The Alchemist: Creating MMA Point Projections - Silva vs Mir

The alchemist is not a scientist, nor a mathematician. He is an artist, searching for truth and order where others only find chaos.

The Alchemist – Diary Entry 23/02/2015

I am thoroughly enjoying playing MMA daily fantasy. As an MMA fan and an avid DFS player, the two make a perfect combination. The events are almost weekly, the small pool of fighters makes research relatively straightforward, the action is 12 months a year, and like any sport, having something at stake makes can make watching UFC events all the more enjoyable.

Wagering on combat sports is a tradition as old as civilization itself. From the moment early Paleolithic humans competed to see who could pin the other down, those watching competed to pick the winner. The sport of Pankration, one of the earliest known precursors to mixed martial arts, was introduced in the ancient Greek Olympic Games in 648 B.C. In Greek Mythology, Theseus, mythical founder-king of Athens, is said to have submitted the powerful Minotaur using his advanced pankration skills. The sharp spectators who picked Theseus that day likely made a small fortune.

And so on through the gladiatorial games of ancient Rome, the jousting contests of medieval Europe, and the boxing rings of the early 20th century. The ancient tradition of wagering on combat sports lives on today; a fascinating Muay Thai kickboxing gambling culture thrives in Thailand, and the prominence of sportsbooks in Great Britain ensures classic betting on combat sports will continue to exist, so long as the sports they serve retain their place.

For most people however, this timeless art of wagering on combatants has been lost through boxing’s decline in mainstream culture. Yet it remains a wholly unique spectacle; two people, standing toe to toe, fighting for you to make a profit or enjoy bragging rights. My hope is MMA fans and DFS players everywhere will soon experience the thrill of watching their underdog fighter land a first round knockout, and the player base will continue to grow. Based on the UFC posting odds prior to the start of a fight, it seems they have a similar theory.

And so, it is in this modern age that I find myself asking two questions. Is it possible to create an accurate, useful MMA point projection system? Can I build it?

The problem is obvious; not only is MMA one of the more unpredictable sports to begin with (we just witnessed an event where underdogs won nine consecutive fights), but the sample size is absolutely tiny. A basketball player can play 70, 80, or 100 games in a single year, but most fighters will only fight a handful of times. Can you really extrapolate any useful data from such tiny sample? Is it even possible to get a representative sample of fights that can provide the basis for predicting say, how many strikes a fighter will land in his next bout? As far as I’m concerned, there’s only way to find out.

This blog will catalog that journey. Over the past three weeks I have been building and tweaking such a system, and will now put it to the test. Through a combination of bookmaker odds, fightmetric data, fighter profiles, and a touch of intuition and experience, I produce point projections for every fighter according to DraftKings’ scoring system, and build cash lineups with that data. Without further delay:

Point Projections and Results

What a night to begin testing! Of the entire fighter pool, underdogs were perfect on the night! Absolutely incredible. Time after time the Brazilian crowd was left stunned as their hometown fighters disappointed. As you can plainly see, in matchups that went the distance, predictions for total points scored in the fight were relatively accurate. The sheer number of first round finishes meant the point totals of several fights were underestimated.

Why total points scored? The system is such that when a fighter wins his fight, he will outperform his projection, and vice versa. Thus, the total points scored in the fight is the best metric to determine the accuracy of the projections. It may prove to be interesting to take the current projection system and also include projected point totals for scenarios where a fighter wins his fight and loses his fight. I will keep this in mind. Of course, all this information is only as useful as the lineups constructed from it.

Optimal Lineup and Outcomes

The entire goal of the projection system is to identify the best value plays of the night, and combine those into the most cost effective lineup possible. The fighters of note here were William Patolino, Marion Reneau, and Michael Johnson:

1) Patolino, as by far the heaviest favourite to finish his fight, was the most obvious play of the night. Suffice to say this did not exactly go as planned as he was knocked out in the first round by Matt Dwyer.

2) Reneau was my favourite underdog pick, and the most valuable underdog pick on the board. All the metrics predicted a fast-paced, high action fight that would go the distance. Typically, this is the kind of performance you seek from your underdogs: lots of strikes thrown, and a chance to squeeze out a decision. The fight began as predicted, but Reneau quickly made Andrade, who had made disparaging comments about Reneau’s brazilian jiu-jitsu belt, eat her words with an excellent first round triangle choke.

