In §1, we examined the basic mechanics of prediction markets. In this section, we turn to some common objections people make; the responses will bring out more precisely some important aspects of how they work. I try to go through these in a logical order1 to let the argument “unfold” naturally; but of course, you can just skip to whichever issue bothers you most. Only the first is slightly involved and technical.

Objection: Low liquidity makes prediction market prices meaningless

Nobody has ever actually said this to me, but readers with a background in finance may be wondering if we don’t just end up with big gaps between bid and ask, and if this means we can’t really talk about a single “price”. Basically, yes, but this is a solved technical problem. Let’s explain the worry for everyone else.

Stocks are traded in order book markets: buyers and sellers post limit orders (“I’ll buy 100 shares at £99”; “I’ll sell at £101”). The highest price any buyer is currently willing to pay is called the bid, and the lowest price any seller is currently willing to accept is the ask; the gap between them is the spread. Trades clear when a bid meets an ask. Liquidity is deep, both because many people already want to trade, such as hedgers and investors reallocating their portfolios; and because professional market makers continuously post bids and asks, earning the spread. In liquid equities, this depth means the spreads are tight, and the last price is a meaningful consensus about value.

One might worry that unlike equity markets, prediction markets lack these features. There is no natural base of hedgers or portfolio allocators; nobody “needs” to hold these securities the way fund managers need to hold stocks and bonds2. Thus, order books are thin: if one person offers to sell at £0.60 and there are no bids, the “price” is just that lone offer. If another trader then posts £0.40, it may appear the probability crashed from 60% to 40%3, even though there was no shift in collective judgement — just two small players posting small orders. Contrast this with how many millions you would need to move the price of Apple stock even slightly! So maybe the relationship between price and probability in prediction markets is less robust than I claimed.

It’s a reasonable concern, and it’s exactly why modern prediction market platforms use automated market makers, such as Robin Hanson’s logarithmic market scoring rule (LMSR). Rather than waiting for a counterparty, you trade against the LMSR market maker — a piece of code that always stands ready to buy or sell, at a price determined by its current “inventory” of shares on each side of the bet. A liquidity parameter in the cost function4 sets how sharply the price responds: small trades move it only slightly, and the cost of pushing the price from, say, 40% to 70% grows nonlinearly; forcing it to 99% becomes prohibitively expensive. The result is that prices adjust smoothly rather than jumping around with each isolated order, and there is always a well-defined quote. In other words, the automated market maker guarantees continuous liquidity (that is, it subsidises trading without adding noise5), and eliminates the jumpiness of illiquid order book markets. This makes it meaningful to interpret the displayed price as the market’s best probability estimate.

Objection: Don’t financial markets and betting markets already do this?

Occasionally when I make the comparison to financial markets, it will be objected that since we already have those, prediction markets are a redundant addition. If I want to predict the future cash flows of a company, that’s a solved problem; I can just look at its stock price. Betting markets likewise already exist to provide odds on popular events like sports results. So what’s the big deal?

The response is straightforward. Financial markets were built to solve a specific problem, that of deciding how to allocate scarce capital. And as a by-product, they generate market prices that contain valuable relevant information. But many other domains present informational needs that aren’t satisfied simply as a spillover from some other system — policy effects, project timelines, likelihood of a study replicating, and so on. In these cases, it’s clear that we should generalise the principle and deliberately create markets with the purpose of providing that information.

Objection: Stupid people add noise; the result is just experts but diluted

It may have occurred to you that the putative validity of prediction markets comes from aggregating the judgements of actually informed traders. But if anyone is allowed to bet, can’t their knowledge be swamped out by traders who don’t know anything? Sure, I suggested mistakes would tend to cancel out, but what if amateurs are biased in the same direction? Worse, what if there are just no informed traders in the market at all? Fear not: as I will show, the presence of uninformed traders in the market makes the price more, not less efficient. More generally, anyone betting for any reason other than that they have information — so-called noise traders — increases the efficiency of the market.

To see why, imagine you’re in a market with no noise traders, in which everyone is well-informed and rational, and that this is common knowledge. You obtain private information about the value of the asset (that is, about the probability of the event). Do you trade on it? Well, no. You assume that since some other informed, rational participant is taking the other side of your trade, they know something you don’t. Everyone else reasons the same way, so in this market, no trades happen; nobody contributes their knowledge. This is essentially the no-trade theorem (Milgrom & Stokey, 1982) applied to prediction markets.

But in the real world, there are “stupid”6 (uninformed) participants in these markets. These noise traders create the possibility of trade, since informed traders wanting to take the dumb money become willing to act on their private signal. Of course, any individual noise trader is, by definition, adding noise, pushing the price in random directions. But the more dumb money there is, the greater the incentive for smart money to enter the market to correct it. Again, just as in financial markets, noise simply subsidises the process by which informed traders push the price to be well-calibrated.

Objection: Prediction markets can be gamed by the rich

Another common worry about using prediction markets for decision-making is that wealthy participants could manipulate them to support their preferred narrative. When Polymarket prices began indicating that Trump’s chances of winning the 2024 election were rising, many were concerned that a few traders with deep pockets might be distorting perceptions of his actual prospects7. One can only imagine how much more concern there would be if prediction markets were used to guide policy choices — say, on tax rates.

