Prediction Markets Redux

For those very long-time readers this blog extensively covered prediction markets from 2006 to 2008. In a prediction market, you have a future event, such as the winner of an election, and a market that pays off one dollar if that event happens and nothing if it doesn't. The price of this market corresponds to a predicted probability that the event will happen. 

I worked with David Pennock and Yiling Chen to create an interactive map that colored every state based on how likely it was to go Democratic or Republican for both the presidential, gubernatorial and Senate races.

This all ended when Intrade, the source of the data that we used, went out of business after its CEO died climbing Mount Everest and may have embezzled money from the company.

In 2016 prediction markets went out of favor after badly predicting against Brexit once and Trump twice. 

But what's old becomes new again. Prediction markets are back with a bang, with Kalshi and Polymarket providing real money markets open to U.S. citizens, something we didn't legally have, except for very low stakes, twenty years ago.

Prediction markets play two distinct roles:

1. As a betting system so people can put their "money where their mouth is" based on their beliefs.
2. As an information aggregation system, to use the wisdom of the crowds to predict future events.

Sometimes these roles conflict with each other. With real money comes real greed and we've seen some recent issues of insider trading, most notably a US soldier who used classified knowledge about the then upcoming raid in Venezuela to net $400K on Polymarket. 

Insider trading will likely lead to better predictions, but those who have bet on that market might feel cheated. In the short run, insider trading might be a good thing, but in the long run, if we scare away participants because they're afraid others who have more information will take advantage of them, then we have fewer people in the market making it harder to draw upon the wisdom of the crowds. 

Another challenge for prediction markets is determining how the market pays off. You need to come up with a very specific definition of what it means for the market to pay off, but strange circumstances can lead to unexpected results. Back in 2006 North Korea launched a test missile outside the country but the market for that event paid off zero because the U.S. Department of Defense never verified that the missile actually left North Korean airspace. 

The market I like to see is factoring the RSA numbers by a given date. Very easy to verify. Some people might buy such a security, expecting quantum computing to be factoring numbers soon. And then I can make some money by betting against it. 

Prediction Markets Review

Yesterday the Electoral College delegates voted, 365 for Barack Obama and 173 for John McCain. How did the markets do?

To compare, here is my map the night before the election and the final results. The leaning category had Obama at 364. The markets leaned the wrong way for Missouri and Indiana, their 11 electoral votes canceling each other out. The extra vote for Obama came from a quirk in Nebraska that the Intrade markets didn't cover: Nebraska splits their votes based on congressional delegations, one of which went to Obama.

Indiana and Missouri were the most likely Republican and Democratic states to switch sides according to the markets, which mean the markets did very well this year again. Had every state leaned the right way (again), one would wonder if the probabilities in each state had any meaning beyond being above or below 50%.

Many argue the markets just followed the predictions based on polls like Nate Silver's fivethirtyeight.com. True to a point, Silver did amazingly well and the markets smartly trusted him. But the markets also did very well in 2004 without Silver. One can aggregate polls and other information using hours upon hours of analysis or one can just trust the markets to get essentially equally good results with little effort.

In some other prediction market news: A couple of years ago, Tradesports spun off their non-sports markets into a new site Intrade. Last month Tradesports shut down. Intrade still lives and hopefully it will continue to provide an exciting collection of political and other prediction markets.

Cantor Fitzgerald, owners of the Hollywood Stock Exchange, have filed with the CFTC to create real money markets based on box office receipts. Each share will pay off one-millionth of the first four weeks of the film's gross. They don't call it a prediction market since it doesn't have a binary outcome but I do because it is based on a specific time-limited event and the price should accurate reflect the expected gross of the movie.

Not market related but the big news in Illinois is of course the embarrassment of our governor right after the Obama high we've been on. At least Blagojevich will, hopefully, fade from view soon while Obama should leave a strong legacy as president that we in Illinois will remember for a long time.

