Staturdays

Betting

  • Pebble Beach Weather Leads to Favorable Odds

    Pebble Beach Weather Leads to Favorable Odds

    I just recently started following the PGA Tour, and even more recently started dabbling in betting on golf.

    Thankfully, there’s a site called datagolf.com—probably one of the greatest sports statistics and betting sites I’ve come across—which makes it very accessible to get started. They offer live finishing odds for every player in every tournament. For paid subscribers, they even offer betting tools and expected values for various bets using live odds from popular sportsbooks. This makes it really easy to evaluate bets in realtime. (The PGA Tour site, on the other hand, is one of the worst and most poorly-organized sites for stat nerds I’ve ever used.)

    The bet I was following on Saturday was the leader Wyndham Clark’s odds of winning. He had a two-stroke lead on the field, which by the end of the round was down to one stroke to second and two-strokes to third.

    My philosophy so far has been to put more trust in the players who have been there before as opposed to the newcomers. Whether this is valid or not is to be determined. And Clark, just this past season, had won his first two pro tournaments. Second place Ludvig Åberg had four wins, but just one on the PGA tour, and third place Matthieu Pavon had just one win as well. Further bolstering my confidence was the fact the Clark was playing great golf, posting a 60 in the third round. Perhaps I was falling into the hot-hand fallacy, if that exists in golf.

    At the end of the third round, datagolf had Clark’s odds around 29% to win. DraftKings was offering him to win at +200 (33% implied probability). This initally looked like a negative expected value (EV) bet.

    Now, my impression of his play that day, his previous experience, and who he was up against bolstered my internal odds to about 33%. But there was one element that was not factored into either of these odds yet: the elements.

    The weather forecast for Sunday was dire, with heavy storms throughout the day not expected to let up until the late afternoon. There was a chance they would postpone until Monday, but there was also the chance that the tournament would be suspended entirely. It seemed almost certain that they wouldn’t play on Sunday, but whether or not they would cancel I had no way to know. However, I knew it was at least a possibility, meaning I should bump my internal odds up higher still, turning this into a positive EV bet.

    If there was a 33% chance of Clark winning if they played, and a 100% chance of him winning if they cancelled, and a 25% chance that they cancelled, then his win probability should really have been around 50%! This turns into a high EV bet. It appeared that DraftKings, and possibly other sports books, didn’t account for weather and early cancellations in their odds models.

    A probability tree of the possible outcomes for Clark if they played or if they cancelled the final round.

    Even if you disagree with my upping the initial odds from 29%, as long as you agree with at least a four percent chance of cancellation given the weather forecast, that puts positive EV on the +200 odds.

    At 33% implied odds and 50% actual odds on the current leader, DraftKings was essentially offering people free money.

    It’ll be interesting to keep an eye on weather-affected tournaments in the future to see if they’ve learned anything from this situation.

  • In-Depth: Predicting Spreads

    In-Depth: Predicting Spreads

    This is an excerpt from a recent edition of the Staturdays newsletter. Subscribe to get weekly content like this in your inbox.

    Last week we talked at length about Moneylines. This week we’re talking spreads, a topic I tend to avoid. Reason: it’s a lot easier to predict which team is going to win, especially in college football—as opposed to the NFL where there’s more parity—than it is to predict how much they’re going to win by.

    Vegas spreads are off by about 12.5 points per game, and Vegas has access to more data than you or I, or almost anyone in the public does. So it is a tall task to try to beat them at their own game. Today we’ll look at the economics of spread bets and what makes them so tricky.

    A Moving Target

    Believe it or not, Vegas doesn’t really need to get the lines right at all. They just need to get equal money on each side of them. So there’s advantage number one. While we need to be laser-accurate to make sure our prediction lands on the right side of the spread, Vegas is less concerned with that level of accuracy (although they are fairly accurate).

    The reason for this is vig. Most spreads pay out around -110, or an implied probability of 52%. So, even though there’s only one right score in the end, the probability of +2.5 covering on either side adds up to about 104%. This means that even if something only happens 50% of the time, Vegas pays out like it happens 52% of the time. If you’re wondering where your extra 2% went, the answer is: the sportsbooks have it. That’s their commission fee for setting up the market, essentially.

    We see this in moneylines too. Ohio State may have a 96% implied probability of winning, but Maryland will still have an 8% implied probability of winning too. Even though that’s not possible, it covers the sportsbook when the improbable does occur.