3) With the cancellation of the Oliveira- Waldburger bout, Michael Johnson was the only fighter available in the middling price range of $9k-$10.5k, and consequently was a near must play in order to maximize the value of the four other slots. The field concurred, and he was the highest owned fighter on the night at almost 60% ownership.

Mike de la Torre filled out the lineup as the most valuable underdog play remaining. I made the decision to substitute Iuri Alcantara for Antonio Silva, and kept the rest of the optimized lineup. I thought Silva, with his knockout potential and five round bout, had a comparable floor to Alcantara but a higher ceiling. This felt like a great decision for most of the night… until Bigfoot came out and seemingly just stood there waiting to get knocked out.

Nevertheless, my projection system and lineup optimizer combined for 286.5 points, cashing all 50:50s, triple ups, GPPs, and beating all the players who had the misfortune of playing against it head to head. My primary lineup scored a respectable 264.5, good enough to be slightly profitable on the night.

Analysis and Conclusion

The results are certainly encouraging, but hardly conclusive. As nine underdogs won nine consecutive fights, I do not expect to find anything conclusive digging through the rubble. There can only be so much truth and order on a night where every single professional prediction was flat out wrong. Thus, the questions remain. Was this a fluke? Was it a coincidence that underdogs selected by the optimizer scored two first round knockouts? Or was that the reward for identifying the best values? Is there any predictability here whatsoever? Am I just staring at white noise, finding patterns that don’t exist? Only time will tell…

I hope you enjoyed the read and are interested to see if we can find answers to these questions. Future blogs will examine various aspects of the system in more detail, discuss DFS and MMA in broader terms, and keep track of the only metric that ultimately matters: Is all this work resulting in profitable lineups?

Follow me on twitter @huitcinqDFS

About the Author

Comments

  • gravycakes

    Great stuff! Interesting read.

    I like your idea about looking at floor/ceiling projections on when a guy wins his fight and floor/ceiling porjections on when he loses his fight.

  • huitcinq

    Yeah the point projections as they are are useful for making cash lineups (assuming, you know, they make any sense in the first place), but since my goal is also to measure their accuracy I think that would be a novel way to do it.

    You raise an interesting point though. Since win/loss projections would essentially be floor/ceiling projections, it could prove useful for building GPP lineups.

  • lilscrapnike

    huitcinq, great read, and I’ see a few things in your chart that I may consider adding to my projection system (such as more precise time of fights). My biggest question is how to project guys with limited data from the UFC (i.e Matt Dwyer who was KO’d in the first minute of his only fight and threw no SS or had no TD), or a fighters making their UFC debut’s. That is probably my biggest gripe with fightmetric is that they only show data for fights from Strikeforce/WEC/UFC.

  • huitcinq

    Yup that’s where I get caught up as well, even doubting the validity of a projection system in the first place. DK seems to use the fight odds to determine pricing, so there is only so much value to be found by comparing odds to salaries.

    As for FightMetric, not only is it a subjective system in the first place with the evaluation of significant strikes, but the sample size is absolutely tiny most of the time, particularly for newcomers as you pointed out. Since newcomers are so often in play, a projection system that ignores them would be near useless.

    That’s where projecting becomes more and more subjective… can I act as a handicapper and fill in the blanks? Will my results be any good? It’s an interesting problem and it’s why I started the blog. Glad you enjoyed it.

  • AvgAmericanM

    I think MMA is the one (maybe golf as well, idk?) sport where analytics will rarely fair better than good capping. Styles make fights and numbers don’t really see that. So many intangibles in fighting. One off occurrences, flash knockouts, subs, camp changes, notice, gameplan adjustments, etc. skew the numbers significantly… especially when as you mentioned you consider small sample sizes and quality of opponents.

    I think there’s a tiny bit of value to be gleaned from analytics (work rate, tendencies), but you really have to do your homework/tape study and add context to the numbers to gain any advantage with them… and in the end like Rogan says, it’s still two dudes throwing bones at each other. Anything can happen.

  • Jasen777

    I think the sample size issue is going to be one that always limits the usefulness of a statistical approach to MMA. I wonder if having a “replacement level” fighter (by weight level) would be of any use – making a stat composite from UFC debuting (and/or limit experience). A problem is of course they often fight each other…

    Looks like you gave Johnson too many points in your “huitcinq’s lineup.”