Electronic Commerce and Prediction Markets

Registration has opened for the upcoming ACM Electronic Commerce Conference (which I am general chair) in Chicago July 10-12. The conference is immediately followed by AAAI and The Third World Congress of the Game Theory Society both also in the Chicago area. Before the EC conference is a series of workshops and tutorials covering topics from on-line advertising to social networks.

One of those workshops covers an area that has excited me for several years now, The Third Workshop on Prediction Markets. Prediction markets aggregate information quite efficiently in ways we don't yet fully understand and remains a fertile area of study. Legal limitations on betting have restricted the applications of prediction markets, particularly in the US, but that might change soon. The Commodity Futures Trading Commission (CFTC) is asking for public comment for regulations of prediction markets. Their concept release gives a nice discussion of the legal issues. Will this lead to more legitimate real money markets in the US? Time will tell.

More from Pennock and Masse.

Crowdsourcing the Truth

A new project Augur aims to create a decentralized prediction market. If this post so moves you, Augur is in the midst of a reputation sale. Don't miss out if you would like to be an Augur reporter.

A prediction market takes a future event, such as Hillary Clinton winning the 2016 Presidential Election, and creates a security that pays off $100 if Hillary wins and $0 otherwise. The market allow buying, selling and short selling the security and the price of the security represents that probability the event will happen. Predictwise, which aggregates prediction markets, has the probability of Hillary winning at 47% as I write this. But there are a limited amount of markets out there for Predictwise to draw from.

Intrade, which shut down due to financial improprieties in 2013, used to run markets on all aspects of elections and other current events. Many other prediction markets have disappeared over time. The Augur team put out a white paper describing their fully decentralized prediction market immune to individuals bringing it down. They build on cryptocurrencies for buying and selling and a group of "reporters" financially incentivized to announce the correct answer for each market.

It's that last part I find the most interesting, instead of having an authority that reports the results, Augur will crowdsource the truth.

A key feature of Augur is tradeable Reputation. The total amount of Reputation is a fixed quantity, determined upon the launch of Augur. Holding Reputation entitles its owner to report on the outcomes of events, after the events occur...Reputation tokens are gained and lost depending on how reliably their owner votes with the consensus.

Consensus may not be "the truth". Reporters are not incentivized to report the truth but what the other reporters will report as the consensus. Those who buy and sell on the market are not betting on the truth but the outcome as it is decided by the consensus of reporters. We have an ungrounded market.

The purchasers of reputation tokens likely won't represent the public at large and biases may come to play. Would this consensus have agreed that Bush won the 2000 election?

What would have the consensus done on the North Korea controversy?

[The Intrade security] allowed speculation on whether North Korea would, by 31 July 2006, successfully fire ballistic missiles that would land outside its airspace. On 5 July 2006 the North Korean government claimed a successful test launch that would have satisfied the prediction, a launch widely reported by world media. [Intrade] declared that the contract's conditions had not been met, because the US Department of Defense had not confirmed the action, and this confirmation was specifically required by the contract. (Other government sources had confirmed the claim, but these were not the sources referenced in the contract.) Traders considered this to be in strict compliance with the stated rule but contrary to the intention of the market (which was to predict the launch event, and not whether the US Defense Department would confirm it).

Since Augur will allow arbitrarily worded contracts on topics such as proofs of P v NP, the consensus reporting may lead to some interesting future blog posts.

Despite my reservations, I'm quite excited to see Augur set up a prediction market system that can't be shut down and wish them all the luck. You have to until October 1 to buy reputation if you want to be deciding the truth instead of just predicting it.

Predictions Markets Map of Senate Races

David Pennock, Yiling Chen and I created a Prediction Markets Map of the 2006 US Senates Elections. When you load the map it downloads the latest prices from Tradesports and colors each state appropriately, a mix of red (Republican), blue (democrats) and green (other candidates). We created the map because predictions based on market prices tend to do better than experts or polls. As Wisdom of Crowds author James Surowiecki wrote recently in the New Yorker

In the past few years, a host of prediction markets, as they're usually called, have appeared online, offering people the chance to speculate on subjects ranging from the box-office performance of Hollywood films to the outcome of Presidential elections and the spread of bird flu. These markets' forecasts have proved remarkably accurate—just as bettors collectively do an exceptionally good job of predicting sports results. (In 2004, for instance, Tradesports, a Dublin-based prediction market, called thirty-three out of thirty-four races in the Senate correctly, and called all fifty states correctly in the results for the electoral college.)

