Jonathan Zubkoff – aka ZubbyBadger – is a top prediction market trader. A professional trader in events markets, he’s made over six figures on PredictIt, serves as Community Manager for Kalshi, and earlier this year won the CSPI/Salem Forecasting Tournament with $25,000 in prize money.
The second-place finisher of that tournament, Robert Grosse, wrote a summary (Part 1, Part 2) of his participation – his very first time participating in a prediction market. Mr. Grosse’s narrative is fascinating because prediction market and forecasting analyses tend to focus on collective outputs rather than individual experiences and motivations. Below we’re going take a similar granular look at Mr. Zubkoff’s trades that won him the CSPI/Salem Tournament.
Background
Manifold Markets is a play-money prediction market platform that partnered with both the Center for the Study of Partisanship and Ideology (CSPI) and the Salem Center of the University of Texas at Austin to host the forecasting tournament from August 7, 2022, to July 31, 2023. Like a typical play-money prediction market, users start with a set amount of “play money,” in this case 1,000 Salem bucks (denoted henceforward by “S$”) to buy and sell binary (Yes or No) event contracts. A correct prediction wins the value of the shares they purchased, and a wrong prediction wins nothing. The price of the contracts can indicate what traders believe to be the probability of that event happening. There was a total of 91 markets offered, ranging from topics like current events (Will Elon Musk Buy Twitter?), political (Will Trump Announce in 2022?), war (Will Russia Control Kherson on 5/31/23?), economics (US GDP Growth 1% or More in 2023 Q2?), and elections (Will Republicans Win the Senate in Arizona?).
Manifold operates slightly differently from other prediction markets like PredictIt. Rather than go into the details, here are links to Manifold’s description and Robert Grosse’s summaries of basic rules (Grosse’s summary of Manifold is actually better than Manifold’s):
What differs the most between play-money prediction markets (Manifold, Futuur, HyperMind) and real-money prediction markets (PredictIt, Kalshi, Polymarket) is the incentives of trading with fake money versus traders’ own money. However, in the CSPI/Salem Tournament, there was prize money and other incentives. The top 5 individuals would be finalists for a fellowship at the Salem Center, with a grant of $25,000 that “does not require teaching or in-person residency. Rather, it will provide an academic job and financial support for a researcher to do whatever they want with their time, in order to advance their career or work on other projects.” Other top traders could win various honors (with no financial reward.) Mr. Zubkoff finished first in the tournament and won the fellowship/grant.
Competition
A total of 1,000 accounts were created for the tournament (including one account for the Salem Center that made 46 trades from July 27 to August 9, 2022.) As pointed out by Mr. Zubkoff, there were liquidity issues from a lack of participation and skill:
Competition was soft. Not trying to be mean, but there just weren’t enough people playing after the midterms. Markets that would have moved on PI in a few minutes on news took twice as long in this tournament
Astral Codex Ten
To his point about soft competition, only 19 traders made more than 500 trades or earned more than $4K. Below is a chart of individual traders’ profit versus the number of traders with Zubkoff in red. The remaining 981 traders are in that cluster in the bottom left corner:
*In full disclosure, the author also participated in the tournament and is in the bottom left corner.
Regarding the lack of participation, here we can see how trading trailed off over time, with a small spike around the 2022 mid-term elections:
More tournament engagement metrics are here.
Markets & Trades
In his very first trade of the tournament, Zubkoff bet everything (all S$1,000) on Yes in the “Biden at 40% Approval on Labor Day Weekend?” market on August 25th. The next day he sold a few shares for a small gain and waited for the market to resolve. After gaining S$72.53 on top of his initial S$1,000, Zubkoff went back to trading on September 5th. He considered going all-in on the first trade as necessary given his late entry in the tournament:
If this bet had lost, I would have completely ignored the contest for the rest of the competition. At this point, I figured I had a late start and had a very small chance of winning. I had no idea how much money the leaders had, who was in the competition, or how sharp the markets would be.
