MMA ELO IS THE GOAT DISCUSSION

Continuing the discussion from Jose Aldo has more of a claim to GOAT than Khabib:

Was having a chat with @MMApurist about a buddy of mine who goes by the moniker MMA ELO on YouTube.

As his bio says, he’s a data analyst by trade and an MMA aficionado in his off time. His ELO Rating model, for as long as I’ve been aware of it, has always intrigued me as a system that I always knew I wanted to build, but never knew I actually wanted to.

By that I mean I always was in my own mostly failed search for an unbeatable betting system that would flawlessly predict outcomes of fights, and allow me to profit from it.

Mine was built around trying to memorize as many statistical outcomes and trends as I could and leveraging it to my advantage, his on the other hand, revolved allowing a high end computer to do all that for him in real time. Points to ELO.

And if I’m really being honest, this software/system/thing he’s created is something I don’t actually think I would have actually ever be able to build. Or even think of myself. And while ELO is the curator and creator of this sacred set of code, even he borrowed a thing or two in order to help build his ‘model.’

He took the bare and basic core of his rating/predicting system from the Elo Rating System used to rank players in chess which basically uses the results of previous chess matches between chess players to predict the outcome of future ones. Or, at least that’s how I think it works.

Plainly put, in chess, a player with a higher ELO rating is statistically more likely to win against a player with a lower ELO rating. The higher the discrepancy, the greater the probability of winning for the higher tiered player.

The same basic concepts of that system in chess can be used to predict the outcome of a fight in mma.

The thing with applying this to mma, and with applying it to betting in particular, is that he doesn’t simply decide who will ultimately win a fight, he measures the likelihood of any and every outcome of a fight happening against the implied odds on various sportsbooks around the web. And attacks where he sees weakness.

So he isn’t simply betting on the fighter with the higher ELO rating every time, he is betting on a line where a bookie fucked up and gave an implied win probability beneath what his own internal model predicted.

So if his model predicts a 33% (+200) chance of a fighter winning inside the distance, but there’s a line out there that gives a 22.2% chance (+350), then he inputs that bet. I think he did something similar to that with Clay Guida this past weekend.

It seems pretty simple, but it just gets more and more complicated and confusing the more questions I ask him about it.

He bets on more than UFC and track all his winnings (and his losses) on his Instagram page. He’s on something of a downswing atm relative to where he was, the same happens in my world of poker (it’s almost expected), but he’s still massively up from his initial $300 investment and quadrupled his bankroll this past weekend if I’m remembering correctly. Little tip: He’s basically all in on Charles Oliveira this weekend.

Fair warning that I could be wrong about literally all of this.

beard GIF

2 Likes

Here’s the basic genesis of the post. ELO is the GOAT. Look into him.

Wrapped up like a douche.

1 Like

Me or him?

Context.

My reply to above.^

Again, I could be wrong about all of this.

@MMApurist

I’m not 100%, but I’m pretty certain his model understands that there are more KO’s at 265 and that Ngannou, for example, is more likely to get a KO in a fight. I guess a better way to say it is that I’m pretty certain that it creates it’s own measured probability of say Ngannou knocking out Cyril Gane using data behind every fighter the two of them have faced, how they fared against those fighters and even how those fighters Cyril and Francis both fought fared against other fighters themselves.

That’s my basic understanding of it, it’s like a giant game of M&M Duels except instead of M&M’s, you get CTE.

But yeah, like I’m telling you, I barely even understand how this code works, and I’m not even entirely sure he does either. That’s why I almost consider it this sacred and mysterious code that seems like it has a mind of it’s own. Probably because it does.

Those are some interesting questions/remarks. I’m really not sure what he does or if he has anything to control or measure for that. One thing I do remember critiquing him for in the past (not sure if critiquing is even the right word), is that his model didn’t seem to offer any measurement for damage incurred over the course of a fighter’s career. Basically, I don’t think that it has a chin-o-meter, or something that will measure the depletion of a fighter’s chin over time and as he or she takes more and more shots to the head.

I’ve been mulling over thinking of something that would help him create some sort of tool for that, but I’m probably way too incompetent to come across anything that may help him… Frankly, I barely understand how his code works, so implementing something will help him would be asking a lot.

If you have any suggestions, I’d definitely pass them off to him.

No problem with the rambling. It’s why I made this thread, so we can ramble all we want.

But to answer you back, there are certainly more knockouts in the heavyweight division and it’s definitely something to include and incorporate into a scenario when you’re looking at two HWs.

I forget the exact number, but something like half of all HW fights end in KO’s.

Edit: It’s just tricky because you’re always effectively measuring the implied odds from the bookies against what the actual likelihood of one actually happening is and if you aren’t MMA ELO, I think that becomes a lot more difficult to measure.

I know he makes (and wins) plenty of bets on heavyweight fights to go to decision for example and it’s simply that he made that bet instead of one on a KO because his model ultimately saw a decision happening as more likely than the line with the bookie, so he made the bet. Granted he’s lost plenty of those too, but again, it’s a pretty complicated (but also logical) system. It’s pretty similar to poker in a lot of regards. Entirely like something else on most the others.

I made a program to calculate Elo ratings for fighters about four years ago, and made a thread about it.

I wrote another program to test it against the betting odds, which was easy because tapology lists the odds for every fight. Unfortunately, it didn’t beat the odds in the long run, so I stopped using it to bet.

The program on github doesn’t work anymore because tapology changed the format of their database.

1 Like

How in-depth were the numbers behind the aggregation of the final ELO ratings? I noticed your ranking system is swayed towards having much higher ELO ratings on average than my pal MMA ELO’s. For example, your all time greats had ELO ratings as high 2500, but I don’t think anyone has broken 2200 in my buddy’s. What do you think might cause that?

1 Like

You can choose the K-factor to make the numbers change to a greater or lesser degree from wins or losses, and you can also choose what the initial rating is. It’s possible that I chose it so that a very small group of fighters would reach the 2500 mark so that I could make a thread about “Grandmasters of MMA.”

If I were to properly calibrate to give the most accurate ratings, I would probably choose the numbers that minimize losses versus the betting odds. That might be what your buddy did. I don’t remember what the best numbers were.

1 Like

Fightmatrix uses an Elo-based rating system, and I think they set K to 170.

1 Like

Funnily enough, MMA ELO is the reason why. :smirk:

Why is that? I was thinking of rewriting the program and using mixedmartialarts.com database. It’s pretty conveniently formatted for data mining.

Sorry for the hold up, I actually just got off the phone with him now.

He’s told me this one before, but I linked him this thread in case he wanted to comment and he mentioned your comment about your script not working on tapology anymore.

He said it was probably because of him.

The story goes that Tapology actually banned his home and phone IP addresses after he got caught scraping their data.

And he actually had the same thing happen to him afterwards with his script where he used it to scrape some data off the site one day, went back to scrape some more about a month later and found that his script wasn’t working.

He found exactly what you did: That they somehow or another reworked the system they used to log and catalog data from fights to prevent him from scraping anything.

It didn’t work for long. *By that, I mean he found a workaround.

1 Like

Btw, I didn’t know mixedmartialarts.com had its own database. How do I access it, I’d definitely like to poke around.

That’s pretty weird. But fortunately the database here has gotten a lot better over the years.

https://www.mixedmartialarts.com/fighter/

Do you know why he is betting in fixed unit sizes, rather than using the Kelly criterion?

Sent him the relevant info last night.

He said thanks and that he’d look into it.

1 Like