Base rates forecasting: the outside view that beats gut feel
Ask a fan whether their team wins the final and you get a story. Ask history how often teams in that exact situation have won and you get a number. In prediction markets, where every contract is already a number, the second habit is the entire game. This guide covers what base rates are, why the outside view is so hard to adopt and so valuable, and how to turn historical frequencies into a disciplined read on whether a market is mispriced.
What a base rate actually is
A base rate is the historical frequency of an outcome across comparable cases. Not the details of this election, this match, or this product launch, but the track record of the whole category it belongs to. How often do incumbents in this position win? How often do projects like this ship on time? How often does the favorite actually deliver?
The power of a base rate is that it prices in everything you cannot see. Every scandal that broke late, every injury in training, every polling miss, every October surprise across dozens of past cases is already baked into the frequency. Your gut, working from a single vivid story, has none of that. A base rate is a crude instrument, but it is crude in an honest way: it tells you what usually happens before you convince yourself that this time is different.
The inside view and the outside view
Daniel Kahneman drew the distinction that organizes all of this. The inside view builds a forecast from the particulars of the case in front of you: the candidate's charisma, the team's form, the roadmap on the wall. The outside view ignores the particulars at first and asks a colder question: in the reference class of similar situations, what fraction turned out this way?
Kahneman's famous example was his own textbook project. The team, looking inside, estimated about two years to finish. The one member who knew the track record of comparable teams reported that similar projects took seven to ten years and that many never finished at all. The book took eight. The inside view was not stupid, it was just built from a sample of one, and samples of one are where overconfidence lives.
Prediction markets punish the inside view with unusual speed, because the market price is itself a probability claim you are betting against. If you have not read the price as a probability and compared it to a base rate, you are trading a story against a number, and the number usually wins.
Three worked examples
How often does the pre-tournament favorite win the World Cup?
Fans treat the favorite as close to destiny. History disagrees. FIFA's own review of the first 22 men's World Cups found that the pre-tournament favorite went on to lift the trophy only about 23 percent of the time, and betting-history reviews of the modern era put the figure near 30 percent since 1978. Roughly speaking, the single most fancied team wins about one tournament in four.
That is the whole outside-view move in one line. A market pricing the favorite at 45 cents is claiming this favorite is nearly twice as likely to win as favorites historically have been. Sometimes that claim is justified, but the burden of proof sits on the inside view, not on the base rate.
How often does the president's party lose House seats in midterms?
Since World War II there have been 20 midterm elections, and the president's party lost House seats in 18 of them, a 90 percent base rate. The only exceptions were 1998, when backlash against the Clinton impeachment helped Democrats gain five seats, and 2002, when George W. Bush's party gained eight in the aftermath of September 11. Both exceptions required an unusual national shock, which tells you what kind of evidence it takes to beat this pattern.
The class can also be sharpened. Gallup's analysis of the same era found that presidents with approval below 50 percent saw their party lose about 37 House seats on average, while presidents above 50 percent lost far fewer, around 14. Same reference class, one conditioning variable, and a much more useful number. If a market on "president's party loses House seats" trades meaningfully below the low 80s in percent under an unpopular president, the base rate is asking why.
How often does the early front-runner win the nomination?
Here the honest answer is: it depends, and the class matters enormously. Pew's review of roughly five decades of primary polling found that in seven open Republican contests since 1960, the front-runner in year-before polls won six times. Early Democratic poll leaders won only four of eight open contests from 1960 through 2004, and famous nominees like McGovern and Carter were barely registering in polls taken the January of their winning years.
Move the clock back to two years out and the signal degrades further, because polls that early mostly measure name recognition. Hillary Clinton led Democratic polling for most of the cycle before 2008 and lost the nomination. Rudy Giuliani led Republican polling for much of 2007 and won zero states. A first-time front-runner two years out is holding a real but weak advantage, and a market pricing that lead like a near-certainty is leaning on the inside view.
How to build a reference class
The examples above all followed the same three steps, and you can run them for almost any market.
- Define the class before you look at the answer. Write down the membership rule first: "open-seat nominations since 1960" or "post-war midterms." If you pick the class after peeking at the outcomes, you are not measuring history, you are laundering your prior.