Email 12/30/2023
The next 34 trades (September 5th to December 7th) were in 15 markets, 10 of which were based on the 2022 midterm elections. Zubkoff netted him S$4,089.53 and took the right position in 25 out of 30 buys. He bet against Republicans in the midterm (to his great benefit) and only bet incorrectly in the following markets:
- Will Trump Announce in 2022?
- Will Elon Musk Buy Twitter?
- Will Russia control Kherson on 10/31/22?
- Will Republicans win the House of Representatives?
Small amounts were lost in three of the markets, but as you will see in the table below, Mr. Zubkoff still profited in the Musk Twitter market.
By Spring 2023, liquidity was so low that the top traders essentially sat around for traders out of contention to make longshot Hail Mary trades that had little hope of succeeding but was the rational decision for them: If they didn’t finish in the top 5, then they had zero chance of winning the $25,000 prize. The 999th place finisher would win as much money as 6th place. Zubkoff described the final months of the contest as rather tedious:
They really dropped the ball the last month. There weren’t any new markets put up for the last month plus of the contest. The only one that was remotely in play was Trump getting indicted a third time. The rest of them were just bonds. This meant that the contest turned into a who can monitor the webpage the best and wait for someone to make a stupid bet to buy the other side. As the person who was likely to win, this sucked.
Astral Codex Ten
While overall trading declined over time, Zubkoff’s activity increased as the contest went on, with the last full week of July being his busiest week of the tournament:
Zubkoff tended to trade in markets before they were scheduled (or seemed likely) to close, minimizing opportunity costs by leaving less money tied up waiting for a market to close, and allowing him to reinvest his winnings.
Most of his trades (264 out of 514) were less than three weeks before the market resolved. If we measure the time between when a trade was created and the resolution of a market, Zubkoff averaged roughly 35 days versus 105 days for the entire field of 1,000 traders, giving him the 80th shortest average time. If we limit the field to the 131 traders who made 100 or more trades, Zubkoff had the 8th shortest time.
Overall, Zubkoff only traded in 48 of the 91 markets offered, showing a decided preference for election markets:
| Topic | Markets | Profit |
| Elections | 21 | S$5,892.33 |
| Trump (non-election) | 5 | S$9,408.11 |
| Ukraine War | 4 | -S$80.75 |
| Biden Approval | 3 | S$1,641.18 |
| Congress | 3 | S$3,735.93 |
| SCOTUS | 2 | -S$55.13 |
| Musk | 2 | S$49.85 |
| China | 2 | -S$57.58 |
| GDP Growth | 2 | S$448.33 |
| PredictIt | 1 | S$1,288.09 |
| Iran | 1 | -S$390.80 |
| Biden Cabinet | 1 | S$299.59 |
| Bolsonaro | 1 | S$2.91 |
| TOTAL | 48 | S$22,182.05 |
In looking at market topics, we see nearly all his profit came from four political markets: Elections, Trump, Congress, and Biden approval ratings. In responding to a question about his choice of markets, Zubkoff said that he tried to stick with the areas he bet on in real money markets and largely avoided the ones new to him like war. Indeed, six out of the 12 markets where he lost money were about Ukraine (3), Iran, China, or Nigeria.
Looking at individual markets, Zubkoff netted a profit in 33 of all the markets he traded in (+S$24,684.33), broke even in two markets, and lost money in 13 markets (-S$2,502.28). Here is a breakdown of his net profit and unique trades by market:
*Please note that we’re counting all transactions as a “trade.”
Below we’re going to look at Mr. Zubkoff’s trades throughout a few markets: the three Trump indictment markets (his three most profitable), one Biden approval market (his most trades), and SCOTUS Student Loan forgiveness (his largest loss of the tournament).
Will Donald Trump Be Indicted for a Crime by July 2023?
This was by far his most profitable market; even when he was wrong, Mr. Zubkoff made money. After buying S$200 worth of No in December 2022, Zubkoff sold off his No in January for a small profit (+S$17.42), and then excluding one unfilled No limit order, his next 25 trades throughout February and March were all Yes: 19 buy, 4 sell, and 2 unfilled orders (+S$4,574.67).
