictable. Just like in real life favourites lose all the time, but ML prices are a good indicator.
Below is some data from Season 20, with all races containing exactly 12 runners (this data is used for all subsequent tables). You couldn't have bet on all these races, it is purely for illustration. If all we have to go on is Morning Line prices, how well do they predict the winner?
The above table covers over 2k races. price_rank is the order the horse is in the ML prices for each race, with 1 being the favourite and 12 being whatever the opposite of the favourite is (rank outsider I guess). The ML favourite wins over 17% of the time, which in a 12 horse race is more than double the random chance of picking the winner (1/12 or 8.333%). So all those people that tell me on Discord that ML is BS because their horse was a ML fav three times in a row and it didn't win once, your chance of winning at least once in those three races, given these stats is only 42.8% (1-.17^3).
If you could place bets at ML prices, congratulations you could just bet on the favourite and be 26.6% up. A 1 Derby bet on each race would yield you 610 Derby in profit. Sadly we can't bet at ML prices, and we can't bet on all of these races. Those pesky other players that know this insight, if they are rational, would bet on the ML favourite, destroying the odds for everyone, until we will all gave up. Nevertheless its good to know that ML prices are somewhat connected to reality. If you compare the win% column to the implied column (which is calculated as earlier by dividing 1 by the price) what you see is that the implied chance of winning for the favourite at 13.4% is significantly lower than the actual win rate. These are situations that bettors love. If you knew the closing line price of an event, and you can regularly beat it, congratulations you are one step closer to being a pro. This suggests we should prefer ML favourites when we don't know anything else. If the price on offer is higher than the ML price for a favourite - there might be a value there and a good entry point. At the other end compare the winning percentage of the outsiders to their implied prices. Outsiders are treat far too optimistically by ML prices than real life would suggest. The outsider wins 2.8% of the time - which means its true odds should be 35.7. Good luck finding those odds. It is rarely profitable to bet on outsiders in PFL because of the way pots are currently juiced. The ML prices and the juice overestimate the likelihood of an outsider winning. In this example the ML prices are giving odds of 13.5 for what we know is a 35.7 bet. You won't get rich taking those odds.
The insight here is that whilst it would be great to predict the winner of every race, that is unlikely. A smart bettor should be looking for the biggest gap between what they think the true odds are and what's on offer. The ML favourite might be the most likely to win, but if the odds on offer are 3 and the second favourite is at 15, the second favourite is by far the better option to bet on. Looking for positive EV bets is the name of the game, and you can only tell that by looking at the odds you are given.
Past behaviour is the best predictor of future behaviour
In any walk of life, when it comes to predicting the future, the past is the best place to look. People are creatures of habit, and so are horses in PFL. Whilst not as individually predictive as the ML price rank, historic behaviour is fairly predictive. The first thing to look at is win rate - wins / runs. Good horses win 20%+ of the time, great horses win 30-50% of the time, outliers win more often than that. Statistically the best way to look at win rate is to actually look at the percentage of the time a horse misses the podium, this is more predictive than win or podium (top 3) rate. Bad horses don't hit the podium and its a big red flag for the prospects of a horse. This data is freely available on the form guide and on screen, and you just need to be able to calculate 1 - [(wins+places+shows) / runs]. What you can see is that the lower the non podium rate, the more likely the horse is to win. It's clearly not as strong as following the ML prices, but horses with high non podium rates rarely win (about half as often). Remember, we are using ranks not the actual rates because races are pvp events, and it doesn't matter what your actual rate is - its where it ranks against the competition that is more important in predictive modelling. Below betting on the three horses with the lowest non podium rank would make you a profit if you could get on at ML prices (you can't).
Beyond winning and losing
Whilst studying podium rate is a good metric, there are better metrics out that explain what is going on and capture more signal. The problem with win rates is that they can be manipulated. Some owners are great at racing their horses only in 4-5 horse races, artificially boosting their win rates, only to disappoint the horse player in that big 20 runner major final. Always eyeball the history of a horse to ensure it hasn't been selectively run in small, uncompetitive fields.
A better metric than win rate is percentage of horses beaten. This is neutral to field size, and quite literally determines across a career what percentage of horses that have previously been beaten. IRL betting this is a great feature, and it is in PFL too. To calculate it is a bit more effort. For each race you need to the following; horses_beaten = 1 - [(position -1)/(number of runners -1)]. So a horse wins a 10 horse race, percentage of horses beaten = 1-[0/9) = 100%. if it came second it would be 1-[1/9]=89%. Extremely great horses beat 80% of the opposition. Good horses beat 70% of the competition. Ultimately the horse with a clearly higher percentage than the competition has the best chance to win a race, regardless of the actual percentage.