    So back to getting an equal audience on both sides, this ensures that people bet on something that happens 100% of the time, but only pays at a discount as if it happens 104% of the time. Let’s look at an example of what happens when the audiences aren’t equal.

    $100,000 total bet on game 1

    Team A: -110, 50% of Public Money

    Team B: -110, 50% of Public Money

    Team A wins: $95,454

    Sportsbook Profits: $4,546

    $100,000 total bet on game 2

    Team A: -110, 55% of Public Money

    Team B: -110, 45% of Public Money

    Team A wins: $105,000

    Sportsbook Losses: $5,000

    So now you can see why they want to balance the line. Even the slightest bit of the audience being off can lead to a loss on a matchup, because they have to pay out more than they took in. Of course, if the sportsbook is really confident in the line they set and they think that the public is wrong, then they could be looking at a big payday if the line is unbalanced in their favor, but I’m not sure if that’s a risk they like to take. This way is a sound strategy to consistently make money, no matter the result of the game.

    Predicting the Spread

    Last week, I showed you this chart of Vegas spreads vs. actual scores and it was surprisingly all over the place.

    So let’s look at how elo differences correlate to actual spreads and see if it’s worth trying to use to predict spreads.

    So we definitely have a clear linear relationship here. That’s good. We’re going to have a significant model. However the problem is the variance of the spread. It’s huge. The error on any given spread is going to be really wide. Let’s build the model and see what happens.

    When we build the model, we get a highly significant model, but not a very accurate one. 50% of all the predictions fall within 11.5 points of the actual score, which is not bad, but the average error on all the data is closer to 17.3 points.

    We also get a formula that tells us how many Elo points translate into 1 actual point in the game. Here’s the formula we get.

    Home Team Actual Win Margin = Home Elo Advantage * .05 + .92

    So, dividing 1/.05, we get 20 Elo points is equal to 1 actual point in the game. The .92 represents the intercept, or in our case, the number of points the home team will win by if their elo advantage is 0, meaning they’re evenly matched with their opponent. Keep in mind, we already include 55 Elo points as the home field advantage by default, so this would indicate that our home field advantage is actually too low. On average, home teams have almost a 4-point advantage all else being equal. That’s pretty incredible!

    So if a team has a 1680 Elo rating and their opponent has a 1600 Elo rating, we’d predict the home team to win by 4.92 points.

    So really this is a very similar graph, but we’re just applying the model to the data (and the same data we trained the model on, which is generally a no-no). So even though the model has seen this data before, this is the best it can do given only the elo margin, leaving a lot of variability (almost 14 points to be exact.)

    So we can’t really use this unless our spread is > 13 points different from the Vegas spread, which should very rarely be the case. That’s also only the average residual, so there’s still a good chance that our prediction is 17 or 20 points off, which won’t help us against Vegas. And even then, we’d wonder whether Vegas knows something we don’t with such a large gap in our predictions.

    The Bottom Line

    I hope I’ve painted a picture of why spreads are such a difficult bet to profit on. I’ll certainly keep an eye out for times when the spreads seem way off and there’s opportunity, but unfortunately those are few and far between.

    Of course, we have the option of adding more data to our model. Elo is one factor and accounts for about 40% of variability in the spread, but maybe with more data we can more accurately predict spreads. Well folks, that’s what I’m working on right now, and so far I haven’t had much success. But I’ll keep plugging away looking for more data that is significant in helping predict spreads and winners, and update the model when it starts performing better.

  • Calculating Expected Value Using Moneylines

    Calculating Expected Value Using Moneylines

    This is an excerpt from a recent edition of the Staturdays newsletter. Subscribe to get weekly content like this in your inbox.

    Let’s talk moneylines and expected value, something that’s essential to understand if you want to stand a chance at profiting off sports betting in the long run.

    Before I go into too much detail, let me just clarify some key terms for any new people to sports betting:

