Why You Should Be Top-Down Betting

Unfamiliar with top-down betting? Read more in the Gambly FAQ.

At 1:18 p.m. you bet on the Cardinals at -120 in their matchup with the Cubs. At 4:18 p.m. you bet the same amount on the Cubs at -105, guaranteeing a loss for the game.

This might seem like the dumbest thing ever, but if you don’t have a process that allows for the possibility of this being a sound decision, you’re doing it wrong.

Let me explain.

At any given moment in time, a team has a certain probability of winning their next game, covering the spread, scoring X points in the first quarter, and so on. There’s a range of outcomes for all these events, and the percentages are constantly changing based on things like injuries, weather, coaching decisions, etc.

Your job as a sports bettor is to estimate the probability of all sorts of different potential outcomes, compare them to odds being offered by sportsbooks, and see where their projections are most off from reality.

Thus, you can see how the initial example of betting against yourself can be smart. Perhaps the Cardinals were really 58.3% to win their matchup when you initially bet them at -120, meaning you got your money in good, and then their starting pitcher got scratched, dropping their win probability to 44.5%. That new breakeven percentage means you also got it in good on the Cubs at -105. The fact you already bet against the Cubs earlier in the day is irrelevant; you work with all the evidence you have at any given time to extract an edge. If that evidence changes, your views must shift, too.

Betting sports can become incredibly complex, but the fundamental process is simple:

Find odds on events.
Create your own odds of those events occurring.
Calculate the differences and expected ROIs.
Bet.

The way you go about doing this could be as simple as signing up for an account at DraftKings, looking at which bets look most “off,” and betting what you like. Even if you’re betting on one sportsbook and using your gut, you’re still completing this process; you’re just using your intuition to (informally) predict probabilities and calculate your ROI, right? I mean presumably you’re betting stuff you think will be profitable, and while you might not be saying “this bet is worth +3.2% to me long-term,” you’re still implicitly stating “this bet has a positive expected return.”

And that’s a bold claim to make given that most sports bettors are long-term losers (and actually not particularly close to becoming winners). Maybe your gut read on the Falcons is enough to overcome betting -110 into an event that a very sophisticated market has deemed a 50% proposition, but my hunch is it’s probably not.

But there’s a better way to do it.

Modeling Sports Contests: Very Difficult, Very Time-Consuming … Unless You Do This

So part of betting on sports is estimating the probability of various events occurring. There’s really no way around it. It could be as complex as acquiring vast amounts of data and building sophisticated models to predict on-field performance, or it could be as simple as using your gut (which is still a crude form of modeling).

This is where most people sort of break down. Like I said, you might think you have some instinctual edge in betting sports, and I believe you that you “know football inside and out,” but you’re wrong that you can just sign up for a sportsbook and start randomly betting games you like and win. You might win for a bit, but you’re not going to continue to win.

Modeling games is really hard, right? Are you really going to sit down and put in the time to build, say, a soccer betting model that can not only accurately predict on-field performance, but also do it better than sportsbooks (who are incentivized to make accurate lines with incredible financial downside if they don’t) and other bettors who are racing you to bet bad lines (and shaping odds to become more accurate across the market).

That’s a big task. Sports betting isn’t a pure peer-to-peer game, but it has peer-to-peer elements in that you need to beat lines that are shaped by many other sharp bettors.

But there’s another way to do it. What if you could tap into the market to gain access to the opinions of very incentivized parties? Not like your cousin Tommy who “has a read on the Cavs,” but serious bettors.

Well, you can. When DraftKings puts out a -110 line on both sides, what they’re really saying is “based on all the evidence we have right now, we think this bet is a coin flip, and you have to bet $110 to win $100 if you disagree.” And if FanDuel has the same line, they’re saying that, too. And if MGM has the same bet at even money on one side and -120 on the other, they’re saying something a little different: “We think one side of this bet is actually 52.2% likely to happen, and so we are charging you -120 on that side, and even money on the side we think is 47.8%.” Note that if DraftKings and FanDuel are right that the bet is truly a coin flip, the even money side of the bet on MGM would be breakeven long-term. If they were offering just one cent more (+101), it would be profitable.

Let’s say there are a bunch of other sportsbooks that agree with DraftKings and FanDuel, too. Each independent odds creator we add to the list increases the probability that the game is truly a coinflip and MGM is off, right? They all have a financial incentive to get it right, and they’re all constantly receiving information on the merits of their prediction (in the form of money from bettors).

In analyzing the differences in odds across the market, we’re in effect beginning to create our own predictions, too. This doesn’t need to be complex as is the case with originating odds from scratch via in an on-field model. By comparing odds, we’re actually leveraging the work of many sophisticated models, harnessing the wisdom of the (sharp) crowd to help build a betting foundation.

Synthetic Hold

To understand the math behind how market data can be utilized to find good bets, it’s imperative to understand hold, or the percentage a sportsbook “charges” to bet. Again, when DraftKings sets a -110/-110 line, they’re saying a bet is a coin flip and charging you $10 on your $100 bet to win $100 more back. That difference—the spread—is their hold and, assuming their odds are right, it’s how they make money. The size of the hold acts as a margin of error for the sportsbook; the smaller it is, the more accurate they need to be.