A similar and simpler metric to calculate is the average finish position. I prefer % of horses beaten because it doesn't get biased by field size, but if you are staring at a race you want to bet on - quickly looking at average finish position is a good ready reckoner.
Even more complex metrics
So what if a horse beat a bunch of crap horses previously, what matters more is how historically a horse has performed against the rivals it is running against today. This is clearly a more complicated metric to calculate, but guess what? Understanding head to head performance is also predictive of which horse will win a race. If you compare every previous race where 2+ runners from today's race ran together, you can create a set of head to head 1v1 battles and sum up how many times each horse won. This is a good metric to know, but it has a couple of weaknesses. It gives you no information on unexposed horses - often in majors, some horses have never run against each other before because the population mostly runs in its own archetype (LDS v LDF). Secondly it doesn't differentiate between which horses a horse raced against and beat. It could be 100% against the poorer horses in a field, inflating the head-to-head metric. It is also a pain in the backside to calculate and with thousands and thousands of historic races to review, this metric is computational expensive. A shortcut to this metric is just checking to see if your preferred horses have previously ran against each other at today's conditions and who won and by how much.
Your rating systems
If you follow me on Twitter/X you will be familiar with Trueskill ratings. Trueskill is a pvp rating system designed by Microsoft for xbox online gaming. It's a single metric designed to capture the underlying skill of a horse, with luck removed. It's a Bayesian system, meaning that we start each race with rating for each horse, and based on the results we update those ratings and carry them forward to the next race. Each race is a series of 1v1 battles between the horses, If you beat the majority of opponents your score goes up. If those horses had high ratings, your score goes up more. Running your horse off preference or at the wrong distance and losing will negatively affect your score. It's biased against horses that take a while to lock in on their ideal distance. Overall, as a single metric its still pretty good to give you one more piece of information,but I wouldn't rely on it solely to make bets. Sometimes its great, other times it misses the obvious. As a single metric, global model. that is bound to happen, but people love seeing their horses in top 40 lists, so I will keep spamming them for a while.
Many gamblers have their own rating systems. I don't bet on something just because it's top rated on Trueskill, but I don't like calling out horses that are top rated that I have already bet on in fear of reducing the odds.
Handicap races yourself
Ultimately if you want to predict horse races and make money longer term, you need to figure out your own prices. Traditionally just building a model to predict which horse will win a race is tough to make money out of because the bookmaker has a price advantage over you. Some very skilled data scientists proudly build betting prediction models, only to discover that the better their model becomes, the more it converges with the prices of a bookmaker, albeit at a price disadvantage because the bookmaker has to make a margin on the prices they make available.
There are ways around this. In the world of deep learning you can develop your own custom loss function. Instead of optimising your model to predict a binary outcome (win/lose), you can optimise the model for any real world outcome. You could set your loss function to be the negative of profit, so, as the loss is minimised, profit is maximised. This reduces your win rate, so you have to endure more draw downs, but it can improve your expected value. You could use the Sharpe ratio in your custom loss function to adjust your returns for risk. Sometimes this makes your model prefer short odds favourites over every other type of bet as it seeks to minimise risk in your returns. There is no free lunch in machine learning.
Predicting is one part of the problem, the other part is the trade execution strategy. Do you bet when the model score is above a certain threshold? Do you just bet the top rated horse in every race? Do you bet on the top 3 horses in every race? Just as predictions can be optimised, so can the execution strategy. Generally the best strategy is to bet when the available odds are far longer than your predicted odds. The maximum EV bet is to follow this strategy. As mentioned earlier, there are no real longshots in PFL due to the way pools are juiced, so finding the best priced runner in your top three is a good plan, particularly when your fellow bettors have piled onto one horse in particular, making their odds very unattractive.
I am sure some players have built models like this, using machine learning to make predictions and price horses accurately. For a gambler, it's not just about picking the winner, but finding the biggest +EV moves, where your prediction of the true odds are out of sync with the marketplace. I am sure gamblers would want more betting opportunities, more players betting, maybe more exotic bets (tricasts or full-card predictions) and more juiced pots - and I am sure that will come over time as this part of the economy matures.