    • Moneyline: A bet placed on a team to win outright, regardless of the score. Usually you’ll see these formatted like “-350” or “+200”. These are called “American Odds”. Seeing “-350” means that you’d need to bet $350 to win $100, plus your initial bet back, so $450 total. Seeing “+200” means you would win $200 if you bet $100, for $300 total. The team with the positive odds are the underdog.
    • Spread: The points a team is expected to win or lose by. It is usually written in terms of the home team, so if it’s Alabama @ Georgia and the spread is -3, that means Georgia is expected to win by 3 points. If you see a line and it just says “Alabama +3”, that would mean that Alabama is expected to lose by 3 points in whatever game they’re playing that week.
      • If you were to bet on Alabama +3, that means you expect them to “cover” the spread, or lose by less than 3 (or win outright). If you bet on Georgia -3, that means you expect them to cover, winning by more than 3.
      • Oftentimes you’ll see the spread at half-points, like “-5.5”. That just means that if the favored team wins by 6, they cover, and if they win by 5, they don’t. This just ensures that no bet is a “push“, meaning everyone get’s their money back because the spread and actual score margin were the same.
    • Implied Win Probability: This is the win probability (WP) calculated based on the moneyline odds set by sportsbooks. You can actually take the fact that a team has +400 odds and convert that into a likelihood of them winning. To do this, you follow two simple steps:
      • If the odds are positive, do 100 / (odds + 100)
        • So for +400, it would be 100 / (400 + 100) = 100/500 = 20% Implied WP
      • If the odds are negative, do odds / (odds + 100)
        • So for -300, it would be 300 / (300 + 100) = 300/400 = 75% Implied WP

    Calculating Expected Value

    So one thing you need to do when evaluating potential bets is look at expected value. If not, you’ll end up placing bad bets and getting underpaid for your wins and overpaying on your losses, which is a good formula for going bankrupt. Expected value is basically the average you can expect to win or lose if you place the same bet many times. So if you bet $10 on a -200 moneyline (66% implied win probability), and that team actually wins that matchup 66% of the time, then you can expect to win nothing on that bet on average.

    But, if you can find a team that has +200 odds (33% implied win probability) that you actually think you know has more like a 45-50% win probability, then you can expect to make a profit on bets like that in the long run. Maybe that first one, or three fails, but over time, assuming your win probability model is tuned right—meaning teams with a 50% win probability actually win 50% of the time, not more, not less—you will profit.

    Let’s clarify this with an actual formula real quick:

    Expected Value = (Potential Profit * Predicted Win Probability) – (Potential Loss * Predicted Lose Probability)

    “Predicted Lose Probability” can also be written as (1 – Predicted Win Probability) since they’re inverses.

    Let’s also do an example real quick. I’m placing a $10 bet on a team with +200 (33% Implied Win Probability) with my predicted win probability being 50%.

    If I win the bet, my profit will be $20 ($30 total minus the $10 I bet). If I lose the bet, my loss will be the $10 I put down.

    So the formula get’s us this.

    Expected Value = ($20 * .5) – ($10 * .5) = $10 – $5 = $5 Expected Value.

    So this would be a positive expected value bet because a team that Vegas is saying wins 33% of the time, I have winning 50% of the time, and if my model is tuned correctly, then I will make money on bets like this over time.

    Elo and Moneylines This Season

    So let’s say you took every bet that Elo has recommended that had positive expected value, even if it was just 10 cents. Well, bad news: You’d be down nearly $70 after almost 200 bets, assuming each bet you placed was $10 (not advisable).

    So far this season, right around $6 in expected value has been the breakeven point, meaning you’d be profiting if you bet on all the games where a team had greater than $6 in expected value. Now ideally, when the season ends, we’d want the breakeven point to be $0, so hopefully by then that’ll be the case once the sample size has increased.

    Unfortunately, $6+ expected value opportunities don’t come along too often (only 38 times so far this year), because Vegas is pretty in-tune with what’s going on with these teams, and they also set the lines slightly in their favor no matter what (this is referred to as vig).

    Here’s an example of applying the Expected Value calculation using this week’s games. You’ll notice that a lot of the teams with the highest expected values are the undesirable teams like Navy, Georgia Southern, Hawai’i, and Kentucky, who are huge underdogs this weekend against #1 Georgia, despite being undefeated themselves. These teams have the highest expected values because Vegas pays the best when underdogs win, and can tend to overlook the improbable happening, leading to massive profit opportunities.

    A list of all the games from Week 7 of college football with positive expected value.

    If you look at the data, it’s these boring teams that pay the best. Georgia Southern, Navy, Hawai’i: these are the teams that get the good-paying odds, and when they hit, it makes up for a lot of the misses and then some. These can also be the ones where you take both the moneyline and the spread, and you hedge if they lose but keep it closer than Vegas thought, or boost your profit if they win.

    The Bottom Line

    The majority of games will have a negative expected value for both teams, so that’s why it’s important to calculate the expected value on each game before you place any bets, because otherwise you might bet on something that happens 80% of the time, and only get paid as if it happens 95% of the time, and over time that will come back and bite you (and your wallet.

    We share all the positive expected value college football games each week in our newsletter, delivered to you every Thursday morning. You can sign up for free here.