Scaling colors according to probabilities is a surprisingly hard problem. It relates to human perception, we want our eyes to distinguish colors related to probabilities that are not that close. For the map I used a transformation based on the XYZ color space that seems to give reasonable though far from perfect results. Even small projects like this can really bring up complex issues from outside the theory world.

Electoral Markets Map

Two years ago we created maps to show state-by-state how the 2006 US Senate and Gubernatorial races were shaping up, based on security prices at Tradesports. These markets did very well in predicting the outcomes of those races.

Now in a presidential election year, we have recreated a map, this time for the electoral college and gave it its own URL electoralmarkets.com. This map takes its prices from Intrade, Tradesports current site for their non-sports securities. Once again darker blue indicates a more likely vote for the Democratic candidate (Obama) and darker red more likely for the Republican McCain. Roll over a state for more information or click on that state to go to the Intrade site for that security with historical information.

We created these maps to promote prediction markets as a useful tool for information aggregation. The Third Workshop on Prediction Markets was held at EC yesterday.

As the summer goes on we plan to add several new features so keep checking back and watch the election play out in probabilities in real time.

The Big Aggregators

Friday morning I wanted to know where the rumors were pointing to for McCain's running mate selection. I could have searched various political blogs, but instead I went to Intrade and checked out the current prices on VP candidates. Since Intrade has constant trading, these prices do aggregate the various rumors and their veracity. Sarah Palin was running at about 60%. Apparantly I was not the only one with this idea has Intrade had major performance problems on Friday.

After seeing the price for Palin, I had a question many other Americans were asking: Who is Sarah Palin? So I went to that other great aggregator Wikipedia and read up on her. The scariest part: For the first time, someone on a major party ticket is younger than me.

The wisdom of crowds boiled down to a number on a trading site and a constantly updated page with much more than I need to know. The rest of the Internet is just commentary.

Some graphs of Intrade's market on the Palin market lifetime and in the last day.

Two things to note: The markets didn't predict Palin until close to the end. Also big volatility in the closing hours. Market aggregate public information—they don't predict what isn't out there. In the last hours, even little rumors, accurate or inaccurate, can drive prices as some people try to make a fast buck.

Chris Maase's blog as always has much more Palin and all other things about prediction markets. Also check out the new Intrade.net which now has do it yourself prediction markets.

I created a simple widget for our Electoral Markets Map. You can see in on the left sidebar until the election and always have the most up to date account of who's ahead state by state.

A Failure of Social Networks?

One of Northwestern's professors made quite a splash a month ago with his group's predictions on the spread of the Swine Flu (now called H1N1). Dirk Brockmann claimed a worst-case scenario of 1700 flu cases by the end of May. Another group in Indiana made similar predictions. The research got picked up on a front page New York Times article. Brockmann's group used data from Where's George, a site that tracks movement of dollar bills, to understand interactions that would lead to the spreading of disease.

So how did these groups do? Not even in the right ballpark, off by two orders of magnitude. As a follow-up article in Tuesday's Science Times the CDC estimated there were well over 100,000 cases by the end of last month. What went wrong? In short Brockmann claims faulty initial data. Nevertheless I always worry that bad predictions from scientists make it harder to have the public trust us when we really need them to.

Might this be an instance where prediction markets greatly out-performed the experts? In short, no. There were relevant markets but two big problems: 

• No one thought to create a market for the number of flu cases over a couple of thousand. 
• Prediction markets require a verifiable outcome so they were based on CDC confirmed cases. But after the flu turned out not to be that dangerous, the CDC stopped confirming most cases and there were less than 7500 confirmed cases by the end of May.

