Solving the Real Problem: When to Bet for Profit

Introduction:

WolfTickets.AI started out as a project to rid myself of imposter syndrome while working with brilliant coworkers and customers at AWS as a Solution Architect, and it led me all the way to being a Principal SA focusing on the launch and success of many of their AI and ML services.

To that end, it has been wildly successful. However, I took the project further as I looked to develop and release products full time at Protect AI, WTAI became a laboratory for how to build a successful product which engaged with AI and ML tech. While amazingly educational and helpful for me, to date I have not delivered on the promise to my users. This blog covers my recent work to do better for my customers and to focus on what they truly care about in this project.

My customers care about one primary goal: making money by betting on UFC fights.

Is that goal even possible? Perhaps, perhaps not, the point is to focus on solving that as best as I can.

Making Money by Betting on UFC Fights… Successfully

Betting on UFC fights at best would qualify as speculative behavior compared to more long term value based investing, however that doesn’t mean it is impossible to do effectively, or that we shouldn’t try. But before trying the initial scope needs to be stated.

The goal: Provide recommendations on bets to be placed on upcoming fights, these bets should deliver positive financial returns over time, and should exceed the performance of saner investments to justify their risk.

Note I don’t condone throwing your retirement on this approach, I’m certainly not.

Delivering a solid performance here is the most important function of the site for customers, more important than:

• Single prediction accuracy
• Explaining a given prediction
• Showcasing fighter trends over time

Now, to determine the baseline.

What is the target to beat?

For my primary investing I’m a [Boglehead] , meaning an automated index fund is the ideal target to beat.

Taking VTI and the range of time that the 3 models on WTAI have been active, let’s see the performance that investing $1,000.00 would have delivered:

That’s the base plot of VTI from May 5, 2023 until April 19, 2024, $1,000.00 would have turned into $1,215.21 for a 21.52% ROI.

What’s Our Baseline Performance?

To keep things incredibly simple I’ll start with a simple setup for betting:

1. $1,000.00 placed into a theoretical sports book account.
2. $50.00 placed on each prediction from the WTAI model using the odds available at the time the predictions were created.

Placing those bets evenly yields a final cash balance of $1,839.58 or 83.958%!

Plotting these returns for the year vs accuracy of the model looks like this:

There’s no point in the year where the initial stake of $1,000 is eroded, and it delivers a 83% ROI over the evaluation period.

To better visualize the returns on each event, the next plot just shows the positive or negative cash returns by each event.

Mostly gains, some losses, and some losses that build in volume. One key thing to note that drives performance over time is that we’re not re-evaluating the scale that we win or loose on. It stays at $50/prediction each time, so as we climb higher, we’re not as negatively impacted by bad week as we are earlier into our usage of the betting strategy.

Can We Do Better?

Better Via Even Betting

A starter idea is; What if we limit bets based on the confidence of the prediction? To get a feel for what model performance in accuracy looks like based on the confidence, here’s a plot for the WTAI model’s results over the window:

It certainly looks like there’s an accuracy gain starting in values above 15, but to be thorough the results are here in terms of profitability:

The magic number is where the confidence is 18 or greater. This gives a balance of a number of bets to place so that we aren’t losing too much on our losses, but also a high enough accuracy ratio to drive profitability to the max over our evaluation period. Digging into the exact data behind the plot shows 18 as a threshold for even betting delivering an astounding $1192.01!

Including Other Models

There are 3 models active on WTAI, the primary model which has been used so far in this analysis, another optimized for “profitability” which behaves a little differently, and a plain model which lacks any data about odds at all on a given fight. Unfortunately the performance of the plain model isn’t very good, so I’ll leave that one out but work to explore the returns when both models agree.

To build the next dataset, I combined the predictions of the WTAI model and Profit model with the following conditions:

1. The models must both agree on the prediction.
2. The confidence for WTAI must be 18 or greater.
3. The confidence for Profit must be 12 or greater.