Of course, a version of this objection is immediately answered by the previous point. Trading because you want to influence perceptions is just one example of “betting for any reason other than that you have information”. It’s just offering free money to informed counterparties who see that there’s no basis for the swing. But a more sophisticated worry might be that some actors (including large corporations) may have lots of capital relative to the size of the market, and therefore be willing to lose large amounts of money for a long time in order to sustain a misleading price.

This is another legitimate consideration, that also appears solved if you consider the analogous case of traditional financial markets. There too, some players have trillions of dollars of assets under management, and comprise a non-negligible percentage of the markets they trade in. But manipulation is near impossible, not only because of how expensive8 it is to keep burning money against informed traders correcting you, but also because regulation ensures attempts to do so quickly run into legal walls.

Objection: Prediction markets prey on gullible fools & incentivise gambling (which is a vice)

Fine, so neither the existence of stupid people nor the existence of rich people deals a decisive blow to prediction markets. Still, perhaps you were put off by the explanation of how the noise generated by uninformed participants is the reward to informed traders. Doesn’t this mean that to function, they depend on bettors who have no business participating in these markets? Is that so different from gambling? Should we really be encouraging that kind of degeneracy?

Let’s first take the concern that this is a vice society ought not encourage. There is a considerable body of opinion that maintains that there is nothing contemptible about betting on one’s beliefs — on the contrary, that it is virtuous to put one’s money where one’s mouth is and own the consequences of being mistaken. Still, opinions differ, probably irreconcilably.

I offer the following points as solace: first, there are many good reasons that participants may have for trading in these markets, including that they regard themselves as being well-calibrated (and wish to test their mettle and find out), or to practise their forecasting skills. Second, demand for speculative risk-taking exists regardless, and it seems better if those energies go towards the pro-social end of providing information rather than towards zero-sum endeavours like casino slot machines or meme coins. Third, the category of noise trader is not solely made up of uninformed traders. It also includes hedgers and those with heterogeneous risk preferences, though admittedly, unlike in financial markets, these types are a minority.

However, the truly important point here is that the worry simply isn’t true; prediction markets need not depend on noise traders in this way. Notice that most prediction markets that could exist, don’t. What explains that we have high-volume markets for the outcomes of presidential elections, but none for the effect of raising the minimum wage on unemployment? The answer is that the former is just much more interesting to people. So lots of amateurs trade — for fun, for vibes, whatever — and their money attracts informed traders in the manner described. But they don’t subsidise all these other markets in the same way. If we want these markets to exist and be informative, we can’t just rely on these “gullible fools” to subsidise the market — some interested stakeholder would have to provide the subsidy. Once they do, this looks much less like clever insiders feasting on hapless rubes, and more like institutions deliberately underwriting a public good. This relates in an obvious way to the next objection…

Objection: Prediction markets need subsidies, or there won’t be enough volume to be informative

So as we have seen, not all prediction markets gain traction. And sometimes the information we want isn’t supplied for free. Snowberg, Wolfers & Zitzewitz complain in their paper Prediction Markets for Economic Forecasting, in a section titled Why They (Sometimes) Fail:

[…] design flaws sometimes prevent reliable forecasts. These flaws generally lead to a lack of noise traders (or thin markets) that reduces incentives for discovering, and trading on the basis of, private information (Snowberg, Wolfers and Zitzewitz, 2005). In order to attract noise traders, the subject of a prediction market must be interesting and information must be widely dispersed.

This by itself, you will notice, isn’t really an objection. It’s just a statement of the dynamic I’ve already described. But it then infers from the fact that noise traders often end up subsidising markets, the more questionable claim that if they are not designed to attract noise traders, they must fail. This is untrue. As Robin Hanson responds:

Noise traders are traders who subsidize your market for free, for reasons of their own, such as risk-hedging, idiocy, etc. If you fail to attract noise traders, you fail to get their free subsidy. But you can still offer to directly pay for your info, by subsidizing the market […] Here we are only talking about a “failure” of prediction markets to mooch stuff for free!

He is, of course, correct. Sometimes information is a public good, in which case it makes sense for governments to fund it. Other times, it is a private good that companies might want to pay for. Except in the rare cases where noise traders sustain a market because of their own interest and enjoyment, you must pay for what you want. It is, in this regard, like any other good. The claim of prediction markets isn’t that they can magically obtain information for free; it is that they are the most efficient way to spend one’s time and resources to acquire trustworthy information.

Specifically, the concerned actor can subsidise the market by sponsoring liquidity through the automated market maker I discussed in response to the first objection. Even a modest ongoing subsidy can yield outsized informational gains. Crucially, the LMSR mechanism not only subsidises trading, but also ensures that pay-outs flow only to traders who move prices closer to truth — guaranteeing that the subsidy rewards accuracy rather than just noise.

Objection: Prediction markets reflect popularity, not truth

Your hackles may be raised at this point. You might have to pay for this information? Won’t you just buy the crowd’s favourite story, and find out what’s popular to believe? What guarantees you get your money’s worth, and learn something about reality?