On March 28th another trader drove the probability down to 28%; four minutes and seven seconds later, Zubkoff bought S$1,000 Yes (2,321.9 shares) driving the market back up to 55%. Exactly one minute after that, he spent another S$291 on Yes (481.6 shares) pushing the probability up to 61%. The former trade was the single most profitable of the tournament for Mr. Zubkoff.
Total Profit: S$4,592.09
Third Trump Indictment?
The second most profitable market for Mr. Zubkoff. He started off betting on the right side (Yes), switched to No, and then quickly corrected and went back to Yes. Beginning on June 9th Zubkoff exclusively traded in Yes shares, eventually selling off in mid-to-late July for a small gain (+S$97.67) and switching to No on July 22nd. After going the wrong way from July 22nd to 27th, Zubkoff quickly sold off for a tiny loss (-S$52.41) on the evening of Trump’s indictment and a few hours later made seven Yes buys in 14 minutes (+S$2,208.47) that netted him nearly all his winnings in the market. This was the last ‘uncertain’ market of the competition and the only realistic chance he had at losing. Zubkoff described his strategy as “just trying to block the other 2 guys who possibly could have beaten me. At the end, I saw the news first and bought as much as I could.”
Total Profit: S$2,253.73
Trump Indicted Again?
The second Trump indictments were the third most profitable market for Mr. Zubkoff. He initially started with a S$0.31/share Yes order that went unfilled, and then the next seven trades from March 31st to April 19th were all No on Trump being indicted again. Despite history proving this forecast incorrect, Mr. Zubkoff still profited (+S$94.67). After April 19th, the remaining 14 trades through June 8th were all Yes (+S$2,079.58) on Trump being indicted again. He later said:
For the first 2 indictments, I think this would have been very underpriced in a real money Prediction Market. I just thought he would be indicted both times and did a good job reacting to the NYT stories that made it pretty clear that it was likely to happen. An example of the markets being soft would be that a real money prediction market would have gone to 80-90 within minutes when it came out that Alvin Bragg invited Trump to testify at the Grand Jury. This market might have gotten to 60 within 30 minutes. That is part due to the structure (there was liquidity injected preventing massive swings), part due to a small amount of people playing, and part due to weak players.
Email 12/30/2023
Total Profit: S$2,174.24
Biden 42% Approval Rating on February 1?
Zubkoff predominantly bought and sold No shares (which was the wrong prediction) for 30 of his 40 trades in this market. From January 8th to January 31st, all his completed trades were No (there were 6 unfilled Yes limited orders). After trading nearly exclusively in No positions (+S$1,320.29) Zubkoff switched to Yes (+S$354.96) with four trades over the final two days. True to form, Zubkoff profited both betting on the right outcome and the wrong outcome. He described it later as:
[T]his is another example of inexperienced traders. I trade the [FiveThirtyEight Biden] Approval markets every week on Kalshi. I have a system set up where I basically find 80% of polls within minutes when they are released publicly. In the real money markets, maybe this gives me an edge. Maybe it doesn’t. Frequently other traders will see the polls immediately as well. In the contest, it was very clear I was the only person who would see the polls at all before they were added to [FiveThirtyEight]. [If I recall correctly], this market was priced at 90 [92 in fact] when it was clear Biden’s approval was starting to drop. While it did take some predictive ability to guess what polls were coming and think it was possible that they would continue to be low, I still had a massive advantage… [If I recall correctly], the reason the side I made all the money on lost was because there were 2 Sunday polls that bumped Biden up last second. Again though, I saw these before anyone else and was able to sell/switch sides pretty quickly.
Email 12/30/2023
Total Profit: S$1,675.26
SCOTUS Strikes Down Student Loan Forgiveness?
To make the other traders feel better about themselves, we’re going to look at Mr. Zubkoff’s biggest failure from the CSPI/Salem Tournament. All nine trades were on the wrong side (No). He began with an unfilled limit at S$0.85/share order on March 2nd and then didn’t trade again in the market until late June. From June 23rd to 30th, he bought S$1,417 of No.
