35 thoughts on “How to Make $1 Billion Betting on Horse Racing with Machine Learning

  1. Thanks for watching everyone! Do you like data science related stories like this? Let me know other ones you would like me to deep dive into in the comments below!

  2. At least in the US, data availability isn't that great unless you're willing to spend significant amounts of money. Also, most of the race tracks in the US not only don't ban the computer model betters, they actually make rakeback deals with them to give them a big advantage over the everyday better.

  3. one very important aspect that he acknowledges at the outset is that Happy Valley is a 'closed system' , meaning horses arrive on the island and stayed for the whole season (which was not that long, and they did not have races every night). This certainly would have allowed him to better establish an accurate hierarchy (ie horse A has placed higher than horse B every time) compared to the typical scenario of horses coming in, racing, and then off to another track (as happens in the US and UK). Regarding suitability to today, you forget that the UK has Exchanges which operate differently than the Tote system Benter had to bet under. The HK Jockey Club took out 18% off the top, and Benter still overcame that 'hole' and walked away a rich man. The UK exchanges take out 1% plus a commission of 3-5% depending on who you bet with. A much shallower hole to deal with.

  4. Ken, have you ever attended the Sloan Analytics Conference in Boston? A professional bettor that appeared on the “Betting With an Edge.” Podcast and YouTube channel mentioned it. He indicated it was a conference for the analytics folks from the American pro sports leagues.

  5. I'm sorry and I don't mean to be rude or distract from your message because I truly respect and admire what do. But I've been a follower of yours for a while now, and I think you're really hot. ❤

  6. A quick note on Correlated/Colinear Features. These can be a huge pain for applications such as Sports Betting, where we are weighing certain factors relative to other factors. Concepts such as introducing dropout, random noise, and using clustering to engineer/select features might be helpful. Doing that would reduce the costs of running a model, improving ROIs even if drops absolute performance

  7. Great video Ken. I want to ask you for an opinion. I started an internship in a data virtualization project, but my goal Is to work as a data scientist or data analyst. What you think about It? Is It good starting point to enter in this area?

  8. Hey ken, Can you check the 2nd example of Kelly criterion. It looks like, you have mistakenly put 1 in down. I am really getting confused when doing this math. Please, help..

  9. Really enjoyed this style of video! Such an informative and interesting breakdown, crazy to think that model was super advanced a while ago! + title and thumbnail game is on point! Nice one, Ken!

  10. Damn it. This was my backup plan if I ever decided to have my own "breaking bad" story. I've been beaten. Time to find a new market to do this in. Perhaps slap fighting. I didn't get my degrees in math and stats for nothing! R e e e e e e e.

    Great video though, Ken. Hahaha, I love stories like this. I know ultimately at the end of the day you're only really supposed to make content that you enjoy and are proud of, but if that involves seeing more stuff like this, than I'm all for it!

  11. Finding a way to beat horse racing was the reason I got into data science to begin with.

    Since you aren't against a sportsbook employing Stanford alums, you just have to beat people and the vig.

    Needless to say, it's far easier said than done. I knew some of this, but not nearly all. This is incredible.

Leave a Reply

Your email address will not be published. Required fields are marked *