When former Philadelphia Phillies pitcher, Michael Schwimer, made cannon-ball-sized waves in the sports gaming space by debuting JAMBOS Picks, a lot of major industry stakeholders took notice. The new tipster service he founded was offering 17 weeks worth of sports betting picks at a price of $3,000. Claiming with great confidence a competitive advantage derived from proprietary data sets and superior algorithms, JAMBOS offered a stunning $10,000 guaranteed refund in the event that the picks distributed over the course of the subscription do not yield a net profit.
As the 2019 NFL season approached, JAMBOS leveraged the names of key investors such as New York Racing Association board member and former Goldman Sachs Partner, Steve Duncker, along with Schwimer’s stature as a former professional athlete to garner media exposure from the likes of Bloomberg and Fox Business. Schwimer is also a contributor on ESPN’s sports gambling program “Daily Wager.”
In this article we explore JAMBOS’ offering, its very misleading advertised metrics (accuracy, ROI, profitability), and consider the fate of the service. Given the scope and magnitude of JAMBOS’ early media reach, and the probability of an imminent collapse, what follows may create some negative impressions in a young and growing legal sports wagering market in the U.S. Hence the concerns, when JAMBOS and the purported guarantee first entered the stage.
Similar “touting” services have long existed
Former Philadelphia Phillies relief pitcher and local entrepreneur Michael Schwimer has raised $23 million in funding to forge a new type of business in an industry even he admits is rife with issues — selling picks for sports bets. https://t.co/kulp1wCZ9S
— PHL Business Journal (@PHLBizJournal) August 27, 2019
So-called “touting” services that charge an upfront fee for sports wagering recommendations are as old as horse racing and sports betting itself. While some “touts” may offer real value and many make handsome amounts of money, it has generally been a business model that the industry has viewed unfavorably for a number of reasons, one of which is the lack of financial accountability that touts have for what effectively amounts to investment advice.
There is, of course, nothing inherently wrong with using a data-driven approach to making sports predictions and then monetizing that work in a transaction with willing customers. One reason why touts have been viewed with skepticism over the years is the fact that often, but not always, it is the case that a person who can predict the outcomes of sporting events so proficiently that their advice justifies a significant financial cost, could make far more money if they simply placed their own capital at risk and made the bets on their own.
As such, critics feel that the tout can not possibly be as confident in their own predictive abilities as they claim to be in advertisements. One counter-argument, which JAMBOS has made, is that placing very large wagers can be challenging (see: Mattress Mack’s World Series Adventure). In Schwimer’s words, “I can’t bet $300,000 a game, but 1,000 people can bet $300 on a game.”
There is some truth and validity to Schwimer’s claim. Many operators around the world have been known for limiting or even rejecting what they deem sharp action and many profitable sports gamblers have reported similar difficulties. For touts who fundamentally lack the means to place large wagers, this claim holds water.
However, some have expressed the sentiment that they would imagine a former professional athlete with Wall Street connections, currently living in the fastest-growing sports wagering market in the world, would have the ability to easily place large enough wagers, completely within the confines of the law, such that the touting model would seem nonsensical.
There are a number of easily conceivable scenarios in which Schwimer and JAMBOS could benefit more from operating a touting service as opposed to placing wagers themselves. That said, there is another concern that critics often voice about paying for picks, which has nothing to do with the intentions of the business, but the economics of the model more broadly.
Perhaps it really is the case that JAMBOS has models that can outsmart oddsmakers in excess of 8 times per day and that Schwimer’s circumstances are such that running a tout service is easier or more preferable than getting bets down himself. In that case, especially given the high-profile nature of his business, a large number of customers acting on the recommendations will force the odds to become less favorable for every subsequent customer. The table below shows the difference in expected profit given a model with 60% accuracy, where the only difference is the average odds at which wagers are placed:
Even if customers were geographically distributed such that their bets did not make a significant impact, tales of success would reach every major operator and odds supplier, who could easily put measures in place to automatically move odds against the JAMBOS selections as soon as they came out, before customers could bet opening lines.
