Presenting Fractalized Tennis Bets™ – a decision support application to help you scientifically evaluate your intuitions about tennis!

Fractalized Tennis Bets™, designed by FDm, demonstrates how Generative Data Modeling, Bayesian Inference and Decision Making can be used to evaluate whether serious capital allocation decisions should be made based on subjective opinions.

The Tennis Context

As an ardent follower of tennis, you may believe your astute understanding of the sport helps you predict how a specific tournament will play out or how a particular set of players will perform in the tournament.

For example, you may believe:

• If Djokovic crosses the quarter finals of a grand slam, he will go on to win it; OR
• The Women’s Australian Open in 2024 will be won by a player outside the top 10 seeds.

If you are in a betting mood, you may use this belief to risk your own capital and place a bet on one of the many sports betting websites available. Even if you weren’t in that mood, you may feel regret or relief when you observe the reality of the tournament validating or invalidating your opinion.

What if, instead of going on this emotional roller coaster ride, you could answer some important risk management questions based on which you could take a well calculated risk:

• Does historical evidence support your opinion?
• If such supporting evidence exists, what is the probability that your opinion will play out in an upcoming tournament?
• If these probability estimates are encouraging, should you go with a naive betting strategy that “follows your opinion”?
• With such a strategy:
  • What is the probability that you will make positive returns when your opinion is validated?
  • What is the probability that your losses won’t exceed your maximum tolerance?

To achieve this you would need access to a reliable simulation and backtesting framework, which is powered by models that bear the validation of skilled data scientists.

With Fractalized Tennis Bets™, you delegate the job of building a trustworthy simulation and backtesting framework to a skilled data scientist, while retaining control of the key decision variables (i.e the answers to the above questions) which guide your evaluation of your intuition.

Does this go beyond tennis? What else can this be used for?

The motivating risk management questions above can be mapped to questions that organizations, communities and individuals should be asking themselves when evaluating any high stakes decisions.

A few examples of high stakes decisions faced by individuals, families and organizations are outlined below:

• As a family or individual: We / I have an opportunity to relocate to a specific country – should we / I go ahead?
• As an investment advisor: Should I recommend this specific portfolio of financial products to my customers?
• As a tech startup: Which areas should I build in-house expertise in, and which areas should I buy expertise?
• As an enterprise: Should I spend more on customer acquisition or on customer retention?

The varied domains and scales of these questions often obfuscate the underlying similarities in their structure. Identifying these analogies and being able to map the underlying mathematical constructs to different domains and human-interpretable metrics requires hands-on practice with a relatable example.

With Fractalized Tennis Bets™, FDm presents a worked example that illustrates how a risk management mindset, combined with Generative Data Modeling, adds high conviction to high stakes decision-making.

Who are the users of this app?

The Fractalized Tennis Bets™ roadmap speaks to 3 groups:

• the Tennis Enthusiast (User Persona representing Business Domain Experts)
• the Data Scientist (User Persona representing Technical Domain Experts )
• Machines (i.e the algorithms and visualizations) which enable the user personas above to take smart decisions.

For a bite sized preview of what the experience will look like..:

• Take a look at this whirlwind tour of the end to end app 
• Double click into the proposed UX for the Tennis Enthusiast persona
• Double click into the proposed UX for the Data Scientist persona

The Roadmap & Current Status

The Roadmap below outlines our plan to build out this app.

We have recently launched v0.1 of the app. In this version, we have a working implementation of one user interface and descriptive stubs of user interfaces which will be released in subsequent milestones.

Look out for subsequent updates as we work through the roadmap and build out decision support systems aligned with each target user story per version.

Speak to Us!

Have any questions, want to suggest an “opinion” or a “betting strategy”?

Or maybe you want to find out more about becoming an FDm subscriber or collaborator?

We’d love to hear from you, please get in touch at info@fractal-data.com or join our dedicated LinkedIn community.

———–

FAQs

I have come across the term ‘scenario’ quite a few times – what exactly is a scenario? 

A scenario, in the tennis bets context, is a specific way in which a tournament is played out. 

For example, if you look at the completed draw of Wimbledon 2023, it represents a historical scenario, i.e. this is how the tournament was actually played out. 

With generative data modeling, we use models to attempt to predict multiple possible ways in which a given tournament could be played out. For example, you could request our generative Tennis simulator to give us 1000 possible ways in which Wimbledon 2023 could have been played out. Each of these 1000 possible outcomes for the tournament is also a scenario, albeit a generated one. 

What are some of the betting strategies that could be applied in tennis betting? 

Betting strategies can range from very simple to very complex ones. Some simple betting strategies that could be used are:

• Always bet on the odds-on favorite (or the underdog)  in any match 
• Always bet on the higher ranked player in any match
• Always bet on the player with a better Win/Loss ratio looking back X months

Some complex betting strategies could take a bunch of the simplistic ones and fire them at different points in a tournament. 

Some other complex strategies could aim to optimize PnL against multiple possible scenarios in which a future tournament may be played out. This may be something we explore further in a future version of this app.

How much historical tennis data are you dealing with? 

For the sake of simplicity, we have constrained the dataset powering this version of the app to

• Men’s tennis tournaments
• From 2001 onwards (though models may be trained only on very recent years for the learnings to be relevant for current players)
• Tournaments that follow the Grand Slam format, with 32 to 128 matches per tournament

Acknowledgments

Our hat-tips to the giants on whose shoulders this app is being setup!

Open source packages 

• pandas: https://pypi.org/project/pandas/ 
• pytest: https://pypi.org/project/pytest/ 
• matplotlib: https://pypi.org/project/matplotlib/ 
• streamlit: https://pypi.org/project/streamlit/ 
• openpyxl: https://pypi.org/project/openpyxl/ 
• xlrd: https://pypi.org/project/xlrd/ 
• playwright: https://pypi.org/project/playwright/ 
• pytest-playwright: https://pypi.org/project/pytest-playwright/ 
• pytest-cov: https://pypi.org/project/pytest-cov/ 
• pymc: https://pypi.org/project/pymc/ 
• graphviz: https://pypi.org/project/graphviz/ 
• plotly: https://pypi.org/project/plotly/ 
• flake8: https://pypi.org/project/flake8/ 
• flake8-docstrings: https://pypi.org/project/flake8-docstrings/ 
• requests: https://pypi.org/project/requests/

Tennis Data Source 

All historical data powering this app is sourced from http://www.tennis-data.co.uk.