For one thing, market prices aren’t headcounts. They’re weighted by how much capital participants are willing to risk, which in turn depends on how confident they are in their convictions. Whereas a poll tells you how many people say they believe something, a market tells you how much they will stake on being right. Expressive trading is costly: if you buy because it’s fashionable, you pay. This skin in the game leads to strong selection pressures: participants who are systematically wrong lose money and fade from influence9.

For another, popular narratives with no grounding in reality just create incentives for market participants to take the free money on the other side of the trade. If you’ve been reading linearly, you may be tired of this point, but it bears repeating: whenever prices diverge from reality for any reason, they create a bounty for anyone who knows better10 to step in and correct them. The greater the divergence, the greater the incentive, and the faster they will act to correct it.

Objection: Prediction markets have made mistakes before!

That’s all well and good theoretically, you may say. But theory can only take us so far; and haven’t prediction markets often just failed in reality? Weren’t prediction markets straightforwardly mistaken about, for instance, who would become Pope in 202511? And doesn’t this sort of mistake undermine everything I’ve said?

Let’s suppose a prediction market tells us event Y has a 20% chance of happening. Then Y happens. Was the market wrong? Not necessarily. If the market were efficient, we expect to see this outcome sometimes, about one time in five to be exact. Observations like this are clearly not sufficient to claim a market was mispriced.

Ah, but you say the price was obviously wrong, ex ante. It was always clear that Y was going to happen. Nevertheless the stupidity of the market reigned. The wisdom of crowds became the madness of crowds, and despite the future being totally clear to you, prices stubbornly refused to take note. If indeed this is your view, I don’t see why you’re complaining. If true, this means you are being offered free money by prediction markets12! If you believe prediction markets are habitually offering up ridiculous results, and you know better, pick a platform of your choosing, bet on what you know to be obviously true, and extract your reward. Good luck!

Objection: Prediction markets are reflexive; they incentivise rooting for, or causing, bad outcomes

Hopefully you’re by now convinced of the epistemic efficiency of prediction markets. They’re mechanically sound, distinct from vibes, and probabilistically meaningful. However, that isn’t the only thing we care about. You might worry they have a corrupting effect on incentives. Markets, after all, sit inside the systems they predict, and so can influence the events they track. What if someone places a large bet on some event, and then attempts to bring that event about in order to profit?

A useful place to start is to consider that a similar worry holds in existing financial markets. Investors short companies they think will decline, or buy insurance against disasters. We allow this because they serve useful functions. But this already makes it possible to, say, take out a large short position against a publicly traded company, then have its CEO assassinated, and collect your profit when the share price inevitably collapses13. It’s noteworthy that this seldom happens.

You will respond, correctly, that this likely has something to do with assassination being a rather big deal, with high potential costs. But what about markets with perverse incentives for lower cost activities? For example, Polymarket had a market on whether a dildo would be thrown onto the court of a WNBA game on some specified dates. Clearly it would be all too easy for someone to bet on this happening, then do it themselves, and profit. And clearly that would be bad.

The response is quite simply that prediction markets, like all markets, have to be designed with this problem in mind. We already don’t allow several kinds of contracts because of perverse incentives: you can’t take out fire insurance against your neighbour’s home, for instance, because then you’d have reason to strike the match. Likewise, well-run prediction markets should exclude topics where individual participants can cheaply influence the outcome. I think it’s bad that Polymarket ran the aforementioned dildo market. But the most valuable use cases tend to involve large-scale, many-cause events, and such markets do not have this problem. The crucial point is to set boundaries on what contracts are permissible, and to enforce conduct rules around manipulation or conflict of interest.

Objection: Prediction markets aren’t widely used, so there’s some other problem I’m neglecting

It’s difficult to believe, but I have been accused of being “prone to rhetorical sleight of hand”. I reject such accusations, for what it’s worth, but you may nevertheless feel that I must be misleading you somehow. After all, prediction markets aren’t in widespread use; and if what I’ve said here is right, they should be. Therefore, I’m wrong.

Though prediction markets like Polymarket, Kalshi, and Manifold exist and are growing in popularity, the observation is clearly true. Part of this is about legal and regulatory issues. One thrust of my argument has been that prediction markets, as financial markets like any other, will function well when regulated accordingly. Instead, they are often just illegal, because of online gambling laws (yes, really). If these misguided laws didn’t exist, I bet they’d be used more.

But that doesn’t answer the question entirely. After all, there don’t seem to be many companies using internal prediction markets to help guide decisions. What am I missing? Hanson’s view, to which I have already alluded, is that people’s real motivations often diverge from their stated ones. Prediction markets help them achieve what they say they want — better forecasts — but in practice they care more about their status. And if a worker asked about a project deadline responded, “I don’t know, but maybe we could use a prediction market”, they’d lose status; promotions would go instead to those who confidently claimed to know.

This certainly may be part of the story, but I am obliged to regard it as something of a cop-out. In the next part, we will explore why, in spite of all the arguments in this post, rational actors choose not to use prediction markets as much as Hanson expects14.