Assessing model accuracy
Not good.
We have made some significant updates to our NCAAB model. Hopefully it improves performance. Only time will tell. https://t.co/J3unz1EA5E
— Michael Schwimer (@mschwimer) December 3, 2019
At the crux of any prediction service is its ability to…predict. Just as a person comparing financial advisors would look at their historical ability to predict market movements, a person comparing touting services might look at their track record of picking games.
One hurdle is establishing such a track record. Verifying picks, much less the specific odds and stakes at which they were placed, is not a straightforward process. Even a tool like Twitter, which allows users to make immutable, timestamped predictions, can be subverted by scammers.
Imagine a person creates eight Twitter accounts dedicated to Monday Night Football predictions. In Week 1, four pick the home team and four pick the away. In either event, four accounts will exist in Week 2 that had the correct prediction time stamped from the week prior. In Week 2, two of those accounts will pick the home team and two will pick the away team. Finally, in Week 3, one of the remaining accounts will pick the home team and one will pick the away team.
Now, the person can delete the other seven accounts, and start advertising on the one that has seen a 100% success rate (Account #4, above). They can show users the timestamped predictions they made previously and use the purported success to justify charging money for next week’s pick. For any number of Twitter accounts, N, that a user creates ahead of Week 1, they will be guaranteed success through Week log2N.
JAMBOS has made the claim, “We provide premier sports betting and analytics backed by algorithms. Our models have been verified to pick at 59.38%.”
JAMBOS further states on their website that:
“CohnReznick was engaged to perform an examination of JAMBOS, LLC’s (JAMBOS) assertion that its Overall Total Percentage of correct predictions of for [sic] the period December 8, 2018 to February 12, 2019 (the Period) was correct…Past results are not a predictor of future results and the accuracy of JAMBOS LLC in making ANY sports type predictions may be materially different in future periods. A copy of the Independent Accountants Report examining management’s assertions as noted above, dated February 12, 2019 can be obtained through request of [email protected]”
Requesting a copy of that report via the means described has not yielded any response.
While we don’t know anything about the conditions under which this metric was generated nor the type of verification that exists, we can still investigate its merit from a purely statistical standpoint. We can look at their performance over any given period and then ask ourselves, “If they really were as good as they claim to be, how surprised should we be to find them performing in the way they are currently performing?”
A great way to think about this approach is to use a basketball example. Suppose you were a scout watching a player take foul shots. Also suppose that player claims to be a 90% free throw shooter. He takes 10 foul shots and you are surprised to see he only makes 7 of them.
You figure this could be a fluke, but when he gets to 100 shots and he has only made 70, you start to suspect that he was never actually a 90% free throw shooter, or at least that the conditions in which he achieved that success rate are different enough from the conditions you are used to such that it is a meaningless figure.
You probably have the intuition that going 70-for-100 is much more surprising (i.e. less likely) than going 7-for-10, if the player really is a 90% free throw shooter. To quantify this difference, there is at a 7.019083% chance of a 90% free throw shooter making 7 out of 10 free throws, but only a 0.000002% chance of a 90% free throw shooter making 70 out of 100.
Measuring and advertising profitability
We can use this same technique to assess claims that JAMBOS has made. One of the more recent detailed updates we received from JAMBOS came on November 1:
Financial update for our 17-week package.
If the 17-week package ended today, subscribers would be making a profit of $6,388 after fees/guarantees. Subscribers would have had to invest $29,670 (including fees). ROI of the package is +21.53%. Annualized ROI is +65.86%.
— Jambos (@JambosPicks) November 1, 2019
Over the period in question (August 27, 2019 – October 31, 2019), JAMBOS made 718 picks. Of those, 357 were winners, 336 were losers, and 25 were pushes. We can disregard the pushes and look at the 357 wins and 336 losses, which would equate to a winning percentage or accuracy of 51.52%. Recall that JAMBOS has claimed, “Our models have been verified to pick at 59.38%.” This mark of 51.52% is, obviously, significantly lower than that figure. How surprised should we be by this?
