Tag Archives: Prediction Markets

Don’t Be “Rationalist”

The first principle is that you must not fool yourself — and you are the easiest person to fool. Richard Feynman

This blog is called “Overcoming Bias,” and many of you readers consider yourselves “rationalists,” i.e., folks who try harder than usual to overcome your biases. But even if you want to devote yourself to being more honest and accurate, and to avoiding bias, there’s a good reason for you not to present yourself as a “rationalist” in general. The reason is this: you must allocate a very limited budget of rationality.

It seems obvious to me that almost no humans are able to force themselves to see honestly and without substantial bias on all topics. Even for the best of us, the biasing forces in and around us are often much stronger than our will to avoid bias. Because it takes effort to overcome these forces, we must choose our battles, i.e., we must choose where to focus our efforts to attend carefully to avoiding possible biases. I see four key issues:

1. Priorities – You should spend your rationality budget where truth matters most to you. You can’t have it all, so you must decide what matters most. For example, if you care mainly about helping others, and if they mainly rely on you via a particular topic, then you should focus your honesty on that topic. In particular, if you help the world mainly via your plumbing, then you should try to be honest about plumbing. Present yourself to the world as someone who is honest on plumbing, but not necessarily on other things. In this scenario we work together by being honest on different topics. We aren’t “rationalists”; instead, we are each at best “rationalist on X.”

2. Costs – All else equal, it is harder to be honest on more and wider topics, on topics where people tend to have emotional attachments, and on topics close to the key bias issues of the value and morality of you and your associates and rivals. You can reasonably expect to be honest about a wide range of topics that few people care much about, but only on a few narrow topics where many people care lots. The close you get to dangerous topics, the smaller your focus of honesty can be. You can’t be both a generalist and a rationalist; specialize in something.

3. Contamination – You should try to avoid dependencies between your beliefs on focus topics where you will try to protect your honesty, and the topics where you are prone to bias. Try not to have your opinions on focus topics depend on a belief that you or your associates are especially smart, perceptive, or moral. If you must think on risky topics about people, try to first study other people you don’t care much about. If you must have an opinion on yourself, assume you are like most other people.

4. Incentives – I’m not a big fan of the “study examples of bias and then will yourself to avoid them” approach; it has a place, but gains there seem small compared to changing your environment to improve your incentives. Instead of pulling yourself up by your bootstraps, step onto higher ground. For example, by creating and participating in a prediction market on a topic, you can induce yourself to become more honest on that topic. The more you can create personal direct costs of your dishonesty, the more honest you will become. And if you get paid to work on a certain topic, maybe you should give up on honesty about who if anyone should be paid to do that.

So my advice is to choose a focus for your honesty, a narrow enough focus to have a decent chance at achieving honesty. Make your focus more narrow the more dangerous is your focus area. Try to insulate beliefs on your focus topics from beliefs on risky topics like your own value, and try to arrange things so you will be penalized for dishonesty. Don’t persent yourself as a “rationalist” who is more honest on all topics, but instead as at best “rationalist on X.”

So, what is your X?

GD Star Rating
Tagged as: ,

Big Signals

Between $6 and $9 trillion dollars—about 8% of annual world-wide economic production—is currently being spent on projects that individually cost more than $1 billion. These mega-projects (including everything from buildings to transportation systems to digital infrastructure) represent the biggest investment boom in human history, and a lot of that money will be wasted. …

Over the course of the last fifteen years, [Flyvbjerg] has looked at hundreds of mega-projects, and he found that projects costing more than $1 billion almost always face massive cost overruns. Nine out of ten projects faces a cost overrun, with costs 50% higher than expected in real terms not unusual. …

In fact, the number of mega-projects completed successfully—on time, on budget, and with the promised benefits—is actually too small for Flyvbjerg to determine why they succeeded with any statistical validity. He estimates that only one in a thousand mega-projects fit that criteria. (more; paper)

You can probably throw most big firm mergers into this big inefficient project pot.

There’s a simple signaling explanation here. We like to do big things, as they make us seem big. We don’t want to be obvious about this motive, so we pretend to have financial calculations to justify them. But we are purposely sloppy about those calculations, so that we can justify the big projects we want.

It would be possible to make prediction markets that accurately told us on average that these financial calculations are systematically wrong. That could enable us to reject big projects that can’t be justified by reasonable calculations. But the people initiating these projects don’t want that, so it would have to be outsiders who set up these whistleblowing prediction markets. But alas as with most whistleblowers, the supply of these sort of whistleblowers is quite limited.

GD Star Rating
Tagged as: , ,

SciCast Pays Out Big!

