(You can also read this post on Substack if you prefer substack cookies....)

Broadly speaking, there are two ways to be wrong about the future.

The first is assigning the wrong probability to an event. To be right about the future you need to assign the correct probability. Manifold, Polymarket, Metaculus and others reward this and are very good at it. If Polymarket says Donald Trump has a 60% chance of becoming president, he probably has a 60% chance of becoming president.

This extends to conditional markets and more complicated constructions as well ("If Trump wins, what will the tax rate for the top 10% be?"). We can call this ability "Question Answering". You have a question; Manifold answers it.

But what if you wanted to know the geopolitical risks of owning TSMC? You can't arbitrarily divide the question into three sub-questions. You have to ask what options China has at its disposal.

Say you had an oracle that said, "The odds of a Chinese invasion of Taiwan by 2028 are 14%," that still wouldn't give you enough information. What about a full-scale blockade? The oracle answers 8%. Could China just blockade materials important to TSMC? What about assassinating TSMC scientists? Can China get an anti-US politician elected?

No amount of time with the oracle will tell you how much geopolitical risk is priced into TSMC's price. The second way to be wrong is to ask the wrong questions. To be right you need to be able to categorize the possible outcomes, judge their importance to the question and determine which outcomes are mutually exclusive. We can call this "Question Framing".

So far, I've assumed that Question Framing is followed by Question Answering; after the forecaster generates and categorizes the scenarios, he gives you the odds. But notice it doesn't have to be that way. We can have one person who's good at thinking through scenarios and another who's good at judging the likeliness of each scenario.

Therefore, a reasonable approach would be to find a good Question Framer and use their framing to make a market "which of these will come to pass".

A "superforecaster" is an expert Question Answerer. According to Wikipedia they're good at "inductive reasoning, pattern detection, cognitive flexibility, and open-mindedness" which boil down to eliminating biases, disregarding noise and statistical thinking. Notice, there's no reason to assume that he is a good Framer. Question Framing takes domain expertise and creativity. Incidentally, I don't think biases hurt Question Framing.

Because Framing and Answering draw on different skill sets, a superforecaster adds little once prediction markets exist. What we need are good Question Framers and the market will take care of the rest.