Just as we could ask how likely it is for someone who is really a 90% free throw shooter to go 7-for-10 from the charity stripe, we can ask how likely it is for a model that is really 59.38% accurate to make 326 successful predictions out of a total of 627. The answer is 0.0017%. In other words, if the JAMBOS model were truly 59.38% accurate, and that model was used to make predictions over 60,000 different 17-week periods, we would only expect them to perform this poorly in 1 out of those periods (1 out of 59,128, to be precise).
For anyone following along at home, you can download a simple spreadsheet that lets you play with the numbers at your leisure here.
Of course, “Black Swan” events do occur, oddsmakers change strategies, and the betting public exhibits varying preferences that move lines. So, while readers can decide for themselves whether they think the 59.38% number was genuine, massaged, completely falsified, or something else, what seems clear is that going forward, it would be foolish, at a statistically significant level, to expect accuracy near 60% in future picks.
Calculating Return on Investment (ROI)
There is one major confounding variable that makes the comparison of a model’s performance over two different time periods a bit difficult, and potentially casts doubt on any claims made against JAMBOS in regards to their historical accuracy. Simply looking at wins and losses does not tell us anything about the type of events that were wagered on. A person who is mostly predicting favorites to win straight up, all other things equal, would be expected to be more accurate than a person who is predicting mostly underdogs. The person wagering on favorites would also need to be much more accurate in order to be profitable, given the shorter odds at which they are wagered on.
On November 19th, at 3:00pm ET, DraftKings Sportsbook had the Oklahoma City Thunder playing at the Los Angeles Lakers with respective money lines of +525 and -715. If every wager a person placed was at odds of +525, they would need a 16% success rate to break even. On the other hand, a person who places wagers only on teams at -715 would need an 88% success rate to break even.
It should be noted that by 6:00pm ET that same day, just 3 hours later, the spread had not moved but the moneylines moved to +460 and -590, respectively.
Had JAMBOS recommended that users place wagers on Oklahoma City +525 earlier in the day, it is likely that many of those users would have had to settle for substantially shorter odds if they wanted to take action. A model that is 60% accurate at predicting games in which a +525 team will win would need to be 67% accurate to achieve the same level of profitability if those teams were +460.
Without knowing the average odds at which wagers were placed during the period for which JAMBOS states its model was 59.38% accurate, it is difficult to make claims with absolute certainty. For example, if the average odds at which JAMBOS placed wagers during the period of alleged 59.38% accuracy were -155 while the average odds during the period over which the accuracy rate was observed to be 51.52% were -110, the latter situation would actually be the one in which the greater profit (or smaller loss) was achieved.
This can be drawn analogous to the process of selecting a financial advisor. Simply looking at the percentage of instances in which a hypothetical financial advisor successfully predicted that specific stocks would increase in value ignores the magnitude of that increase (or decrease). An individual who is 60% accurate in selecting stocks that tend to increase by a small amount would not deliver the same amount of profit as an individual who is 60% accurate in selecting stocks that tend to increase by a larger amount. This is part of the reason why the measure of Return on Investment (ROI) is used so frequently, both in the investment world and the gambling space.
ROI and wrapping paper
One specific matter within the nebulous menagerie of JAMBOS-related criticisms that seems to require the most immediate attention is the way in which the historical ROI of their packages are advertised. It is a matter that has gained attention within a small corner of Twitterverse as well as a message board known as BettingTalk. It may not necessarily be the case that the misrepresentation outlined below was deliberate and it is entirely possible that the deception to be discussed was unintended. Schwimer is clearly aware of the matter and stands by the calculation, but it is possible that his misconception is accidental or derived from information being fed to him via sources unknown to us.
He has no clue how to calculate it but think he does. At the most basic form, they have released 698 plays at $300 a play. That’s $209,400 in total capital deployment. Yet somehow he thinks his ROI and other numbers are correct. That’s ridiculous.
