from Hacker News

Revisiting the algorithm that changed horse race betting (2023)

by areoform on 5/27/25, 10:03 AM with 66 comments

• by AnotherGoodName on 5/27/25, 3:22 PM

  One thing that i don't see in automated horse betting is the work to look at the horses before the race.

  You can get a sense of a horses energy and condition on the day (it can be different on different days). Horse racing even has a parade ring before the race for this reason. I don't gamble but i'm absolutely sure there's signal there in visually looking at the horse on the day that these systems don't take into account.

  I wouldn't at all be surprised if a system could be far more successful by using vision of the parade ring (whether machine vision or expert human) as additional signal.

  In fact given my experience with horses (grew up on a farm) and knowing the day to day differences i'd be surprised if any pure algorithmic system could beat the marker without the above since those who do bet on horses absolutely use this signal.

• by gosub100 on 5/27/25, 1:45 PM

  Keep in mind if you found a profitable algorithm today, you would have to put a lot of care into concealing your bets. Casinos in the US will ban you from playing if you consistently win. And if they catch you structuring your bets they can deny your payout on winning tickets. You might get away with it for a year or so, but in any sportsbook, if they see your face daily and figure out you're consistently winning, (and not just playing for entertainment) don't expect it to last.

  Edit: this was a great video about online casinos and several accounts of people who were blocked for winning too much or structuring at in-person casinos:

  https://youtu.be/unksvJ1JfDE?si=u1tUH285geBykfrC

• by bitshiftfaced on 5/27/25, 2:16 PM

  This reminds me of an anecdote out of one of Thorp's papers (2007). He joined forces with a computer science Ph.D and started sports betting. They were very profitable, but he listed three reasons why he stopped: 1) it required having someone in Vegas placing bets, 2) he believed it was risky to have someone going around with large amounts of cash in order to make bets, and 3) it wasn't competitive with his stock market trading strategy. He mentions how his belief about #2 unfortunately turned out to be correct, as apparently something happened in another group's operation.

• by PaulHoule on 5/27/25, 4:23 PM

  I play the Dr. Z system [1] by the seat of my pants when I go to the track (got family who are always inviting me to the racino.)

  The basic idea is that more people are in the win pool so the odds to win are more accurate (as a probability estimator) than the odds to place and show (finishing second and third.) If you compute probability estimates for place and show based on the win probabilities you sometimes find that place or show are underbet on the favorites. Maybe one to three races a day have a good betting opportunity.

  Dr. Z's books have a number of tables for what you should do heuristically that he got from doing simulations. I have a Python script that simulates all race outcomes and points out good opportunities that I've been thinking of recoding for my HP Alpha calculator. Practically I will do a hand-waving calculation based on the ratio of the win to place and show odds for the favorite and just bet that way.

  Th trouble with it is you only get to bet a few times a day and most people want more action that that. You tend to win most of the time because you are betting on the favorite [2] and (roughly) you have three times as many ways to win betting the favorite to show than you do the win. You don't win very much when you win, however, so just a few losses can eat up what you win.

  You can't really make a living this way and I think if you've got the grindset to do this one seriously you can apply the same skills and attitude to currency options or something like that and make a lot more money.

  [1] https://www.barnesandnoble.com/w/dr-zs-beat-the-racetrack-wi...

  [2] actually the favorite is underbet so it would be a winning strategy to always bet on the favorite if it wasn't for the track taking a cut

• by elpakal on 5/27/25, 12:56 PM

  > Four of the five seasons resulted in net profits, the loss incurred during the losing season being approximately 20% of starting capital. A strong upward trend in rate of return has been observed as improvements were made to the handicapping model.

  For anyone interested in the results

• by tlyleung on 5/27/25, 1:05 PM

  Author here - funny to see this posted after a few years. Happy to answer any questions!

• by jeffreyrogers on 5/27/25, 4:31 PM

  Algorithmic betting is widespread in horse racing now. I can't remember the exact figures but I think it's estimated to be about 40% of the total handle. There is a company that will allow individuals to connect directly to the betting pools and wager automatically. It's rumored that the biggest two bettors are betting over $1bn a year (of course a lot of that is recycled from prior payouts).

• by WalterBright on 5/27/25, 6:01 PM

  If you bet on horses, you are betting against the tide (the house cut). If you bet on stocks, you are betting with the tide (a growing economy).

  I prefer the latter.

• by msarrel on 5/28/25, 4:03 PM

  I see a lot of comments here from people saying that how a horse looks and behaves on the day of the race is irrelevant. I suspect those people have never been involved in horse racing. It's comforting to believe that statistics tell the entire story, but it isn't true.

• by farceSpherule on 5/27/25, 4:30 PM

  You can make money on horses if you know what you are doing. You have to have a betting strategy and have to stick to the strategy.

  You go to the track with $X and your strategy. When you are done, you leave. You do not stay and keep betting with your winnings.