    DK most certainly does look at the betting lines to set prices, but not always perhaps as good as they should. Rousey and Holm may be near the same odds to win right now, but given the round over/unders (not all the props seem to be out now) having Holm so close to Rousey in price doesn’t make sense.

  • huitcinq

    @Jasen777 said...

    Looks like you gave Johnson too many points in your “huitcinq’s lineup.”

    Ah thanks for pointing that out. The total is correct but the points are lined up with the wrong fighter. I’ll fix it for next time.

  • huitcinq

    You guys both make great points, and it’s the reason why I want to publicize and follow this process through each event. I certainly built up my system knowing it might be a wild goose chase.

    The sample size is brutal for reasons you’ve both stated. Not only are fluke occurrences common, but most fighters nowadays develop specific game plans for their opponents. So even if you have a pool of 9-10 fights of FightMetric data, it’s very difficult to know if it will be representative of the next fight. This is where scouting and matchups come in.

    The replacement level fighter idea is interesting. I’ll have to think about that one a little more.

    From a DFS perspective the sport remains brand new, so I do think all ideas are still on the table. I’m not even sure if we can conclude with any certainty that a “sharp” MMA DFSer will be profitable in the long run, though I certainly hope that’s the case.

  • headChopper

    RG Contributor (OG Status)

    • 232

      RG Overall Ranking

    • Ranked #24

      RG Tiered Ranking

    • x2

      $1M Prize Winner

    • 2014 DraftKings FBWC Finalist

    Yeah man, good analysis. MMA is a fun sport and I will continue to play it. But I’ve played 3 or 4 slates thus far and they’ve all gone different. This slate had all dogs. Last slate was favorite heavy with 1 early stoppage. Before that was several early stoppages. Tough to get a good grasp on this. For me its a good GPP sport cause I wouldn’t want to risk too much cash game money when 1st rd KO’s mean so much, yet are so unpredictable. Then the GPP cap is limited due to the small number of fighters per card and the tight salaries. But fun sport none the less

  • huitcinq

    @headChopper said...

    For me its a good GPP sport cause I wouldn’t want to risk too much cash game money when 1st rd KO’s mean so much, yet are so unpredictable.

    Interesting… see my logic was that the limited pool of fighters makes it a difficult GPP sport, as any lineup near the cap will have duplicates (the first week I played, I believe 35 or so other people had my primary lineup in the takedown).

    Thus you have to treat GPPs as you would small slates in other sports, meaning your GPP lineups should be significantly under the cap to ensure you’ve differentiated from the field (but perhaps I’m overthinking this as the player field isn’t as large as in other sports). The difficulty here is you’re chasing the random, unexpected outcome of each card. I’d be interested in your thoughts on this.

  • xMiami2Ibiza

    interesting read mr huitcinq and some good feedback from everyone above. sadly for the last couple of months I have spent waaay too much time analyzing UFC from both a numbers standpoint and just a “visual/feel” standpoint if you will for the purposes of DFS. to be honest, i don’t know what the right answer is. as mentioned above, because of the huge reliance on the first round finish it really skews a lot of other things. throw in the limited data you are getting on some of the fighters through fightmetric and you have a real clusterf*ck on your hands.

    i am of the belief that styles make fights, but as you know anything can happen. having just said that though, i am a huge “fightnomics” fan and love the statistical side as well.

    also agree with the statement above that more than other sports i believe handicapping is more important in MMA DFS than other sports IMO. then again, you have a card like last weekend and all i can do is shake my head!

    i will be curious to see how your experiment pans out over the next few events and look forward to further reading.

  • X Unread Thread
  • X Thread with New Replies*
  • *Jumps to your first unread reply

New to DFS?

Be sure to click through our links and use our exclusive promo codes to receive the industry's best sign-up bonuses, including free access to our premium content.

RotoGrinders.com is the home of the daily fantasy sports community. Our content, rankings, member blogs, promotions and forum discussion all cater to the players that like to create a new fantasy team every day of the week.

If you or someone you know has a gambling problem, crisis counseling and referral services can be accessed by calling 1-800-GAMBLER (1-800-426-2537) (IL). Gambling problem? Call 1-800-GAMBLER (NJ/WV/PA/MI), 1-800-9-WITH-IT (IN), 1-800-522-4700 (CO), 1-800-BETS OFF (IA), 1-888-532-3500 (VA) or call/text TN REDLINE 1-800-889-9789 (TN).