— SF492 (@SF4922) November 1, 2019
It seems that users are, justifiably, confused by the claim that $6,388 represents an ROI of 21.53%, specifically because they take issue with the claim that the amount which needed to be invested was only $29,670. With metrics based on a unit wager of $300 and a total of nearly 700 picks, it seems that users would have had to risk or invest at least $300 * 700 =$210,000 by this point, rather than the significantly smaller number suggested by JAMBOS. This brings up the need for a discussion of Return on Investment.
Return on Investment is calculated as a ratio with net profit in the numerator and amount invested in the denominator: ROI = Net Profit ($)/ Amount Invested ($)
With the JAMBOS matter at hand, it appears to be a controversy over how the denominator is tabulated. Schwimer even posted a screenshot of a text message he wrote speaking to these concerns.
To summarize the screenshot Schwimer posted, if you played ten $100 hands of blackjack, you would likely consider the denominator to be 10 * $100 = $1,000, whereas JAMBOS would consider it to be the amount of chips you purchased (i.e. how much actual cash you exchanged for chips), regardless of how much you wagered.
We can use a couple of examples to illustrate the disparity between the way in which JAMBOS reports ROI (again, with the caveat in place that it may not be intentional) and the way most customers are likely to interpret it.
Suppose you sat down at a roulette table planning only to bet on black or red, and only in $10 wagers. Suppose, similar to our basketball player taking free throws earlier, you claimed to have a system for predicting whether the result of any given spin would be black or red with 90% accuracy.
If your first three predictions are all correct but your fourth is incorrect, you will have won $30 and lost $10 (recall that you are only placing wagers in $10 units) for a net profit of $20. At this point, what would you say your ROI is? You risked $10 on four spins for a total of $40 risked. Your profit was $20. So you have a $20 / $40 = 50% ROI. Right?
Not if JAMBOS is doing the accounting. They would agree that your profit was $20, but they would say your amount risked was only $10 because that is the amount of cash you exchanged for chips without needing additional funds to reach this point. So, rather than reporting an ROI of 50%, JAMBOS would call this a $20 / $10 = 200% Return on Investment.
When confronted on Twitter by users alleging that this was disingenuous at best, Schwimer justified his methodology by likening it to the metric many hedge funds use, Return on Invested Capital (ROIC). The thing is, he is not entirely wrong in making this analogy, but it is objectively misleading in the context to which the analogy is being applied, even if accidentally so.
In order to properly convey the danger of misusing these metrics, deliberately or inadvertently, it helps to understand their genesis and use cases. Let us take a step back and consider things conceptually and philosophically. What is the point of a measure like ROI or ROIC? What is each trying to measure, and what sort of decisions are based on each?
To understand why, how, and when ROIC is most appropriate, imagine you have a savvy investor friend who claims he can take money you give him, invest it in the stock market, and make it into more money.
If you gave him $100, and he invested that full $100 in the stock market, assessing his performance would be very easy. If the $100 becomes $110, you received a $10 Profit / $100 Invested = 10% ROI.
But, what if he only finds $50 worth of stocks to buy and your other $50 just sits in a shoe box in cash. If the $50 in the stock market becomes $55, what is your ROI now? You gave your friend $100 and he gave you the $55 from the stock market plus the $50 in cash for a total of $105 back, so it seems like your ROI is $5 Profit / $100 Invested = 5%.
But, is that really an accurate measure of your friend’s ability to make good investments? He might argue that factors over which he had no control limited the number of investment options, but that for the portion your capital he did invest, he made $50 into $55, and ROI of $5 Profit / $50 Invested = 10%.
In the future, he tells you, when comparing his performance to other friends trying to invest your money, you should look at the ROIC, perhaps alongside the ROI.
The metric of ROIC is useful if and only if the friend is both making stock picks AND purchasing those stocks at the right price at the right time. However, if the friend were merely publishing stock picks and charging you a subscription fee, this would be problematic.
Why is using the “ROIC” method inappropriate for JAMBOS to use? If you buy into a service like JAMBOS, you are trusting that their advertised level of accuracy and profitability will persist over the long run. Suppose it really is the case that JAMBOS will predict game winners with ~60% accuracy over the course of 17 weeks, across the 1,000+ picks they guarantee to deliver.
