You should look underneath the hood of prediction markets before citing “the odds”
The choice for chair of the Federal Reserve has drawn a great deal of attention among investors, market watchers, and financial news outlets. Rightfully so — the choice for Fed Chair will take the lead on US monetary policy and bank regulation for at least four years to come. Getting an early glimpse as to who that candidate could be might be worth serious money if you are able to translate that choice into an anticipated market reaction. The mere rumor of the hawkish Kevin Warsh’s rise in the odds led to a sell-off in Treasury bonds. The re-appointment of Yellen could soothe, while the appointment of conservative Stanford economist John Taylor might shock in anticipation of sharp interest rate hikes to come.
In attempting to peer into the future, many have looked to prediction market PredictIt, where a set of contracts opened on May 25, 2017 seeks to determine exactly this question. Twitter is rife with screengrabs of the market to show who leads the betting that day, such as a now-deleted tweet by @LizAnnSonders: “Jerome Powell in the #Fed chair mix now, but with “odds” still lower than Kevin Warsh.”
Talking heads on cable news have referred to PredictIt odds, reproducing price graphs onscreen. Even Paul Krugman referred to the “prediction markets” on Bloomberg.
CNBC has also gotten in on the action, citing PredictIt in narrowing down the candidates to two: Kevin Warsh and Jerome Powell.
PredictIt can indeed be a good source for assigning a general probability to events. When people are actually putting their money where their mouths are, it is as good a proxy as any for likelihood. But it is not perfect. What would traders taking positions based on these odds do if they knew they were rigged? Well, they are rigged — and I know because I rigged them. Sorry.
Too often I see PredictIt stats bandied about without any analysis of the underlying activity or understanding of the factors influencing the apparent odds. Here I’d like to to use the Fed Chair market as a useful example to explain: a) how PredictIt works in markets like this; b) what moves the markets; and c) how to actually track a PredictIt market for insights into the betting.
How the market works
PredictIt uses a simple binary system, where for any bet you can take a ‘yes’ or ‘no’ position — long or short. For example, will Trump be impeached in 2017? You can pay between 1 and 99 cents per share, up to the limit of $850 per contract (imposed for regulatory reasons). If you’re right, and the contract resolves the way you think, you get $1 minus fees. If you’re wrong, you get nothing. You lose. Good day, sir.
After a bettor has taken a position, they are not bound to it forever. They can trade, by selling their shares to others for a profit or for a loss. This ongoing trading is what generates the information seen on the PredictIt pages.
Based on this, you might say that Jerome Powell has a 48% chance of being selected, Warsh 25%, Yellen 13%, and so on. You might also say that Powell and Warsh’s chances appear to be declining, while Yellen’s are increasing.
You can also generate cool graphs like this 30 day view showing relative daily prices plus volume for Warsh:
What moves the markets
The 30-day view, actually, gives some impression of how chaotic these markets actually are. You see Warsh climbing in price, trading paces with Yellen a few times until suddenly…on September 29 it was revealed that Trump and his team had interviewed Jerome Powell in addition to Warsh, Yellen, and Cohn. Powell had not even been listed before — you see his light blue track begin that day when traders requested a new contract be added for him. The news prompted a massive sell-off in Warsh shares, almost 9,000 shares. But the price quickly recovered.
First of all, it should tell you something that the market was completely unaware of Powell’s candidacy until Bloomberg reported it. In a fully efficient, liquid market with lots of traders, the prices should add up to pretty close to 100. Prior to Powell being added in, Cohn, Yellen, and Warsh added up to around 90 — meaning traders assigned a roughly 10% chance of an unlisted name. After he was added, the market re-stabilized once again to add up to nearly 100 with Warsh in the lead by about 10c. New news is the primary mover of markets, and you can often pinpoint to the minute a story filtered out via Twitter. But markets can be mispriced for several reasons:
Bias toward listed options
PredictIt bettors tend to overvalue the listed options even when it is known that they may not represent the complete set of possible outcomes. Hence panic caused by adding Powell into the mix.