You need to reduce this hold percentage at all costs, but for you, the bettor, the “real” hold doesn’t apply to just one sportsbook; it applies across your entire betting world. If you bet at DraftKings and FanDuel, you’re going to bet a side you like where the odds are most favorable, right? The only odds that really matter to you are whatever is best on each side.

Let me show an example. Let’s say you want to bet on the 76ers/Celtics game, with the following odds posted.

If you want to bet on the 76ers, you should be doing it at MGM. If you want to bet on the Celtics, you should be doing it at DraftKings. This might seem obvious, but it’s amazing how many bettors fail to recognize the importance of getting the best price. The difference between +110 and +105 odds might seem small, but it’s literally double the ROI long-term.

You absolutely must get the best price available.

Let me put that in italics and say it again: you absolutely must get the best price available.

Let me put that in bold and say it again: you absolutely must get the best price available.

Let me stand on one foot and say it again: you absolutely must get the best price available. You can’t tell but I was on one foot as I typed that. Okay good, knowing that should really get the point across.

The reason I say most bettors don’t recognize the importance of price is because there’s an easy way to determine if they do: how many sportsbooks they use. If you really, truly understand the importance of getting the best odds, you’ll necessarily have as many “outs” as possible; it is essential to expand the pool of available odds.

It increases the number of potential bets.
It increases the likelihood of you being on the right side by chance.
It decreases the sportsbooks’ available margin for error.
It provides more information on the “real” odds.

In the above example, each sportsbook is charging 20 cents to play, and if you use just one, you’re also effectively paying 20 cents. But if you’re a customer of all three, the effective hold percentage for you is zero. Why? Because you can bet the Celtics at -110 on DraftKings and the 76ers at +110 at MGM.

Over the long run, if you can reduce the synthetic hold to 0%, you’ll break even just guessing sides. Even the teeniest, tiniest bit of knowledge of which side is best would push you into profitable territory. Or, alternatively, if synthetic hold moves negative (say, MGM moves from +110 to +111), you’ll be profitable just guessing a side and betting it.

That’s the power of reducing synthetic hold. That’s the power of having access to as many odds as possible.

A Market Data Model

As we reduce the synthetic hold, we’re getting more bets and improving the odds of getting a bet in good simply by improving the payoffs on each side. The margin for error for the sportsbook gets slashed, and we can analyze these differences between sportsbooks to make assumptions about games in a completely market-driven, sport-agnostic manner.

In the 76ers/Celtics example, let’s say we value the opinion of all three sportsbooks to the same degree. In that case, we can average the implied probability of winning. Based on the odds, the implied probability of the 76ers winning this game is:

DraftKings: 50.00%

FanDuel: 48.92%

MGM: 45.73%

The average of those is 48.22%, which is likely a better estimate of the “true” odds than any of the sportsbooks in isolation. Of course, not all sportsbooks are equally accurate and we can and should weight them accordingly, which Gambly does for you.

We’ll go with a 76ers’ win probability of 48.22% for now. Well, look at what we just did; we made a (crude but very powerful) model to handicap the game using market data and zero basketball knowledge.

Remember, since we found a bet with a synthetic hold of 0%, we only need the slightest bit of knowledge to identify a profitable bet. So let’s test out the effects by betting on the 76ers at MGM. If our true odds are correct, we’re going to lose this bet 51.78% of the time. The 48.22% of the time we win, however, we’ll make $110 for every $100 we bet.

Let’s say we bet $100 on this game and play it out 100 times:

76ers lose 51.78% - lose 51.78 x $100 = $5,178.00

76ers win 48.22% - win 48.22 x $110 = $5,304.20

The difference is $126.20, which is our profit.

Over the course of these 100 bets (the number is irrelevant and just for illustrative purposes), we’ve bet $10,000 and made $126.20. That’s a 1.26% ROI.

Let’s look at the other side of the bet. We’d be betting Boston at -110 on DraftKings, meaning we’d need to wager $110 to make $100.

Celtics lose 48.22% - lose 48.22 x $110 = $5,304.20

Celtics win 51.78% - win 51.78 x $100 = $5,178.00

The difference is $126.20, which is what we’d lose at these implied probabilities and odds. Note that the overall profit on both the 76ers and Celtics is $0, which is the result of a synthetic hold of 0%. In almost all cases, the synthetic hold is positive (meaning just blindly betting both sides would result in a loss). In every case, you’re trying to overcome the vig with your knowledge, and that becomes exponentially easier as you decrease the hold percentage.

Now you see the importance of reducing synthetic hold and getting the best price. Which means you should also see how imperative it is to have access to bet at a multitude of sportsbooks.

Why Market Data

I want to stop to reiterate that this way of betting games is just one method—and really should be just a single piece of the handicapping puzzle—but in my opinion, it’s the most crucial component. Look, someone has to sit down and handicap games, and those who do it well can find larger edges than what I’m talking about here. You see, using market data, it’s quite rare to find a bet that’s good at more than one or two available sportsbooks.