Finish times
Perhaps not as strong an indicator as you would like, it is worth paying attention to a horse's finish times over today's distance and conditions. This can be predictive of performance, but it is not that great in isolation. There is a fair amount of variance in a horse's finish time. Many horses have record amazing, even record breaking times, only to never follow that time up again. Others are inconsistent in the times that they achieve, particularly across distances and under-hoof conditions.
It goes without saying, but racing horses at the optimal distance is a major differentiator. If you learn nothing else about a horse, ask yourself: is this horse running at its optimal distance today? Horses running at their optimal distance are 2-3x more likely to win. A quick glance at their charts will tell you if this is true or not. Some horses can comfortably run at multiple distances, some only have one distance where they are competitive. Pay close attention to 4F and 12F races where specialists thrive.
Preference stars
I have covered preferences in this blog multiple times, so at the risk of repeating myself more than usual, preference stars matter. Low-preference horses rarely win. High-preference horses are still largely unprofitable to bet on, so you need to use this information to filter out the no-hopers, not to pick the winners. Also remember, it's the relative amount of stars you have not the absolute.
Benchmark
The benchmark for horses is very visible and is in fact a good predictor of winning races. At morning line prices, betting on the top 2 benchmark horses in a race is profitable (you can't get these odds though when it comes to betting remember). But if you see a benchmark leader at high odds, and its running on preference, you think its running at its preferred distance, it has a good head to head record with today's rivals, and has a good ML price, it might be worth a shot.
Play the long game
Okay, so that is 10 tips to get your started on your betting journey in PFL. There are 50 things I haven't mentioned because this post is getting too long. Things like, if you are going to bet big, bet late so you are aware of the odds. Or ignore post and gender, they have no influence over the outcome. But we haven't got time or space for those. You are going to have to figure other these things for yourself. Be wary of people selling you tips or systems. Make sure there is verifiable proof of their success. The game seems very sound for the robustness of the algorithm. I see the randomness and I see the underlying skill of the horses; I think it has the opportunity to scale - it just needs the liquidity and active players in each prize fund. There are full sure things in this game, so be prepared for the ups and downs. Today many of the featured races are the majors - and these are some of the hardest races to predict. Big, competitive fields, make the prediction of winners as hard as it gets - so don't get disheartened if you don't win the Kentucky Derby with your first wager.
And none of this was gambling advice. Right? Right.
Join the fun and put these insights into practice at PhotoFinish.Live and if you are considering starting your own stable please consider using my referral code: PADDOCK or just click on this link: https://signup.photofinish.live/?referralCode=PADDOCK
Please remember this is a web3 game where you spend your own money. Nothing I write about should be considered financial or investment advice.
Other blog posts:
Breeding in season 21: https://aipaddock.com/getting-ready-for-breeding-in-season-21
Saving the LTF archetype: https://aipaddock.com/there-is-gold-in-them-thar-hills-a-campaign-to-save-ltf
Season 19 retrospective: https://aipaddock.com/season-19-by-the-numbers
A race selection process flow: https://aipaddock.com/selecting-the-right-races-for-your-horses
How good is your new juvenile: https://aipaddock.com/how-good-is-my-new-juvenile-horse
Tips for breeding strategies: https://aipaddock.com/charting-the-uncharted-in-breeding-within-pfl
Know your odds of breeding success: https://aipaddock.com/know-your-odds-before-you-roll-the-dice
Further down the rabbit hole of ML: https://aipaddock.com/deeper-down-the-morning-line-rabbit-hole-we-go
What can ML prices tell us: https://aipaddock.com/skill-luck-and-morning-lines
Finding the optimal distance: https://aipaddock.com/understanding-optimal-distances
Understanding breeding: https://aipaddock.com/understanding-the-most-successful-breeding-horses-in-pfl
What are subgrades: https://aipaddock.com/what-are-subgrades-and-how-do-they-work
Fastest horses: https://aipaddock.com/who-are-the-fastest-horses-in-the-game
Racing profitability: https://aipaddock.com/how-much-profit-do-you-make-racing-horses
Trueskill pvp ratings: https://aipaddock.com/which-horse-is-the-goat-in-pfl
Breeding: https://aipaddock.com/trying-to-understand-breeding-in-pfl
Evaluating horses: https://aipaddock.com/how-good-is-my-horse
Understanding performance: https://aipaddock.com/understanding-the-true-performance-of-your-horse
Do stars matter: https://aipaddock.com/how-much-do-preferences-matter
FF Rating vs Finish Time: https://aipaddock.com/the-difference-between-ff-rating-and-finishing-time
Are horses getting faster: https://aipaddock.com/are-pfl-horses-getting-faster