Sunk costs
Warsh likely remained in the lead after Powell was added because, well, people had already sunk a lot of money into Warsh shares and were unwilling to part with them for something more closely approximating his “true” odds. This also often prevents shares with closer to a 0% chance of dropping all the way to 1c. People hate to sell hundreds dollars worth of shares for a couple bucks and may stop thinking in terms of odds and simply roll the dice.
Panic
This is the opposite of sunk costs. When new information appears in the market, it often goes haywire as people attempt to find a new equilibrium. This leads to people dumping thousands of shares at any price they can get simply to get out of the market with their shirts. This can temporarily distort both price and volume charts.
Media feedback loops
The CNBC article I mentioned earlier identifies Warsh and Powell as the finalists for the position. But it identifies them as finalists not via sources within the Administration but via…PredictIt. This news then gets aggregated to other sources, and stories like this get posted into the market as truncated tweets, further pushing up the prices of Warsh and Powell and suppressing everyone else. Around and around we go!
Market manipulation
Finally, the crux of the issue. The Fed Chair market on PredictIt is relatively low volume compared to other more popular markets. There are only about 12,500 shares each of Powell, Warsh, Cohn, and Yellen, so about $50,000 bet in the market total on both sides. That comes with a volume between 20–40,000 shares per contract traded back and forth over the lifetime of the market. Compare that to well over 100,000 shares per contract traded in some more popular markets and a volume of nearly 9.5 million shares traded on one contract for Trump’s 2016 election.
With only a handful of people trading, this opens the market to manipulation. If few people are buying, or selling, a cunning trader can create the impression of a market crash by selling only a handful of shares to fill all available buy orders
This is the current order table for Kevin Warsh. The sale of 570 ‘yes’ shares, or the purchase of 570 corresponding ‘no’ shares would send his price to from 25c to 11c. 1,100 shares would take him to 5c. In some cases, the floors (or ceilings) are much thinner, and it becomes possible to spend only a couple bucks to tank the price. Some traders may want to create the impression of a crash in order to spark panic and try to purchase shares at a discount. Some traders with a solid hunch of their pick, or new information, may use subtle manipulation to keep the price from spiking up rather than suddenly buying up all the available shares.
While, in theory, the market should reflect roughly the probability of an event occurring, traders profit when the prices are disconnected from the actual probability — and they benefit from keeping the prices skewed for as long as possible.
People are stupid
If you actually take 30 seconds and read beyond the Twitter headline, he actually says “I think I’m the only guy on God’s green earth that believes it will be Neel Kashkari.” Nevertheless, this one moment did this:
Kashkari remains hovering at around 5% odds, despite not having been interviewed for the position, and multiple Administration sources confirming he is not under consideration. This is partly to blame on PredictIt’s comments section. You will never find a more wretched hive of scum and villainy, and the “pump and dump” is alive and well — misleading or selectively edited quotes, and outright false information can be shared to move the markets.
How to read the market
So, we’ve discovered that markets like this can be mispriced for all sorts of reasons, especially in relatively low volume markets like this one. I hope this will caution investment firms and the Twitterati from pulling the trigger on an investment decision based on the betting odds. But how to use what is clearly a useful source of data?
- First, try to look past the horse race “who’s in first” ranking of candidates. Differences of a few cents are likely meaningless.
- Look at the bid/ask spread. A candidate trading at a certain price with a big jump up before shares are available means that either “yes” holders are reluctant to sell, or “short” buyers are reluctant to buy…and the price may increase. Likewise, a thin floor under a price or a long drop may mean that traders are reluctant to buy “yes” or sell “no”, and the price may soon drop.
- Look to volume in deciding how to weigh pricing. The day after Warsh’s 9k share sell-off, only 900 shares were traded, or about $300. That can easily be one or two nervous traders, or even an unlucky fat-fingered order.
I, for one, think it’s going to Powell. I have 2,000 shares with his name on them, or nearly 20% of the entire contract. So if you’re trusting PredictIt to place him as the front-runner, you’re really trusting me. Good call!