Olympic Markets

What do the Olympics have to do with computational complexity? Not much, so while I have spent much more time watching the games than proving theorems this week I couldn't think of the right way to fit it into the blog. Until I got the following email from David Pennock yesterday.

We implemented Olympics medal count prediction on Yoopick. Since it was your idea, you now have a moral obligation to blog about it, use it, and evangelize it to all your friends. :-)

Where most prediction markets track binary events, like whether the Obama will win the election, the Facebook-plugin Yoopick, developed at Yahoo Research in New York, is a fake-money market that predicts distributions over a range like the number of points scored in a basketball game. I suggested using Yoopick to predict the distribution of medals won by county and now you too can make your predictions and win some Yootles. I hear 100 Yootles and 2 Dollars will get you a subway ride in New York.

In other Prediction Market stuff at Yahoo, Sharad Goel used Amazon's Mechanical Turk to offer 100 people three cents each to predict the probability that Obama will win. The predictions were all over the place but average them up and you get exactly the same value as Intrade. Not the first time we've seen this phenomenon and it seems hard to explain.

And don't forget to check our our Electoral Markets Map. Obama has the slight edge as we get closer to the conventions and start of the real campaign season.

Information Markets and Quantum Computing

The DIMACS Workshop on Information Markets drew a neat crowd, a mix of computer scientists, economists (mostly experimental) and members of industry from companies as big as Microsoft and so small they are run by individuals in their spare time. Information markets create securities that can aggregate people's beliefs and help predict the likelihood of future events.

This workshop reminded me of the early conferences in quantum computing a decade ago. Quantum computing back then had some promising research (factoring algorithms for example) and no one was sure whether it would lead to whole new computing paradigm or just disappear into the ether. Information markets are also a new technology with some promising research (mostly analytic and experimental) and no one knows whether it will revolutionize the way everyone does prediction, information aggregation and decision making or just slowly disappear.

Information markets face different challenges than quantum computing. The technology already exists to run efficient markets and better tools are being developed. However the field needs to convince business and governmental leaders of the value and accuracy of the markets, overcome the stigma from the terrorism futures scare and find ways to overcome the legal issues relating to gambling in running real money markets.

Quantum computing needs to deal merely with the laws of physics but information markets need to deal with the laws of the United States of America.

Update 2/9: Ken Kittlitz's report on the workshop. (Thanks to Chris Masse)

Do not become obsessed with the Polls unless...

I know someone who checks the polls 3 times a day to see who looks like they will be elected prez. She cares A LOT about the election. It is irrelevant to this post who she supports. 

I've asked her `if you find out that, say, Biden is up  8 points instead of 9 in Penn. or that Georgia is looking pretty safe for Trump, or that Texas is in play (really?) how will that change your life? What will YOU do differently?'

She had no answer. Unfortunately she is still a poll-watcher (I know that means something else usually, but you know what I mean.)

So who should be poll-watching or poll-obsessing?

1) To be fair to my friend, she might decide to GIVE MORE to her candidate if the polls are saying that he will lose (whoops- by saying `he' I gave away that her candidate is NOT Jo Jorgenson- Libertarian).  I doubt my friend could say to the campaign `I want to you to spend it in state X since I read that its close there' (I read that some big donors in 2016 demanded more say in where the money was spend. I doubt that's a good idea since I suspect the party knows more about how to best spend the money then the donor does.) 

2) The Biden and the Trump Campaigns SHOULD be poll-watching to decide where to put their efforts. And I suspect they are doing just that.

3) A really big donor (my friend is not one of those) MIGHT want to poll watch to decide if the candidate they want needs money. (I wonder if EITHER candidate needs money since they get so much free media.)

4) Nate Silver-being a poll-watcher is kind-of his job. And of course writing columns about them and making predictions based on what he sees. My friend is not Nate Silver. 

5) Other people who have Nate Silver's job. I can't name any- is Nate Silver the most famous... Gee, not sure what job title he has... SO this is now two questions: What is his job title, call it X, and is he the most famous person who does X?

SO- my point- DO NOT be a poll-obsessive unless the information you get will lead to an action you can take. And I suspect that mostly it does not. 