When I announced SciCast in January, I said we couldn’t pay participants. Alas, many associated folks are skeptical of paying because they’ve heard that “extrinsic” motives just don’t work well relative to “intrinsic” motives. No need to pay folks since what really matters is if they feel involved. This view is quite widespread in academia and government.

But, SciCast will finally do a test:

SciCast is running a special! For four weeks, you can win prizes on some days of the week:
• On Wednesdays, win a badge for your profile.
• On Fridays, win a $25 Amazon Gift Card.
• On Tuesdays, win both a badge and a $25 Amazon Gift Card.
On each prize day 60 valid forecasts and comments made that day will be randomly selected to win (limit of $575 per person).
Be sure to use SciCast from May 26 to June 20!

Since we’ve averaged fewer than 60 of these activities per day, rewarding 60 random activities is huge! Either activity levels will stay the same and pretty much every action on those days will get a big reward, or we’ll get lots more activities on those days. Either you or science will win! :)

So if you or someone you know might be motivated by a relevant extrinsic or intrinsic reward, tell them about our SciCast special, and have them come be active on matching days of the week. We now have 473 questions on science and technology, and you can make conditional forecasts on most of them. Come!

Added 21May: SciCast is mentioned in this Nature article.

GD Star Rating
Tagged as: ,

Who/What Should Get Votes?

Alex T. asks Should the Future Get a Vote? He dislikes suggestions to give more votes to “civic organizations” who claim to represent future folks, since prediction markets could be more trustworthy:

Through a suitable choice of what is to be traded, prediction markets can be designed to be credibly motivated by a variety of goals including the interests of future generations. … If all we cared about was future GDP, a good rule would be to pass a policy if prediction markets estimate that future GDP will be higher with the policy than without the policy. Of course, we care about more than future GDP; perhaps we also care about environmental quality, risk, inequality, liberty and so forth. What Hanson’s futarchy proposes is to incorporate all these ideas into a weighted measure of welfare. … Note, however, that even this assumes that we know what people in the future will care about. Here then is the final meta-twist. We can also incorporate into our measure of welfare predictions of how future generations will define welfare. (more)

For example, we could implement a 2% discount rate by having official welfare be 2% times welfare this next year plus 98% times welfare however it will be defined a year from now. Applied recursively, this can let future folks keep changing their minds about what they care about, even future discount rates.

We could also give votes to people in the past. While one can’t change the experiences of past folks, one can still satisfy their preferences. If past folks expressed particular preferences regarding future outcomes, those preferences could also be given weight in an overall welfare definition.

We could even give votes to animals. One way is to make some assumptions about what outcomes animals seem to care about, pick ways to measure such outcomes, and then include weights on those measures in the welfare definition. Another way is to assume that eventually we’ll “uplift” such animals so that they can talk to us, and put weights on what those uplifted animals will eventually say about the outcomes their ancestors cared about.

We might even put weights on aliens, or on angels. We might just put a weight on what they say about what they want, if they ever show up to tell us. If they never show up, those weights stay set at zero.

Of course just because we could give votes to future folks, past folks, animals, aliens, and angels doesn’t mean we will ever want to do so.

GD Star Rating
Tagged as: , , ,

Michael Covel Interview

Investment advisor Michael Covel interviewed me on prediction markets for his podcast show here. I couldn’t be very encouraging about his main strategy of trend-following, but we covered many interesting issues.

GD Star Rating
Tagged as:

Fixing Academia Via Prediction Markets

When I first got into prediction markets twenty five years ago, I called them “idea futures”, and I focused on using them to reform how we deal with controversies in science and academia (see here, herehere, here). Lately I’ve focused on what I see as the much higher value application of advising decisions and reforming governance (see herehere, here, here). I’ve also talked a lot lately about what I see as the main social functions of academia (see here, here, here, here). Since prediction markets don’t much help to achieve these functions, I’m not optimistic about the demand for using prediction markets to reform academia.

But periodically people do consider using prediction markets to reform academia, as did Andrew Gelman a few months ago. And a few days ago Scott Alexander, who I once praised for his understanding of prediction markets, posted a utopian proposal for using prediction markets to reform academia. These discussions suggest that I revisit the issue of how one might use prediction markets to reform academia, if in fact enough people cared enough about gaining accurate academic beliefs. So let me start by summarizing and critiquing Alexander’s proposal.

