2 Leg Parlays - Over 100 UFC Events Analyzed for Profitability

I’ve been analyzing prediction models across more than 100 UFC events to identify the most reliable betting strategy for WolfTickets.AI. While my previous approach using multi-leg parlays initially showed promise, its performance eventually declined, prompting this comprehensive reevaluation. This analysis revisits my Expected Value (EV) model, which compares predicted win probability against the implied probability from betting odds, to determine if there’s a more consistent path to profitability. After extensive testing across various approaches, I’ve discovered a surprisingly simple strategy that outperforms all others.

Key Finding: The 2-Leg Parlay Strategy

No matter how I slice the data, randomize it, control for other ideas, the final result is that the most consistently profitable approach to using WolfTicketsAI is:

Place a 2 leg parlay on the top 2 highest EV value predictions for each event.

That’s it.

For this week that means:

Predicted Winner	EV	Odds
Davy Grant	18.3	118
ChangHo Lee	15.5	-148

Placed as a bet:

Bet Type	Selections	Combined Odds
Parlay	Davy Grant (118) + ChangHo Lee (-148)	+266

Why I Needed a New Approach

Last year there was a long wave where multi-leg parlays of 3-4 fighters dispersed over all the predictions for an event were highly profitable, unfortunately this did not last and results over time show this drop in performance. It is also worth noting that all of the models also started to degrade for a bit but have been performing within normal bounds this year with no updates or alterations. There’s not a ton of data here but there are all sorts of factors like the number of events at The Apex, the limited history on many modern fighters, and the experiment with different gloves for a few cards. Regardless, I had to change something.

A few weeks ago a member reached out to ask about the bet selection for each week and nudged me to revisit my earlier EV work as that was proving highly successful for him over time. To read into that, check out (Un)Expected Values: Trying to find EV. I concluded here that the best betting strategy was single bets where there was a positive EV value yielding a return of +240.14%.

EV Calculations - A Recap

To identify the most profitable betting opportunities, I developed an Expected Value (EV) calculation that quantifies the difference between my model’s confidence and the betting market’s assessment. Here’s how it works:

1. Converting Model Confidence: My prediction system assigns each fight a confidence score (0-36) which maps to win percentages (51-79.17%) based on historical model performance.

2. Converting Betting Odds: I transform American odds into an implied win probability using standard formulas:

   • For positive odds (+150): probability = 100 / (odds + 100)
   • For negative odds (-200): probability = |odds| / (|odds| + 100)

3. Calculating EV: The formula is straightforward:

   1
   EV = Model's Win Probability - Market's Implied Probability

A positive EV indicates my model believes a fighter has a better chance of winning than what the odds suggest. The higher the EV, the larger the discrepancy between my assessment and the betting market’s valuation, potentially representing a more profitable betting opportunity.

Reviewing EV Performance

First I needed to determine if this was still the case or if there had been another slump as well and performance suffered here too. A quick review of all of the predictions to date yielded:

Metric	Value
Total Predictions	1,319
Positive EV Predictions	991
- Correct Predictions	617
- Positive EV Accuracy	62.26%
Negative EV Predictions	328
- Correct Predictions	244
- Negative EV Accuracy	74.39%
Overall Accuracy	65.28%
Average EV for Correct Predictions	6.46
Average EV for Incorrect Predictions	11.54

Accuracy is interesting but I still need to see how profitable EV can be, so just over the data from 2022 to present:

Neat! In this plot I can see that there are some pretty long historical slumps when using positive EV predictions to select bets but that it does seem to have paid out over time and is leading compared to a mixed or negative EV only approach.

Exploring EV Ranges, Single Bets, and Parlay Strategies

Now that we have an EV value for every prediction, the results, and the odds that were available at the time we can dig through a number of strategies to see if there’s a pattern with EV usage that holds up for profitable betting.

Slicing Through EV Ranges

Starting off, does the value of EV have any impact on ROI:

The negative EVs were not profitable, losing 5% over our period, the positives did better but it is alarming that the 10-15 range yielded an even larger drop, clearly my EV model is not perfect.

Instead of clustering into discrete buckets, what if the value was used as a filter, so only select bets ABOVE the specified threshold: Looks like things are looking up if we’re incredibly selective with our bets and go after things that have an EV larger than 25. Unfortunately this accounts for less than 8% of the total predictions and means we’d often spend weeks with no specific bets. Not ideal.

Exploring Parlays

Parlays are great at amplifying the risk but also the returns when they land. Knowing that EVs can be bucketed or used as a cutoff point, I started to explore what it looks like to explore the total size of a parlay and its performance through EV filters.

Now we’re getting somewhere! An ROI >20% comes from simply parlaying the top 2 EV predictions each week. That’s consistent information we can use to bet for every UFC event and profiting while doing it.

Stress Testing the Strategy

Seeing the results once is great, but we already saw a number of long term slumps in our data, what if we hit a rough patch? To see how reliable this pattern is I borrowed an approach from portfolio management where we take our data(predictions/bets for each UFC event) and then we shuffle it a number of times so that the events are no longer in chronological order. We have no reason to expect that one card will have any relation to the next card, and we need to know what do the results look like over a range of possible outcomes.

To test the approach in the shuffles I evaluated and plotted the following approaches:

1. Parlay the Highest and Lowest EV prediction per event.
2. Parlay the Highest 2 EV predictions per event.
3. Single betting on the highest EV per event.

To further capture how this looks like when sports betting, I started with a bankroll of $1000.00 and placed a $100.00 wager on each bet.(Each strategy received $1,000 at the start and are tracked independently). I shuffled the order of the events 100 times and evaluated each strategy over the duration of all the fights from 2022 until now.

The Results:

Some interesting points here:

1. All approaches after 10 events stay profitable on average.
2. All approaches have some number of losers until 100 events.
3. The Top 2 EV parlay is really impressive taking our bankroll from $1k to over $3k for its median and average performance!
4. I was shocked to see the high/low EV parlay consistently outperform just the single top EV prediction.
5. The least surprising datapoint is that the single prediction had the least amount of volatility(narrowest band in the plot).

Overall, I’m happy to learn there’s a consistently profitable strategy to be had here with the models as they are, freeing me up to have good results now and time to build even better models.

Coming Soon:

The EV figures are live for all of the WTAI predictions on the site, even the historical data, so you can get started selecting your bets today!

Next I’ll have some EV oriented reporting and plots, as well as a better approach for showcasing long term results on the site. I’m also still working on a better navigation/search system that lets you explore the historical data on fighters, see the previous writeups more easily, etc.

Be sure to check out all of the predictions, the writeups and the bets for this week at: WolfTickets.AI - UFC on ESPN: Emmett vs. Murphy.

Good Luck

-Chris