The primaries are different: If a poll says A can beat X but B cannot beat X, that might guide who you vote for. 

Misc thought: 

 I've heard the phrase `democratic pollster' and `republican pollster' These terms do not make sense. Would I call myself a `democratic Muffin Mathematician' ? My political leanings do not affect my search for truth about mathematical Muffins. Similarly, one would think that a pollster wants to find the TRUTH, even if its bad news for their employer, ESPECIALLY if its bad news for their employer, so they can help their employer fix it. The phrase `pollster employed by the X party' would make more sense-- however, whenever they are on TV they seem to always say that their candidate is doing well, even when they are not. 

ADDED LATER: Lance had a great tweet about this post: do not obsess about polls, but DO obsess bout prediction markets. I think in the past prediction markets have been better predictors but some group-think has set in so its no longer clear. (I could be wrong- but thats why I have heard.) 

Expectation of Expectations

Can we quantitatively judge the effect of a debate on the election? One reasonable approach would look at prediction markets. If the value of the security for Obama to win would greatly increase after the debate in the absense of significant other factors, one could conclude that the debate went well for Obama, or at least that he did better than expected.

Intrade, the Irish real-money prediction markets site, went one further and set up a new security 2ND.DEBATE.OBAMA based on the outcome of tonight's presidential debate. In short,

This contract will be settled according to which Presidential Candidate receives the largest increase in value following the Presidential Debate on Tuesday October 7th. The pre-debate value will be calculated for 2008.PRES.OBAMA and 2008.PRES.McCAIN and then compared with the post-debate value for each of these contracts. This contract will expire at 100 if the value of 2008.PRES.OBAMA increases more than the value of 2008.PRES.McCAIN. This contract will expire at 0 if this is not the case.

A similar security for the VP debate paid off zero as there was a slight bump for McCain suggesting Palin did better than expected.

But there is a flaw in the logic for such a security—The market should already take into account the expectation of the debate. In general for any random variable A, the expectation of the expectation of A is the same as the expectation of A, i.e., E(E(A))=E(A).

So, assuming that these securities represent real probabilities, should the value of 2ND.DEBATE.OBAMA before the debate be 50? Not quite. Suppose that there is a 75% chance that the Obama stock goes up 5 and a 25% chance it drops 15. Then the price of 2ND.DEBATE.OBAMA should be 75%. The price of the security indicates who has the least chance of major downside from the debate. But to think one could use this security to predict how well the debate will go makes little sense.

The press release suggests using this security for real-time debate tracking. The price of 2ND.DEBATE.OBAMA should gyrate dramatically on small changes of 2008.PRES.OBAMA. But one could just derive this directly from 2008.PRES.OBAMA and since the later security has much greater liquidity it will likely give better information.

Chris Masse suggested another explanation.

[Intrade] is trying to induce traders into trading more (so as to get more transaction fees), and this new contract offers more upsides for the traders who will bet right. One could double one's initial investment in a matter of 2 days.

In which case all the power to them.

Complexity versus Complexity

For those interested, I've started writing posts for the Predictwise Blog. Predictwise makes predictions of future events such as who will win the Republican Nomination (currently Trump with an 80% probability) based on prediction markets and other betting sites. This has been a fascinating election in terms of predictions, strategies, rules and game theory and I'm happy to try and makes sense out of it over at Predictwise without subjecting my readers here at Computational Complexity with too many political posts.

A reader had asked me to comment on a Slate article The Theory of Everything and Then Some, a book review of John Miller's A Crude Look at the Whole: The Science of Complex Systems in Business, Life, and Society. John Miller is a social scientist who works on the other "complexity theory" that studies that "simple local rules can have complex global implications". Often complex systems work quite well, like the invisible hand of the economy, but sometimes things can go wrong and the article often mentions the "flash crash" of trading programs reacting to each other causing a major drop in stock prices in May of 2010.