Alexander proposes prediction markets where anyone can post any “theory” broadly conceived, like “grapes cure cancer.” (Key quotes below.) Winning payouts in such market suffer a roughly 10% tax to fund experiments to test their theories, and in addition some such markets are subsidized by science patron orgs like the NSF. Bettors in each market vote on representatives who then negotiate to pick someone to pay to test the bet-on theory. This tester, who must not have a strong position on the subject, publishes a detailed test design, at which point bettors could leave the market and avoid the test tax. “Everyone in the field” must make a public prediction on the test. Then the test is done, winners paid, and a new market set up for a new test of the same question. Somewhere along the line private hedge funds would also pay for academic work in order to learn where they should bet.

That was the summary; here are some critiques. First, people willing to bet on theories are not a good source of revenue to pay for research. There aren’t many of them and they should in general be subsidized not taxed. You’d have to legally prohibit other markets to bet on these without the tax, and even then you’d get few takers.

Second, Alexander says to subsidize markets the same way they’d be taxed, by adding money to the betting pot. But while this can work fine to cancel the penalty imposed by a tax, it does not offer an additional incentive to learn about the question. Any net subsidy could be taken by anyone who put money in the pot, regardless of their info efforts. As I’ve discussed often before, the right way to subsidize info efforts for a speculative market is to subsidize a market maker to have a low bid-ask spread.

Third, Alexander’s plan to have bettors vote to agree on a question tester seems quite unworkable to me. It would be expensive, rarely satisfy both sides, and seems easy to game by buying up bets just before the vote. More important, most interesting theories just don’t have very direct ways to test them, and most tests are of whole bundles of theories, not just one theory. Fourth, for most claim tests there is no obvious definition of “everyone in the field,” nor is it obvious that everyone should have opinion on those tests. Forcing a large group to all express a public opinion seems a huge cost with unclear benefits.

OK, now let me review my proposal, the result of twenty five years of thinking about this. The market maker subsidy is a very general and robust mechanism by which research patrons can pay for accurate info on specified questions, at least when answers to those questions will eventually be known. It allows patrons to vary subsidies by questions, answers, time, and conditions.

Of course this approach does require that such markets be legal, and it doesn’t do well at the main academic function of credentialing some folks as having the impressive academic-style mental features with which others like to associate. So only the customers of academia who mainly want accurate info would want to pay for this. And alas such customers seem rare today.

For research patrons using this market-maker subsidy mechanism, their main issues are about which questions to subsidize how much when. One issue is topic. For example, how much does particle physics matter relative to anthropology? This mostly seems to be a matter of patron taste, though if the issue were what topics should be researched to best promote economic growth, decision markets might be used to set priorities.

The biggest issue, I think, is abstraction vs. concreteness. At one extreme one can ask very specific questions like what will be the result of this very specific experiment or future empirical measurement. At the other extreme, one can ask very abstract questions like “do grapes cure cancer” or “is the universe infinite”.

Very specific questions offer bettors the most protection against corruption in the judging process. Bettors need worry less about how a very specific question will be interpreted. However, subsidies of specific questions also target specific researchers pretty directly for funding. For example, subsidizing bets on the results of a very specific experiment mainly subsidizes the people doing that experiment. Also, since the interest of research patrons in very specific questions mainly results from their interest in more general questions, patrons should prefer to directly target the more general questions directly of interest to them.

Fortunately, compared to other areas where one might apply prediction markets, academia offers especially high hopes for using abstract questions. This is because academia tends to house society’s most abstract conversations. That is, academia specializes in talking about abstract topics in ways that let answers be consistent and comparable across wide scopes of time, space, and discipline. This offers hope that one could often simply bet on the long term academic consensus on a question.

That is, one can plausibly just directly express a claim in direct and clear abstract language, and then bet on what the consensus will be on that claim in a century or two, if in fact there is any strong consensus on that claim then. Today we have a strong academic consensus on many claims that were hotly debated centuries ago. And we have good reasons to believe that this process of intellectual progress will continue long into the future.

Of course future consensus is hardly guaranteed. There are many past debates that we’d still find to hard to judge today. But for research patrons interested in creating accurate info, the lack of a future consensus would usually be a good sign that info efforts in that area less were valuable than in other areas. So by subsidizing markets that bet on future consensus conditional on such a consensus existing, patrons could more directly target their funding at topics where info will actually be found.

Large subsidies for market-makers on abstract questions would indirectly result in large subsidies on related specific questions. This is because some bettors would specialize in maintaining coherence relationships between the prices on abstract and specific questions. And this would create incentives for many specific efforts to collect info relevant to answering the many specific questions related to the fewer big abstract questions.

