Prediction Markets -- Beginner to Intermediate

By Cultivate Labs

If you read Wikipedia’s intro paragraph for prediction markets without knowing how they function, I can almost guarantee that you'll walk away confused. I asked Adam Siegel, someone who's been explaining what prediction markets are for over 12 years, what his short explanation would be. He said the following:

Prediction markets are just a fancy name for combining people’s opinions in to a probability, or the chance that something will happen in the future. It’s kind of like a weather report where the weather man says there’s a 70% chance of rain. We can tell you what the chance is of a team winning a game, a project getting done on time, or how sales will be in Q3 just by combining the opinions of a bunch of people (ever see the show Who Wants to Be A Millionaire when the contestant chose “Ask the Audience?” That’s basically what a prediction market is doing.)

An explanation like this is perfect for Prediction Markets 101. It's succinct, to the point and leaves you understanding what a prediction market does. What I want to do in this post, however, is cover what might be taught in Prediction Markets 200 -- giving you a slightly more in depth understanding. I won't go into the math that takes place in the background of a prediction market (aka Prediction Markets 400). That’ll be for a later post, but right now, I’m only talking necessities.

The rest of this post will go through three things. First, what are prediction markets used for, second, the requirements of a prediction market question, and finally, an explanation of how the numbers in a prediction market change, which leads to accuracy.

What are prediction markets used for?

At the absolute most basic level, a prediction market is a tool that is used to forecast the outcome of an event.

“What kind of events can you forecast using prediction markets?” is something you might ask. My answer would be, “Pretty much anything.”

When we work with a company and run a prediction market internally, we ask questions related to various parts of their business, such as:

  • What will be Company X’s reported revenue for the first quarter in Y year?
  • Will Project Y be completed by X date?

Cool! And if you asked these questions to a single person, they might come up with a number. For example, they could answer $1.2 million, and nope.

If you were to ask a bunch of people these questions, take the average or median of everyone’s responses, you might get some useful information. But notice that the answers I gave, and the answers you might give, can vary widely. Prediction markets combat this by formalizing the extraction and summation of everyone’s responses, leading to concrete and actionable results.

Format of prediction market questions

First, each question must have a verifiable result. All the examples above are valid in prediction market land. Some examples of questions that don’t work for prediction markets could be:

“Do wizards exist?” won’t work because you can’t exactly prove that wrong. “Will ISIS exist in 2021?” doesn’t work because it’s too tough to define what ISIS existing means exactly, and “Should we hire a new person?” won’t work since it’s not asking for expected probability (84% we should hire someone doesn’t really makes sense and isn’t concrete).

Second, each question comes with predefined possible answers that people choose from as their answers, and these answers have to cover all possible solutions to the question. For example, they might be choosing from buckets of numbers instead of picking a number themselves. Or they might be picking from a list of candidates, rather than trying to come up with the list of people themselves.

Finally, at all times when a prediction market question is running, the answers have probabilities associated with them, which is the current believed probability that the answer will occur. The sum of these expected probabilities will always add up to 100%.

The best way to explain this is by using examples, so if you haven’t already, check out these links to the markets running on our public site to see what I’m talking about.

You might have noticed from those links that with each answer we list its “Current price”. Considering I’ve only talked about probabilities so far, this might be a little confusing. There are actually a couple different sets of terms people use when discussing prediction markets. The first set of terms is in “normal” language, which I’ve been using up until now. The other set is “market” language, which is going to be important to understand when I talk about how people change the probabilities of the answers in a question. The following sentences mean exactly the same thing in relation to a prediction market:

“Questions have answers, and those answers have probabilities that add up to 100%.”

“Markets have stocks, and those stocks have prices that add up to $100”

The main takeaway is that questions are markets, answers are stocks, and probabilities are prices.

Mechanics of a prediction market

Now that you have an understanding of what a prediction market looks like and how to read the information, I’m going to move on to explaining the secret sauce of a prediction market -- How users can change the expected probabilities of the outcomes according to their beliefs.

When you sign up for one of our public sites, you’re given an initial balance of 15,000 Clinkles, our fake currency. From there, users spend the "money" to buy and sell shares of stocks in order to align the prices with their beliefs.

If you think the current price of a stock is too low relative to your estimate, you can buy shares in that stock to push up the price and therefore the expected probability of that event occurring. If you think the price of a stock is too high compared to your estimate, you can sell shares in that stock and purchase short shares, pushing the price lower.

A question ends when either a single answer is determined to be correct, or the time frame of the question passes. People with long shares in that correct answer are given money for being correct, and people with short shares in incorrect answers are given money for being correct in those as well.

For example, if a person comes in and thinks that a Democrat winning the 2016 election has a 70% chance of happening, and the current price is $65, they can spend money to buy long shares in that stock until the price is $70. If they think the probability is actually 60%, they could spend money to buy shares short and push the price of that stock down to $60. The exact number of shares you’re buying, and the price of the stock after your transaction come from the math behind a prediction market. Overall, a person is incentivized to push markets to what they believe is the correct probability of an answer occurring, maximizing their expected return on their Clinkles.

And that’s it for Prediction Markets 200! Stay tuned for future articles about prediction market, such as the actual math behind a prediction market (Prediction Markets 400), and good mindsets and strategies when forecasting in prediction markets (Prediction Markets 405).

prediction markets crowdsourced forecasting