Our fields with the similar names are not as different as might appear. Much of what they study are inherently computational-like processes and we also look at emergent behavior from simple operations of Turing machine; read, write and move the tape. What they call non-linear we call computation. We do take very different approaches. The computational complexity theory community proves theorems where we can and helps understand the mathematical challenges of when we can't. The other complexity theorists try to explain by examples, simulations and simplified models.

The two communities often, but not always, seem to have disdain for one another and that's a shame. The tools of computational complexity can help understand the power and limitations of complex systems. These collaborations require them to understand how we can help them and for us to be willing to work on problems that may not yield difficult-to-prove theorems. That's what attracts me to prediction markets, a very simple kind of information aggregation system that still is very difficult to analyze as a computational mechanism.

What's missing from the article is how tools like machine learning can play in helping to predict the outcomes of many complex systems. The big deluge of data that starts off the article may add to the complexity but it almost paradoxically also makes it possible to learn from it.

Quickies

David Pennock gives a CS view of prediction markets and gambling on his new weblog Oddhead. For those interested in prediction markets also take a look at the group blog Midas Oracle.

Chris Leonard, former editor of the Elsevier theory journals, returns to academic publishing for the open-access publisher BioMed Central to help them expand into physics, math and CS. He writes about changes in scientific publishing (and other topics) in his weblog.

Should teachers try to make math interesting and relevant? No, according to Tom Loveless.

And I'd be remiss not to mention the lengthy New York Times profile on Shing-Tung Yau.

Goodbye Old Friends

We had two terrible losses this week. Not people but websites, Intrade and Google Reader.

The corporate auditors for the real-money Irish prediction markets site Intrade found improprieties with payments to the late founder John Delaney and have effectively shut down the site, probably for good. I never bet money on Intrade but they made their bet data easily available. I used Intrade data to power my electoral markets map, possibly the most accurate predictor of elections, at least until Nate Silver came along. Even then our map updated in real time based on new information reflected in market prices, while Nate had to wait for poll data. I also used Intrade generated graphs of prediction market prices in dozens of talks I've given over the past decade. There are other sites for prediction aggregation but painful to lose the granddaddy of them all.

Google announced it is closing down Google Reader, Google's newsreader on July 1. I find Reader invaluable for keeping up with the theory blogs, especially those that don't post that often (I'm looking at you Scott and Luca) as well as a various collection of other blogs and my daily Dilbert. Not to mention the 2241 people who subscribe to this blog on Google Reader. There are alternatives (I'm trying Feedly) and I will be more vigilant on sharing posts on Twitter and Google+ but the real question is to why do fewer people use Google Reader. Is the flood of stuff on the Internet gotten so large we don't even try to keep up with it?

Prediction Map Post Mortem

As I've mentioned before, David Pennock, Yiling Chen and I developed a map predicting the 2006 Senate races based on market prices from tradesports.com. So how did those predictions go? In short you can say the markets predicted every individual race correctly but got the senate wrong, but let us look a little more carefully.

At about 9 AM CST on the morning of election day I made a snap shot of the map for a Discovery Channel Website article.

Every state colored blue was won by a democrat and every state colored red went to a republican. But also note the 69% given to GOP (Republican) Senate control although this election will give control to the democrats. No outcome would have made all the states and senate control agree with the 9 AM map.

Were the markets inconsistent? No, because the markets predict not absolutely but probabilistically. For example, the markets gave a probability of winning 60% for each of Virginia and Missouri and the democrats needed both to take the senate. If these races were independent events, the probability that the democrats take both is 36% or a 64% chance of GOP senate control assuming no other surprises.

Of course the races were not independent events and there are other states involved making it more difficult to compare the probabilities of the individual races with that of senate control.

So how did the markets do as predictors? Quite well as the outcome seems quite reasonable given the markets. Other outcomes would have also been reasonable such as the Democrats losing Virginia and the senate remaining in republican hands, a possibility that came very close to happening.

We plan a map with a better design and more features for the 2008 Electoral College and Senate races. Stay tuned.

Markets and Polls

A couple of weeks ago a strange thing happened on our electoral markets map. California turned red for a couple of hours. A few days later Michigan turned red as well. Neither of these states are about to vote republican, rather there were odd trades of Obama at a very low price in California and McCain at a high price in Michigan. Since we used the closing price the states were colored wrong until another trade occured.

