A data project driven by a love for golf, statistics, and a curiosity: can data outsmart the bookmakers?

From Tee Box to Text Editor

I used to play competitive golf, representing Kuala Lumpur, Malaysia’s capital, at the national level. I wasn’t the biggest hitter or the most naturally gifted - I was nearly a foot shorter than the average player in my age group (thanks to a late growth spurt and some unfortunate date cutoffs). So I leaned into what I could control: statistics-based preparation and deliberate course management.

That edge took me further than talent alone might have.

These days, I’m no longer grinding out 36-hole weekends, but my love for golf - and for data - hasn’t gone anywhere. Lately, that’s taken a new form: sports betting. Nothing serious - just small-stakes wagers with friends - but over the past year, I’ve seen a surprising 150% return.

I’ll be the first to admit: a good portion of that was luck. But it got me wondering…

What would it take to build a model that doesn’t just predict results - but consistently outperforms the bookmakers’ implied odds?

Where Do You Even Start?

Like any data scientist, I started with the data.

Sites like DataGolf and FantasyNational, and others frequently cited on Reddit were promising - but behind paywalls. That left me with one major free alternative: the official PGA Tour statistics site.

It’s a goldmine, offering data back to 2004 - but also a logistical nightmare. Every CSV must be downloaded manually, one at a time, by selecting a specific stat, for a specific tournament.

To put this in perspective:

  • ~350 stats per tournament
  • ~39 events per season
  • That’s over 13,650 CSV files for a single year

There weren’t any public scrapers available. Most users either paid for third-party access or gave up.

So I Built One

To solve this, I built a web scraper that automates collection of all ‘Tournament Only’ statistics from the PGA Tour website - across any season from 2004 to the present.

It enables flexible, large-scale data collection without needing manual intervention.

Key Features

  • Scrape across any date range or season between 2004 and present

  • Select specific metrics using the Stat Code Reference List e.g. 02564 = Strokes Gained: Putting

  • Export clean .csv files, organised by:

    • Stat category
    • Season or date range

  • Run as either:

    • Python script (CLI)
    • Jupyter Notebook (interactive)

In addition, the GitHub repository includes fully pre-scraped data covering 01/01/2004 – 10/05/2025, so you can start exploring immediately - no scraping required.

Why This Matters

There’s growing interest in using data to inform sports predictions - but too often, the data itself is locked away, incomplete, or poorly documented. By making this tool (and the data) openly available, I hope to support others who want to:

  • Build predictive models
  • Investigate performance trends
  • Analyse course-specific player stats
  • Create more transparent, hands-on sports analytics projects

In short, this is about lowering the barrier to entry - and encouraging exploration.

What’s Next?

This project lays the groundwork by solving the data access problem - but the real challenge (and excitement) lies ahead. My next step will focus on building predictive models using this dataset, exploring whether performance metrics can reliably anticipate outcomes - and how closely those outcomes align (or diverge) from bookmaker odds.

Stay tuned for the modelling phase.

Explore the GitHub Repository

(Includes full codebase, setup instructions, stat references, and a complete pre-scraped dataset)

Saved some money using this instead of buying data elsewhere?
Consider supporting via Buy Me a Coffee - thank you!