Consider any arbitrary set of 5 games. At 60% accuracy, any combination of 3 wins and 2 losses would be expected with equal likelihood. For example, the following series of pick results would be expected with the same probabilities in a case where the model is shown to be 60% accurate:
Imagine that JAMBOS advertised that, during a trial period, their record was W-L-W-L-W, as shown below:
In considering whether to pay for a subscription, would you be trusting that the next 5 predictions they make will be correct and incorrect that exact order? Or, just that 3 of the 5 will be correct? Hopefully, it’s the latter. There is a 60% chance that a model with 60% accuracy will go 3 for 5, but only a 3% chance that it does so in the exact order of W-L-W-L-W.
Consider the earlier roulette example where you placed four consecutive $10 wagers on Black, won the first 3 and lost on the final spin (W-W-W-L). Conventional wisdom would calculate ROI at $20 Profit / $40 Wagered = 50% whereas JAMBOS would calculate the ROI at $20 Profit / $10 Chips Purchased = 200% .
However, imagine if you lost the first roulette spin instead of the last one. You would have started with $10 and saw your bankroll decrease to $0. In order to place any additional wagers, you would need additional cash. If you opted to purchase an additional $10 chip, and won your bet the subsequent three spins in a row (L-W-W-W), you would have achieved a net profit of $20, just like before, and an ROI of $20 Profit / $40 Wagered = 50%. However, the JAMBOS ROI calculation would now be $20 Profit / $20 Chips Purchased = 100%. Since the series of events W-W-W-L and L-W-W-W are equally likely to occur, does it really make sense to report an ROI metric that changes based on which of those specific sequences happen to occur?
Thus far, this article has discussed the way in which the denominator in the ROI calculation is defined, because it would seem rather unintuitive to talk about variations in calculating the numerator (Net Profit). Typically, one simply takes the amount that they’ve lost and subtracts it from the amount they’ve won and, Voila! – net profit.
Unfortunately, this logic seems to not hold in the case of JAMBOS, at least according to the aforementioned November 1st tweet where subscriber profits were listed at $6,388 at an annualized ROI of 65.86%.
Fortunately, we can easily corroborate this information by looking at the verified historical records that have been published on the site since late August. The particular tweet in question concerned a 17-week package that began September 3rd. What becomes quickly apparent and was soon noticed by respondents to that Twitter thread, is that the $6,388 profit quoted actually includes the $10,000 guarantee. More specifically, it is composed of the net loss taken on predictions made at a rate of $300 per unit, the $3,000 subscription cost, and then the $10,000 refund.
This is EXACTLY correct! That is what makes us COMPLETELY different than other touts. With other touts a subscriber loses and thats it. With our service, we will pay you if our picks can't beat the market. Therefore providing the subscriber with financial value.
— Michael Schwimer (@mschwimer) November 1, 2019
Many people have a gut reaction here and feel it is intentionally misleading. Whether it is intentional or not, what is clear is that this tweet does not accurately represent the financial reality of a person who purchased the picks in question. After the backlash on Twitter, JAMBOS seems to have stopped publishing updates with the same amount of detail. A few days later, another user posted their calculations for the actual levels of profitability for the two 17-week packages that had been sold:
Haven't seen any 17-Week package upd8s from @JambosPicks this morning, so:
17-Week Package 01: -13.17 units, -$6951.00 for a $300 bettor (including fees)
17-week Package 02: -12.64 units, -$6792.00 for a $300 bettor (including fees)
*unvested guarantees excluded bc lol come on
— Cizzle (@CizzlingSports) November 4, 2019
In reality, a person would have had to pay the $3,000 subscription fee and then risk more than $200,000 on the 659 picks recommended over this period, just to be an additional ~$1,000 in the red, and then wait another two months to hopefully receive a refund. Of course, if at this point the subscriber chooses to stop taking the JAMBOS recommendations and somehow the performance improves enough to be profitable, the subscriber would not be eligible for a refund.
