# Nerdy Boring Statistical Analysis: How to Profitably Bet on Lewis vs. Hunt

Discussion in 'UFC Discussion' started by latoya johnson, Jun 8, 2017.

1. ### latoya johnsonDouble Black CardYellow Card

Joined:
Mar 27, 2016
Messages:
1,274
3,316
Location:
Hio de Janeiro
This is part of a series of threads.

1. Introduction
Several days ago, I defined a statistical model to estimate win probability for MMA bouts.
I have since made several critical refinements.
Credit to @razmataz1314 for questioning the temporal linearity of athletic decline, leading to an important change in the model.

The science behind this model has grown increasingly esoteric and diverse, pertaining to integral calculus, continuous interest rates (believe it or not), and combinatorial statistics.
If anyone wants to learn more about its derivation, they may post a request on my profile page and @ me.
(BTW, @wookie brawl , I hope you got what you needed from me. Of course it's somewhat obsolete now).

However, since this model is likely profitable, it behooves me now not to release the software or the most intricate details of its derivation.

With recent improvements and a data sample size of 52 bouts, it may now be accurate enough to be profitable for legal betting purposes.

This probability model (now expanded to include financial statistics) is the basis for the betting guides which I will post.

Statisticians have a similar saying: Reduce your confidence intervals by increasing your sample size.

This means that if you only place very few bets based on this model's recommendation, you're not guaranteeing a profit. There is always risk, but it can be reduced by placing multiple small bets.

The table below shows you the relation between risk and sample size.

As you can see, there is a very high risk of losing money if you only place, say, 2 bets.
You may mitigate the risk by placing 9, though.
In fact, your risk is almost always lower for an odd number of bets than for an even number.
There are three reasons.
1. Combinatorial Statistics
2. Unfair Betting Lines
3. The Model's Accuracy
Basically, the model has a 73% success rate at picking the winning fighter.
However, if you place an even number of bets, then for half of the possible outcomes, you'll lose half of your bets.
Since the oddsmakers' odds are unfairly skewed against you (your returns are generally less than 100% of what you bet), you'll end up losing money in all of these cases.

3. Two Betting Styles
My model includes two betting approaches:

3a. High Risk (not recommended)
The high-risk method has a historical 44% profit margin.
No, I did not make that up.

However, these high returns come with grave risk.
Remember how risk is reduced by placing multiple bets? Well this method disregards that wisdom to some degree.

Bet spreads are heavily weighted towards particular fights based on their individual expected returns.
For example, my model's highest win was a colossal bet on David Teymur to beat Lando Vanatta.

Why did the model bet so much on this fight?

The high-risk betting method prefers weighting bets where it perceives a high expected return, and this happens when there's a great disparity between the oddsmakers' expectations and the probability model's.
Basically, the oddsmakers' were convinced Lando would win that fight, and my model was convinced that David would win. This gave it a +300 line on what it perceived to be a sure-fire win.
The high-risk betting model salivated at this opportunity.

David did win, resulting in massive gains.

However, its greatest loss came when the oddsmakers' were sure that Pedro Munhoz would defeat Damian Stasiak. My model ardently disagreed, and tried to exploit the tantalizing +425 on Damian.

Unless you're either
1. able and willing to lose a lot of money or
2. able and willing to place many dozens of bets to mitigate the risk,
don't use it. It's simply too dangerous.

3b. Low Risk (recommended)
The low-risk method has a historical 35% profit margin.
Obviously this is still very high, so I recommend this approach over the high-risk approach.

The low-risk method is characterized simply by
I. distributing your wager evenly across all bets, and
II. always betting on the fighter that the model predicts will win regardless of betting lines.

Again, refer to this chart for risk evaluation.

Again, place an odd number of bets to minimize risk.
Note that this chart absolutely does not apply to the high-risk method.

4. Gambling Guide for... UFC Fight Night: Lewis vs. Hunt
I was only able to make recommendations for 9 of the bouts on this fight card.
Disperse your wagers according to the instructions below.

High Risk (not recommended)

[TLDR]
Low Risk (recommended)

Last edited: Jun 8, 2017
2. ### latoya johnsonDouble Black CardYellow Card

Joined:
Mar 27, 2016
Messages:
1,274
3,316
Location:
Hio de Janeiro
Closing Considerations

1. I am not encouraging anyone to gamble.

2. The bet spreads recommended here are based on current betting lines. If the lines shift before you place your bets, then these recommendations may become obsolete.

3. I may post the results of both betting spreads after fight night for review.
However, I would still prefer a larger sample size before any judgments are made so no one gets mad at me.
Be aware that the high-risk method may be unprofitable even for very large sample sizes... or not. No one knows, 'cause it's high-risk!

