Mayweather McGregor
A Lesson In Forecasting
Well the fight of the century has ended. The experts were right, luckily. The betting markets were wrong, they are making a new habit of that lately. And here I am to tell you that I scripted the flow of the game… which I don’t expect you to believe. I have no proof. But I suspect I can teach you something regardless.
For starters, don’t believe that too many other people got this right either. This fight had a strong chance of going either way. If that upper cut in the early rounds had been a little harder, if Mayweather had been a bit slower to counter the combo… I never write this article and the entire story of who won and who lost changes. That wasn’t the only pivot point, there were others.
Most experts simply discounted the ability of an MMA fighter to win against a seasoned champion (90% or more). The betting masses saw it otherwise (95% or more). Most members of each group modeled the most logical thinking — McGregor would win early or Mayweather would win late. Assuming a dozen other crazier alternatives didn’t occur… one punch knock-out, McGregor being disqualified, or some other far-fetched possibility. But what does such a bifurcated set of predictions teach us? And how would that change if one of those crazy alternatives had occurred?
The latter were long-tail events. If you utilized a Monte Carlo style simulation to predict this fight, which had I been tasked with a prediction — I certainly would have, these sorts of outcomes would potentially have been modeled in. And somewhere in the betting and analysis space, there were some loan souls who made such predictions. They were low probability.
What would we have learned if Mayweather’s first punch had dropped McGregor like a rock? Plenty about the boxer’s preparation, but little about anyone’s ability to forecast. Some few ‘know-it-alls’ suffering from confirmation bias and inflated egos would have had 15 minutes of fame (sadly a bit more), but predictions of long-tail events deliver little more than awareness that these things occur. A good lesson but a thin one.
The model for a boxing match is fairly simple. The amount of money and power interests behind this fight actually made it even easier to predict. To make my pre-fight prognostication, I simply walked through the fight, round-by-round, determining the weighting I saw from various points of feedback. I built a script with weights and by derivative — probabilities.
My prediction of a late round TKO, could just have easily been wrong. If it had, this article may never have been written. But not because of the outcome, what would have jeopardized this article wasn’t who won, but when. A shorter fight would have allowed me little opportunity for comparison. The value of a feedback scripted approach is learning — and a shortened fight would limit that learning.
Let’s Get Ready To Model!
The weakest point in a feedback model like this is the first round. It is filled with the most external feedback, the least residual feedback (memory), and is therefore most susceptible to complexity (wrongly called dumb luck). A lot could have happened in round one. Here is an example of some of the balancing minus the numbers.
- McGregor had never taken a punch from a champion +MW
- Mayweather would be tempted to test that +MW
- BUT Mayweather would also be inclined to slow the tempo +MG
- Mayweather’s handlers (professional boxing in general) had no interest in an outcome like this +MG
In feedback analysis, this final point is external feedback and critical. The money and importance of this game meant that members of boxing, the MMA, and likely even Vegas were in the ears of trainers and managers (if not the contestants) on both sides. I am not saying the fight was rigged. It didn’t need to be — in the earliest rounds, external feedback was so strong — it couldn’t help but dominate the script.
In the later half of the first round, momentum starts. Avoiding any initial KO, McGregor would begin building confidence. He was the younger fighter, the less experienced fighter, and a boxing novice. Now context is working for us. It would have been harder to script a fight that didn’t include an MMA fighter. Could McGregor have come out fast, avoided the one-punch KO, and delivered a surprise of his own? Certainly!
The benefit of a scripted model is that had that happened, our point of failure becomes clear. The arguments (or weighted feedback) against that was strong however.
- External forces did not want a first round KO (now) +MW
- Mayweather would have been coach to expect and evade +MW
- McGregor would have been cautioned about overextending before “feeling out” his opponent + MW
As the fight progresses, momentum feeds McGregor. External feedback (although still favoring a longer bout) has much less influence once the contest is underway. And so, as round two through four develop, our model (like our betting market) produces the highest probabilities for McGregor to win. And he came close…
Rounds four and five gave us another point of weakness. Two new points of feedback enter the model.
- McGregor is now fighting a fight much longer than he has in the past +MW
- Each corner is making in-fight adjustments. +???
Here the novelty of this fight is a problem. It is really hard to understand what to expect. Although that likely made a KO (or TKO) very unlikely in these middle rounds. With each contestant experience so much change (pace being predicted, Mayweather’s Mexican fighting style less so), feedback encourages caution. My model wasn’t sure who to favor, but it clearly showed the least likelihood of any decision.
As the fight wore on, all the momentum and feedback swung toward Mayweather. He had predicted a long contest and he had fought plenty. It also though support a TKO, rather than a KO. All the feedback now point toward sloppy and although opportunities abounded for a lucky punch or a lenient referee — a robust model would have leaned more toward the TKO.
The possibility that Mayweather would rely on the judges rather than stay aggressive was high on the first cut of the model, but feedback is layered. Those early McGregor leaning rounds created a situation where Mayweather would had every bit of feedback he needed not to risk it. Again, the fight went to script.
There is no need for you to trust the outcome.
Maybe I am lying. Maybe I just have confirmation bias, too. It doesn’t matter. The feedback model I summarized above describes a process rich with factors and weightings. It is a multitude of predictions. Had the script been wrong at any point there was plenty to learn. And, as noted, the fight lasting 10 rounds actually provided the largest window of learning.
So what did I learn? McGregor’s camp made adjustments much earlier than my model expected — that was a failure on my part to realize they were an MMA corner (mostly) and relying on their feedback of the need for an early win. The infamous uppercut was in round one, my model would have favored it in two or three. I didn’t think through how McGregor would react to the series of hits he took in round 10. Like an MMA fighter, he stayed on his feet but dropped his guard (presumably muscle memory of an impending take-down). That could have changed a lot.
There is actually much more to share, but predicting fights is not a high reward opportunity for a model like this. This style model is far better suited for things that will recur — like sales processes, projects, and other more mundane endeavors (no one is putting them on pay-per-view). These opportunities allow you to test the model, garner feedback (meta-feedback if you will), and continue to learn and optimize.
I hope this inspired you. Thanks for reading! And at the very least, try a little ‘Notorious’ whiskey.
Note — the author has not been compensated for this endorsement, nor tried the product. Any effort to remedy that would be greatly appreciated.