The JAMBOS Guarantee
Ultimately, however, some JAMBOS supporters may argue that as long as they make good on their guarantee to make all purchasers of their package financially whole, the record and the rest shouldn’t matter. The Terms & Conditions are worded as follows:
“If at the end of your subscription period, we are down units for that subscription period, you will receive the guarantee amount for that subscription period by receiving a credit to your JAMBOS Wallet. Anytime you have money in your JAMBOS wallet you can cash it out without any restrictions. If at the end of your subscription period we are up units, you will not receive a credit to your JAMBOS wallet in the amount of the guarantee, however you may receive a deposit in your wallet for a smaller amount in accordance with the individual package terms based on how many units the recommendation are up.”
“The guarantee associated with your specific package will be deposited into your JAMBOS wallet by 5:00pm ET on the Tuesday after your package ends.”
“JAMBOS wallet balances are paid out to subscribers via ACH. Assuming that all information provided by the subscriber is accurate, the money will be deposited to the designated bank account within 5 business days. Our goal is to get it out as quickly as possible.”
Critics online have alleged that the secret to JAMBOS’ “success” will lay in planning or finding ways to not pay out on guarantees in the event of a package failing to deliver profitable recommendations. To JAMBOS’ credit, they have stated that the guarantee will be paid to the customer based on the performance of the recommendations, not actual bets that were placed.
Indeed, many customers will likely have trouble getting the exact odds recommended by JAMBOS and perhaps some subscribers are merely betting against the company’s ability to deliver on its promise in hopes of collecting on the “guarantee.” The way the JAMBOS Terms & Conditions are worded, it seems clear that a user who pays $3,000 for the 17-week period will be eligible for the $10,000 refund if those picks would yield a net loss, even if the user did not actually place any bets themselves.
Selling just 100 subscriptions (a milestone JAMBOS is rumored to have failed to reach with their first 17-week package) would result in a liability of 100 * $10,000 = $1,000,000 that JAMBOS would need to have on hand, in cash. It seems to clearly be the case that they raised far more than this amount of capital from investors. They also claim to post weekly affirmations from their bank, Capital One, stating that JAMBOS indeed has the cash on hand to pay these obligations.
This appears to be a step in the direction of integrity, fidelity, and verifiability. Of course, just because a company has the ability to pay an obligation, that does not necessarily mean that they will or can be forced to. Further, to play devil’s advocate, we see that the letter from Capital One states that JAMBOS maintains an escrow account that, “as of the date of this letter, has a balance equal to or greater than the JAMBOS Total Subscription Liability Amount as indicated on the JAMBOS Liability Report…” But, we do not know what they have actually stated that liability amount as nor do we know how it was calculated. One could imagine, just for the sake of argument, an entity in JAMBOS’ position making the claim to their bank that a large number of the guarantees will never have to be paid for X, Y, or Z reasons, and thus the Total Subscription Liability being referred to in the letter could be an amount far less than what the public might assume it to be.
Conclusion
Ultimately, the metrics that JAMBOS advertises (accuracy, ROI, profitability) are very misleading and do not accurately represent their offering, though it is not clear if and to what extent this is deliberate. Supporters of JAMBOS point to the data science team on staff at Schwimer’s other venture, Big League Advance, as evidence of the ability to create betting models with a true edge.
That venture, incidentally, reportedly raised of $150 million in venture funding. Detractors, on the other hand, point to stories such as the lawsuit filed against Big League Advance in which a Cleveland Indians prospect claimed that they “used ‘unconscionable’ tactics to get him to surrender a substantial stake of his future big-league earnings in exchange for an up-front payment in 2016.”
Both prospective investors and potential customers need to approach such offers with extreme caution. There may be a need for regulation or licensure at some point down the road, as the standard remedies available at law are likely not sufficient to cover the potential losses caused by a sophisticated but fraudulent touting service. That is not at all to say JAMBOs is knowingly or unknowingly perpetrating a fraud, but the situation certainly gives rise to a justification for ensuring that consumers can trust and rely on the claims and advertisements of such entities down the road.