So we adjusted our algorithm so that if the closing price is lower than the bid price (the price someone is willing to pay) we use the bid price instead. Also if the closing price is higher than the ask price (the price someone is willing to sell) we use the ask price. That seems to avoid the problems of rouge trades.

Which brings me to a controversial post at fivethirtyeight.com, a site that makes predictions based on poll numbers. Nate Silver notes that the prediction of Obama of the markets to win the presidency (about 58%) is much lower than his site's prediction (about 72%). Silver suggests the disparity is because of a rouge trader or traders that seem to be driving the price down and even buying Hillary stock.

I don't buy it. There is quite a bit of volume on these securities and the prices on Obama should bounce back quickly and in fact it does. The rouge trader cannot explain a difference of 14 points. The Hillary factor can be explained by the long-shot bias (people overweigh low probability events). The markets suggest that the probability that Hillary is president is about the same as McCain winning Illinois which sounds about right (even if they are both too high at around 4%).

I have a different theory: The race is tighter than the Silver analysis suggests. The polls can give you numbers about each state but not the correlation between them. Silver gives an explanation of how he handles the correlations in his simulations:

Our simulation accounts for this tendency by applying a similarity matrix, which evaluates the demographic relationships between different states by of a nearest-neighbor analysis as described here. Our process recognizes, for instance, that as the polling in Ohio moves, the polling in a similar state like Michigan is liable to move in the same direction. On the other hand, there may be little relationship between the polling movement in Ohio and that in a dissimilar state like New Mexico.

Silver admits he has little data to back up this claim. I believe the correlations are much tighter—that most of the states might move in the same direction depending on some future event or ad or gaffe or performance on the yet-to-be-held debates, that there is high future correlation even between Ohio and New Mexico.

The swing states are typically highly correlated and if the four states running about 50% (Nevada, Ohio, Virginia and New Hampshire) all go to McCain then the Electoral college ties and it would just take one other state (like New Mexico) to push it over.

The analysis of each state can be done well with polls and historical data and the market prices and polls for the individual states do not differ that much. But understanding the correlations between states is more guesswork and I trust the wisdom of the crowds over the wisdom of the one.

Predicting in Changing Environments

The New York Times yesterday ran a story connecting climate change to the Louisiana flooding.

The National Weather Service reports that parts of Louisiana have received as much as 31 inches of rain in the last week, a number Dr. Easterling called “pretty staggering,” and one that exceeds an amount of precipitation that his center predicts will occur once every thousand years in the area.
Dr. Easterling said that those sorts of estimates were predicated on the idea that the climate was stable, a principle that has become outdated.
The third National Climate Assessment, released in 2014 by the United States Global Change Research Program, showed that “the amount of rain falling in very heavy precipitation events” had been significantly above average since 1991.
However, the research did not identify the South as one of the areas of greatest concern; the increase was found to be greatest in the Northeast, Midwest and Upper Great Plains regions of the United States.

In short climate change means our old prediction models of the weather no longer apply. On top of that, new models that tried to take into account climate change predicted heavier rains but not in that area of the country.

The weather is hardly the only predictions gone bad this year. From Nate Cohn's What I Got Wrong About Donald Trump

Did he have a 1 percent chance to win when he descended the escalator of Trump Tower last June? Twenty percent? Or should we have known all along?
Was Mr. Trump’s [republican nomination] victory a black swan, the electoral equivalent of World War I or the Depression: an unlikely event with complex causes, some understood at the time but others overlooked, that came together in unexpected ways to produce a result that no one could have reasonably anticipated?
Or did we simply underestimate Mr. Trump from the start? Did we discount him because we assumed that voters would never nominate a reality-TV star for president, let alone a provocateur with iconoclastic policy views like his? Did we put too much stock in “the party decides,” a theory about the role of party elites in influencing the outcome of the primary process?
The answer, as best I can tell, is all of the above.
I do think we — and specifically, I — underestimated Mr. Trump. There were bad assumptions, misinterpretations of the data, and missed connections all along the way.