This reduced us from 401 fights, to a total of 127 over the same time period. Remember the goal is to simply win more when betting, part of that should be reducing losses and volatility. The results of the same even betting strategy for this returns:

Unfortunately it looks like this approach did have a bit of a rough start before it recovered, but it has been performing nicely since September of last year, and has even delivered steady gains in the most recent round of Apex slop cards that have challenged the other approaches. Overall it delivered a profit of $600.68 or a 60.07% ROI!

The really interesting thing about the chart above is the total number of 100% accuracy events that exist.

Better Via Parlays?

A parlay is when you combine 2 or more bets together and they only pay out when both bets are successful. It is a great way to improve your returns if you are accurate, and a great way to lose your money when you are not.

With a fair coin, there’s a 50% chance (1 in 2) of getting heads or tails on a single toss. If you’re placing a parlay bet on the outcomes of two coin tosses, you’re looking at the combined probability of those events. Since each toss is independent, you multiply the probabilities to get the overall odds. For two tosses, the odds are 0.5 x 0.5 = 0.25, or a 25% chance of predicting both correctly. That’s a lot of risk even with a totally fair and balanced system, fights can be much more chaotic, so our accuracy numbers really have to drive performance home.

To keep things simple, I’ll start back with just the single WTAI model, focusing on a parlay of all of the bets when the confidence score was 18 or higher.

No need to plot this one, lots of losses remembering all the times the predictions were not 100% accurate for the day, this yielded a total of $567.58 in profit.

To see if its possible to do a bit better, now I’ll evaluate the following:

1. A parlay strategy always betting $50.00
2. Using the earlier filter of predictions where WTAI and Profit have to agree along with the confidence interval restrictions.

This brings us a return of $804.70! All for a cash exposure of $50.00/week! Viewing the returns over the entire year:

This chart sells it for me. Even with a moderately rocky start, the approach works very well with the models as they are calibrated and delivers a positive trend at the end of nearly a year with over 80% growth in our balance overall.

Putting It To Work

This week the UFC is back at the apex and we have the following fights coming up, and their predictions from the WTAI and Profit Model:

All of the details can be explored here: https://wolftickets.ai/events/ufc-on-espn-nicolau-vs-perez/6207127551655250548/

This approach this week yields only one prediction to bet on:

1
Fight: Ivana Petrovic vs Liang Na
2
Predicted Winner: Ivana Petrovic
3
WTAI Confidence: 20
4
Profit Confidence: 15
5
Odds: -530

To view a complete report on this prediction check out https://wolftickets.ai/events/ufc-on-espn-nicolau-vs-perez/6207127551655250548/ivana-petrovic-vs-liang-na/3163/

Not the most exciting thing, but definitively the MOST likely to yield a positive return over time based on this analysis.

Alternatively, you could take the route of betting evenly over each prediction from the WTAI model if you want to manage a lot more bets for the week or those just over the confidence threshold of 18. That gets you 3 predictions for this week!

Next Steps

Previously I have been working on a hedging strategy that combines parlays into groups of 2-3 fighters to look to mitigate the risk of a singular prediction going wrong. I’ll be developing this further and expanding this report reflect any learnings there.

Conclusion

Betting a parlay when the WTAI and Profit model agree with a high degree of confidence looks to be the most stable path for gains using WolfTicketsAI. This approach delivered 3x the returns compared to Vanguard’s Total Index(VTI) Fund, and did so with the least amount of bets placed compared to our other approaches.

The historical data does provide a number of various profitable approaches to using WolfTickets.AI effectively and they each have their own risk profiles:

1. Single bets on WTAI predictions with a confidence of 18 or higher. - Highest amount of betting but 119.20%.
2. Single bets on WTAI and Profit predictions with a confidence of 18 and 12 or higher respectively. - Medium volume of betting yielding 60.07%
3. Parlay from the shared collection of predictions from the WTAI and Profit models with a confidence of (18/ 12)or higher. - Least amount of betting with an 80.47% ROI.

The important thing when looking to gain from betting is to not place losing bets, where we can restrict the number of bets and increase our accuracy we can really start to achieve positive returns.

Good luck! I’ll be wagering a single unit on Petrovic this weekend!.