- Count outcomes, honestly. Include the failures and the boring cases, not just the memorable ones. 18 of 20 is a number; "midterms are usually bad for the president" is a vibe.
- Adjust for what is genuinely different. Condition on variables with a documented effect, the way approval ratings sharpen the midterm class. Each adjustment should be something you could defend to a skeptic, not a synonym for "but I like this candidate."
Base rate versus price: where edge comes from
Once you have a defensible base rate, the core comparison is simple: base rate versus market price. A contract at 62 cents implies 62 percent. If your reference class says situations like this resolve YES around 45 percent of the time, and you cannot find case-specific evidence strong enough to explain the gap, the market may be overpriced. If the class says 75 percent, it may be underpriced. The gap between an honest frequency and the current price is the raw material of every mispricing worth investigating, and the expected value math turns that gap into a decision about whether the price is worth paying at all.
Two cautions keep this honest. First, liquid markets already contain many people running the same comparison, so most gaps you find will be small or illusory. Second, a base rate is a starting point, not a verdict. The right mental model is Bayesian: start at the base rate, then move away from it only as far as the strength of the case-specific evidence justifies.
When to deviate from the base rate
Deviating is sometimes correct. The base rate said the president's party loses seats, and in 2002 the country had just lived through an unprecedented attack. The discipline is in what counts as a reason to move:
- Strong, verifiable, case-specific evidence. A resolution rule that differs from the historical cases, a structural change in how the event works, a documented shock with a known direction. These justify real movement.
- Not narrative. "Momentum," "this team is special," and "everyone I follow agrees" are the inside view wearing a costume. If your reason to deviate would have applied equally well to past cases that failed anyway, it is not evidence, it is mood.
- Move proportionally. Even good evidence rarely justifies jumping from 25 percent to 70 percent. Large deviations from a well-built base rate should be rare and should feel uncomfortable.
Common traps
- Cherry-picked classes. "Teams wearing red have won three of the last four finals" is a class chosen because it flatters the conclusion. If the membership rule would sound absurd stated in advance, it is not a reference class.
- Tiny samples. Three of the last four is a coin flip's worth of data. Small classes still beat pure intuition, but treat any base rate built on fewer than ten cases as a rough sketch with wide error bars, not a number with a decimal point.
- Survivorship bias. Databases remember winners. The startups that died, the candidates who dropped out before a single poll, and the markets that resolved quietly at NO all vanish from casual recollection, inflating your sense of how often bold outcomes happen.
- Stale classes. History from a world with different rules can mislead. Primary base rates from the era before modern primaries existed, or sports base rates from before a format change, need to be down-weighted or rebuilt.
SmartX is an AI trading terminal for prediction markets. It ranks wallets by realized PnL and win rate and streams their live positions, which is a useful reality check on any outside-view thesis: if your base rate says a market is mispriced, you can see whether consistently sharp traders are positioned the same way.
Open SmartX →FAQ
What is a base rate in forecasting?
A base rate is the historical frequency of an outcome across a class of comparable situations, such as "the president's party lost House seats in 18 of 20 post-war midterms." It anchors a forecast in what usually happens before any case-specific details are considered, which is what forecasters call the outside view.
How many historical cases do I need for a useful base rate?
More is better, but even a dozen honest cases usually beats pure intuition. The key is to widen the class until the sample is meaningful and then adjust for real differences, rather than narrowing the class until only flattering cases remain. Below roughly ten cases, treat the rate as a wide range, not a point estimate.
Should the base rate override the market price?
No. The price already reflects many traders, some of whom know the same history you do. A base rate is your prior, and a gap between it and the price is a reason to investigate, not an automatic trade. The verdict "the market is about right" is a common and legitimate conclusion.
Where do base rates come from for prediction markets?
Public historical records: election archives, sports results databases, academic datasets, and long-running polling series from organizations like Gallup and Pew. The data is rarely hard to find. The work is in defining the reference class honestly before counting, and in citing real numbers instead of remembered impressions.
Base rates will not make any single forecast right, and nothing here is a promise about outcomes. What the outside view offers is a repeatable process: define the class, count, compare to the price, and deviate only with cause. This guide is education, not financial advice. Over many markets, that discipline is the difference between forecasting and storytelling.