We also had bad predictions on Brexit and one factor in the 2008 financial crisis was a heavy reliance on historical patterns of housing prices.

We have at our fingertips incredible prediction tools from machine learning models to prediction markets. Not all things change, our models trained to recognize cat pictures will continue to recognize cat pictures for a long time running. But as we continue to rely more and more on data driven predictions and decisions, be prepared for more and more surprises as underlying changes in the environmental, political and financial climates can pull the rug out from under us.

Robin Hanson's Ems

Robin Hanson an economist at George Mason and author of the Overcoming Bias blog, has a new book The Age of Em: Work, Love and Life when Robots Rule the Earth. Em stands for brain emulation, a computerized version of a human brain processes including consciousness, all the wants, desires and faults of a human brain. An em can be created by copying from a human brain or another em, it can be stored and restored, slightly tweaked and can run faster or slower depending on the power consumption. Reminds me a bit of the cookies in the White Christmas episode of Black Mirror. Hanson also talks about clans, the collection of all the ems that descend from a particular human.

The book has two distinct parts. The first gives a plausible physical and social explanation as to how and why the em world will come to be. Hanson gives the odds of such a world developing at one in a thousand. I put the odds much lower, especially since we have no true understanding of consciousness. For example if consciousness requires quantum entanglement, we would have no hope of copying into an em without destroying the old em (or human) it came from. More likely I expect we would have em-like machines that can perform a number of human-like tasks but won't have any true consciousness. Nevertheless I'd acknowledge that Hanson's world is at least possible given our current knowledge of the brain.

I enjoyed more Hanson's discussion about the world of the ems given that they exist. Hanson takes a science, as opposed to a science fiction, approach to the topic, carefully thinking about how ems would work as a clan, how they interact politically, socially and economically. You can see the variety of topics in the table of contents on the book's website. For example, Hanson argues that it makes economic sense to have ems split off as "spurs" to do more menial tasks in parallel in exchange for a short working life, as short as a few minutes, and a long retirement, with the retirement in a slower low-powered mode. Hanson has done some strong research and advocating for prediction markets (we even have a joint paper on the topic) that there is no surprise that markets show up as a decision making systems for ems.

I don't agree with all of Hanson's conclusions, in particular he expects a certain rationality from ems that we don't often see in humans, and if ems are just human emulations, they may not want a short life and long retirement. Perhaps this book isn't about ems and robots at all, but about Hanson's vision of human-like creatures as true economic beings as he espouses in his blog. Not sure it is a world I'd like to be a part of, but it's a fascinating world nevertheless.

A Predictions Markets Mess

The prediction market exchange Tradesports have themselves a little controversy over North Korean missiles.

I have written about prediction (or information markets) before. Consider some future binary event like whether Hillary Clinton will be the democratic nominee for president in 2006. One can create a security 2008DEM.NOM.CLINTON that pays off $100 if Clinton wins the nomination and $0 if she doesn't. Then allow trading on the security including selling short. The price of the security will correspond to the probability that the event will occur, and studies have shown that these probabilities predict better, in general, than experts and polls. Tradesports has such a security on Clinton and the price as I write this for 2008DEM.NOM.CLINTON is 41.9 indicating Clinton has a 41.9% chance of winning the nomination.

Tradesports had another security N.KOREA.MISSILE.31JUL that would pay off if North Korea launces a test missle that leaves North Korean air space by July 31. As you might remember, North Korea did fire test missles on the fourth of July. So it seems like the security N.KOREA.MISSILE.31JUL should have paid off at $100.

Here's the rub. The fine details of the contract required that the US Department of Defense verify that the missiles left North Korean air space. Tradesports couldn't get the verification so they expired the security at $0.

Those who predicted the North Korean missile launch lost real money on a technicality which risks the accuracy of these markets. They no longer predict whether a launch occurs, just whether the DoD would acknowledge it.

More on the Freakonomics Blog.