4. Just for fun!

Last edited: Jun 8, 2017
3. ### loyalyoloyalGuest

Just tell me who to bet my house/ life savings (\$30) on.

mon, Filthymaggot, myjohnson9 and 6 others like this.
4. ### The Witcherвідродитися

Joined:
Nov 15, 2015
Messages:
9,978
25,649
Location:
Australia
Mark hunt being the underdog is too juicy to pass up

5. ### WormwoodYellow CardYellow Card

Joined:
May 15, 2008
Messages:
2,199
4,756
Location:
Can I borrow \$2.47?

6. ### davidlemonpartyBrown Belt

Joined:
Jan 9, 2017
Messages:
4,412
9,407
Location:
Perth, Australia
Hey man that's a lot of effort you've gone to, hope it works out for you ☺️
Just for fun I'm going to put a \$10 parlay covering every matchup, on all the fighters based on your win probabilities, excepting the 50/50 match up. See what happens!

7. ### CrotonPurple Belt

Joined:
Oct 25, 2009
Messages:
1,596
4,016
Ottow @ 7%
Tim Elliot @ 31%
Pearson @ 10%

Wat

8. ### Rondas PillowBlack Belt

Joined:
Dec 17, 2015
Messages:
5,188
13,749
Did someone just finish their first statistics class?

jrleclerc88 and ArtOfDrowning like this.
9. ### wookie brawlBrown Belt

Joined:
Sep 10, 2013
Messages:
3,570
5,040
Location:
Australia
Just checked my e-mail-I got it! Thanks very much!

10. ### gunlokBlack Belt

Joined:
Feb 26, 2016
Messages:
6,013
6,566
Just tell me where to shove my money

11. ### Mr hankyPurple Belt

Joined:
Jan 11, 2017
Messages:
1,839
5,076
Chan-Mi has 90% of chance of winning? Based on what? Her win over inexperienced cans and last one against 45 years old fighter who didn't win since 2010 ?

Pichel after 3 years layoff has 84%?

Ottow vs Kunimoto has 7%?

LMAO, man except Brunson and Lewis you've got your predictions totally wrong...

Last edited: Jun 9, 2017
12. ### YellowBanenooRed Belt

Joined:
May 28, 2017
Messages:
9,245
27,824

arridude and ArtOfDrowning like this.

Joined:
Sep 8, 2011
Messages:
11,008
14,288
14. ### Ramon AntonioBrown Belt

Joined:
Feb 19, 2010
Messages:
4,911
9,212
52 bouts is not nearly enough data to create a good model. Also, how do you know you're not overfitting? Did you use a different train / test set?

The Clark likes this.
15. ### warriorscommandoI will always protect you

Joined:
Aug 27, 2011
Messages:
16,036
24,121
Crock of shit.

16. ### warriorscommandoI will always protect you

Joined:
Aug 27, 2011
Messages:
16,036
24,121
Oh and 52 bouts. Took a lot more games to do moneyball analysis in a sport with fewer variables (and variables that could be quantified easily).

TS is autistic but not a savant.

Spouting a load of shit (continuous interest rates? What the motherfuck motherfuck? Calculus - you? My ass, unless you're running a histogram). His number crunching is like a kid tossing straws in the air.

Also you idiots talking about parlays - want to calculate your odds of winning based on his bullshit odds?

Calculate the result of ,63x,92x,90x,87x... then multiply by x100 and you'll get a very small figure for your %

Gabe and Threetrees like this.
17. ### latoya johnsonDouble Black CardYellow Card

Joined:
Mar 27, 2016
Messages:
1,274
3,316
Location:
Hio de Janeiro
Wow, you kind of roasted me. I'd like to respond.

1. I agree, a better sample size would be ideal. It just takes so long to compile. I still think the sample was large enough for this model to be profitable. @Ramon Antonio

2. Yes, interest rates. I won't say how they're related specifically, but the idea is that certain advantages depreciate with time. Depreciation is measured very well by interest rates. As for calculus ~ it's a secret.

3. At the end, there, you are actually calculating the odds that a person wins every bet that they place. I neither promised nor now predict that anyone would win all of their bets.

18. ### latoya johnsonDouble Black CardYellow Card

Joined:
Mar 27, 2016
Messages:
1,274
3,316
Location:
Hio de Janeiro
UPDATE:

I have further refined the algorithm.
New spread with just one change: Bet on Volkanovski, not Hirota.

Also, I have drastically improved the accuracy of the prediction confidence by inputting more variables.

Last edited: Jun 10, 2017
19. ### GabeSaturdays are for the Boys

Joined:
Feb 8, 2015
Messages:
11,074
31,693
Location:
Charlotte, USA
Hunt is overrated. People usually always pick him to win and half the time he loses badly.

Joined:
Nov 5, 2004
Messages:
10,740