For all sports bettors, the phrase “it depends on the price” should be second nature. Everything depends on the price, but using a market data model to bet games takes that to the next level. When the vig-free market price for an underdog is +101 and DraftKings has them at +106, that’s a great bet; but we’re still talking about cents on the dollar.

Thus, a market-driven approach relies on more bets with a theoretically lower ROI than an awesome on-field model, which will result in fewer bets (but, if it’s good, a higher ROI per bet). I personally believe the former style of betting should be, at the very least, a foundational starting point for all bettors. For every bettor, casual or pro, getting the best price is as important as finding what to bet.

For casuals, though–which is just about everyone who bets–a top-down approach to betting makes the most sense. The benefits to a sport-agnostic, market-driven betting style:

It opens up the pool of available bets.

If there are enough odds on an event across the market, you can find inefficiencies in them. You don’t need to be an expert on NHL spreads or college football first half lines to bet them.

You can have a high degree of confidence that you’re right.

Again, when you model games from scratch and bet against the entire market, you’re implicitly stating “this book is wrong and this one is wrong and this one is wrong and the entire market who has bet into them is wrong.” And maybe you’re right! But to acquire all that data and take all the time to build a model and assume you have an edge others haven’t found and to constantly adjust everything to make sure it’s still good—well, call me crazy, but I think that should maybe be a Phase 2 kind of thing. Start with the lowest hanging fruit.

More bets = lower variance.

I’m no stranger to taking on risk—and risk in a vacuum isn’t a bad thing—but all else equal, you should want your 2% or 3% or 5% nightly ROI to be as steady as possible. If you’re betting $X on one game, it’s of course going to bring with it higher volatility than that same total amount spread across 20 bets. And, depending how much you’re betting per game, you’ll probably just be able to get more money down betting small inefficiencies in the market on a repeated basis. Thus, you don’t need as high of an ROI to make more total dollars.

Natural hedges also reduce variance.

True story: just minutes ago before I started writing this, I bet the Mariners at +130 on DraftKings and their opponent the Astros at -122 on FanDuel. Why did I do that? Well, all else equal, it’s basically free money.

That opportunity isn’t super common, but that difference in odds represents negative synthetic hold, i.e. just blindly betting either side will be profitable long-term. When comparing those odds to other sharper books, I calculated a vig-free market line of +126/-126, meaning both sides of the bet were good for an expected ROI of 1.78%.

Of course, this is just an expectation. You’re not really going to make nearly 2% on each side of this bet; one team is going to win, and you’ll either make money or not. But the point is that if you continue to make such a bet, you’re going to eventually settle around this 1.8% range, and the fact that you’re on both sides of the bet means you’re not going to have as many bankroll swings, i.e. you’re more likely to actually realize your ROI.

Similarly, you could easily bet on an NCAAF team -3.5 points and on their opponent with the moneyline. As long as the numbers make sense, you can and should be on both sides if the payoffs dictate it. Note that when you take this stance, you’re always going to lose at least one of the bets, and sometimes both (when the favorite wins by three points or fewer). To understand why this can still be smart, consider a scenario where you’re getting even money on the favorite -3.5 but +1900 on the underdog's moneyline. Obviously we know a 3.5-point underdog will win at a rate much higher than 5% of the time (the breakeven percentage). Thus, it follows that the spread on one side and moneyline on another side can both be good—it completely depends on price—and being flexible enough to bet those situations when they arise is a natural way to reduce bankroll variance.

It stands up well to time.

The reasoning behind using market data as a starting point for handicapping is that you’re effectively “copying” off of all the sharpest sportsbooks (with a financial incentive to be correct) and the sharpest bettors (also with a financial incentive to be correct) who are betting into them and providing feedback on how to accurately adjust their lines. Thus, when they make changes, so do you. You don’t need to be worried about how a rule change will affect the usability of your model moving forward (yet it’s still something you can incorporate if you believe it’s being undervalued by the market).

Ultimately, the top-down approach to betting is simply less time-consuming and safer than building a model from scratch.

How Gambly Can Help

Our goal at Gambly is simply to match you with good bets for you. Perhaps you know what you want to bet and just need to find it on a particular sportsbook. Or maybe you want to see an odds comparison. Or maybe you want to get alerted when specific bets are posted.

But if you don’t know what you want to bet, you can ask Gambly for help. We’ve gotten rid of all the friction that comes with actually implementing a top-down betting approach. Simply tell Gambly something like “Show me the top three strikeout props tonight” and we’ll do the rest.

Here’s the current answer for that query, by the way.

The numbers in the green bubbles represent the edge for each bet based on the real-time odds and a weighted calculation of what the “true odds” should be based on current market structure. For more on this topic, check out How Gambly Calculate's a Bet's Edge.

We’ve tried to remove all the hurdles that exist for casual bettors to quickly find the bets they want. Play around and see what you can find!