Yes, we’d  probably end up with some politics and corruption on who qualifies to judge later consensus on any given question – good judges should know the field of the question as well as a bit of history to help them understand what the question meant when it was created. But there’d probably be less politics and lobbying than if research patrons choose very specific questions to subsidize. And that would still probably be less politics than with today’s grant-based research funding.

Of course the real problem, the harder problem, is how to add mechanisms like this to academia in order to please the customers who want accuracy, while not detracting from or interfering too much with the other mechanisms that give the other customers of academia what they want. For example, should we subsidize high relevant prestige participants in the prediction markets, or tax those with low prestige?

Those promised quotes: Continue reading "Fixing Academia Via Prediction Markets" »

GD Star Rating
Tagged as: , ,

Spanish Interview on Prediction Markets

The spanish language Sintetia.com just posted an interview with me here. There’s also an English translation here, but at the moment it is pretty rough.

GD Star Rating
Tagged as:

GiveWell Interview

Alexander Berger from GiveWell interviewed me on prediction markets, and has posted his notes here. Alex and I seem to disagree about the importance of this topic:

Organizational obstacles  The main barrier to wider-scale adoption of prediction markets is that most organizations are reluctant to use them. It is unclear why this is the case. Those currently in power within firms may resist prediction markets because the markets would spread previously privileged information across the company and change perceptions of what is knowable and who knows

I tried to emphasize this topic, but Alex devotes only 60 out of 1800 words to it.

GD Star Rating
Tagged as: ,

Academic Stats Prediction Markets

In a column, Andrew Gelman and Eric Loken note that academia has a problem:

Unfortunately, statistics—and the scientific process more generally—often seems to be used more as a way of laundering uncertainty, processing data until researchers and consumers of research can feel safe acting as if various scientific hypotheses are unquestionably true.

They consider prediction markets as a solution, but largely reject them for reasons both bad and not so bad. I’ll respond here to their article in unusual detail. First the bad:

Would prediction markets (or something like them) help? It’s hard to imagine them working out in practice. Indeed, the housing crisis was magnified by rampant speculation in derivatives that led to a multiplier effect.

Yes, speculative market estimates were mistaken there, as were most other sources, and mistaken estimates caused bad decisions. But speculative markets were the first credible source to correct the mistake, and no other stable source had consistently more accurate estimates. Why should the most accurate source should be blamed for mistakes made by all sources?

Allowing people to bet on the failure of other people’s experiments just invites corruption, and the last thing social psychologists want to worry about is a point-shaving scandal.

What about letting researchers who compete for grants, jobs, and publications write critical referee reports and publish criticism, doesn’t that invite corruption too? If you are going to forbid all conflicts of interest because they invite corruption, you won’t have much left you will allow. Surely you need to argue that bet incentives are more corrupting that other incentives. Continue reading "Academic Stats Prediction Markets" »

GD Star Rating
Tagged as: ,

Fail Faster

It looks bad for a manager to have one of his projects fail. So to “cover his ass”, such a manager often tries to prevent any records showing that people saw failure coming. After a failure, he wants to say “this was just random bad luck; no one could have foreseen seen it.” His bosses up the chain of command tend to allow this, because they also want to avoid being held responsible for failures during their watch. So they also prefer the random back luck story.

Unfortunately, this approach tends to prevent organizations from getting signals that would let them mitigate failures, such as by quitting projects earlier. For example, most startup firms don’t fail until they have spent nearly all of the cash they were given. It is rare for a startup to admit it isn’t going to work out, and give some cash back to investors. Similarly, government agencies created to achieve some purpose rarely recommend to legislatures that they be eliminated when their find that they aren’t achieving their intended purposes.

Of course bosses don’t want to be too obvious about silencing possible signals of failure. They find it hard to silence what have become standard signals, like cost accounting measures.

A great application of prediction markets is to give better and clearer warnings of upcoming failure, to enable better mitigation, such as quitting. Of course project bosses anticipate this, and oppose prediction markets on their projects, for exactly this reason. But we can still hope that prediction market warnings may someday become a standard signal, and thus hard to silence:

I hope prediction markets within firms may someday gain a status like cost accounting today. In a world were no one else did cost accounting, proposing that your firm do it would basically suggest that someone was stealing there. Which would look bad. But in a world where everyone else does cost accounting, suggesting that your firm not do it would suggest that you want to steal from it. Which also looks bad.

Similarly, in a world where few other firms use prediction markets, suggesting that your firm use them on your project suggests that your project has an unusual problem in getting people to tell the truth about it via the usual channels. Which looks bad. But in a world where most firms use prediction markets on most projects, suggesting that your project not use prediction markets would suggest you want to hide something. (more)

GD Star Rating
Tagged as: ,