Being quick on the draw does have disadvantages. A last No buy of S$1,000 occurred on June 30th, because Zubkoff admittedly confused the two released Supreme Court opinions on student loans. Department of Education v. Brown was a unanimous decision in the DOE’s favor and was released by the Court first, while Biden v. Nebraska (6-3 against Biden’s student loan forgiveness) was released second and the basis for the Tournament market. After immediately catching his mistake, Zubkoff sold off S$944.43 in two transactions within two minutes of his mistake.
Total Loss: S$472.57
Zubkoff’s Accuracy
Zubkoff was right more often than he was wrong. As he stated in a comment on Astral Codex Ten, “For basically any prediction market I’ve ever been in, I’ve totally ignored the fees and have just tried to get the answer right.” In the table below we’re breaking down the 514 transactions by type and the ultimate outcome:
| Type | Right | Wrong |
| Unfilled Limit Orders | 38 | 30 |
| Buy | 168 | 117 |
| Sell | 74 | 87 |
Please note that this doesn’t reflect Mr. Zubkoff’s gains and losses, merely the outcome of an event where he owned Yes or No shares. It was possible – and he often succeeded – to profit from positions that ultimately turned out to be wrong. Zubkoff stated, “These markets overvalued whether something would happen, so I would frequently buy no.” Getting the right answer was important, but only as a means to an end. Zubkoff’s ultimate goal was to earn (play) money.
While prediction markets aggregate the collective wisdom, individual traders are motivated to participate by self-interest: money, bragging rights, or a $25,000 prize. In the past CFTC comments over Kalshi’s proposed congressional control market, it was clear that many opponents of election markets conflate price discovery with the profit motive of individual traders. There seemed to be the belief in many of the public comments that any profit motive somehow delegitimizes prediction markets or what they forecast. “Greed” was referenced in 29 public comments for the second Kalshi filing.
The profit motive is a feature, not a bug, of prediction markets. By betting Yes on events he thought would happen, or betting No on events he thought were over-valued, Mr. Zubkoff contributed to the ‘wisdom of the crowd’ in determining event probabilities by seeking profit for himself. If a partisan trader wants to distort an election market to favor a candidate, traders like Mr. Zubkoff will gladly take his money and correct the market.
Conclusion
A cursory look at the keys to Mr. Zubkoff’s success in the CSPI/Salem Tournament (among other prediction markets) include:
- Focusing on the right answer
- Reacting quickly to breaking news
- Identifying mispriced markets
- Benefiting from information asymmetry
- Staying in his lane
- Minimize opportunity costs
Zubkoff’s experience as a full-time trader seems to have given him the skills and experience to operate on a different level from most participants in the tournament. Play-money prediction markets have value and predictive power, and this tournament created unique and valuable information, but the play was not at the same level as those in real-money markets. According to Zubkoff, “The competition was extraordinarily soft. If there were 2 or 3 other people with similar experience to me, it would have been more difficult to win.” This again speaks to the profit incentive of real money markets in price discovery.
Further analysis
There are a few areas this article didn’t touch that would be worth exploring further:
- Mr. Zubkoff and other traders used tools to track price changes and trading activity. It would be extremely interesting to examine how much of an edge technology gives a trader, not just in reaction time but in monitoring other traders’ actions/strategies.
- How accurate was he compared to the tournament as a whole? Future analysis could look at the Brier scores of top individual traders like Zubkoff and Grosse versus the ‘wisdom of the tournament.’
- For all the traders in the tournament, how was success tied to buying shares of the correct forecast versus spotting price inefficiencies?
Further reading
Below are some links and sites that may be of interest:
Manifold documentation
CSPI/Salem Tournament Site
CSPI/Salem Tournament Rules
CSPI/Salem Tournament Results
CSPI/Salem Tournament Data
Python scripts and data used in this article
Robert Grosse post Part 1
Robert Grosse post Part 2
Astral Codex Ten summary
Astral Codex Ten Zubby Comment
Additionally, the team at Manifold Markets is extremely responsive to questions and very patient in explaining their market data.