How to see into the future? | Hacker News Reader

How to see into the future?

ft.com

67 pointsdanboarder32 comments

An excerpt of key points on forecasting from end of the article:

"How to be a superforecaster:

Some participants in the Good Judgment Project were given advice on how to transform their knowledge about the world into a probabilistic forecast – and this training, while brief, led to a sharp improvement in forecasting performance. The advice, a few pages in total, was summarised with the acronym CHAMP:

● Comparisons are important: use relevant comparisons as a starting point;

● Historical trends can help: look at history unless you have a strong reason to expect change;

● Average opinions: experts disagree, so find out what they think and pick a midpoint;

● Mathematical models: when model-based predictions are available, you should take them into account;

● Predictable biases exist and can be allowed for. Don’t let your hopes influence your forecasts, for example; don’t stubbornly cling to old forecasts in the face of news."

As a participant in the Good Judgement Project I recall this training (basically an interactive Powerpoint). I would agree that it was helpful in my forecasting performance last year.

It's an interesting article, but those conclusions seem so weak to me; little more than formalised common sense. Maybe the training you had did a better job of making them tangible.

I think you nailed it, it's formalized common sense (the article hints at this conclusion too). However, the unique addition of a way to track and improve forecasting performance over time is what makes this project work. This was accomplished through a few attempts at custom software in the last few years, and for this year they are using Inkling Markets, so far it is best platform I've experienced for tracking and improving prediction performance.

This whole concept reminds me very much of a character in an Iain M Banks novel (I forget which) whereby this character had--historically--shown an incredible knack for predicting/analysing future events, but there was always doubt whether they were just a statistical anomaly (given enough people, someone has to get it right all the time--until, suddenly, they don't) or a real effect. Interesting idea. Presumably they try and control for this.

> Presumably they try and control for this.

How could they? They're using a retrospective design, which means they wait until someone's past forecast turns out to be correct, then they tally that as a data point in favor of their thesis. In actual science, one must use a prospective design, to avoid the perversities and statistical pitfalls of a retrospective one, and one must craft and then test a falsifiable theory -- an explanation -- about the forecasting ability.

In a prospective design, one being run by people with even a tiny bit of common sense, on the first day of the study the experimenters would say, "Okay, if there were really an ability to forecast the future, anyone possessing the gift would be busy making a billion dollars in equities, and coincidentally they would refuse to talk to us."

In a retrospective study, the falsifiability criterion on which science depends is easy to circumvent -- all one needs to do is tally successful predictions and discard failed ones. The originally huge experimental group will keep getting smaller, but unless everyone has made a bad prediction, the theory still has merit.

Locating an astonishing record of predictions is trivial -- for a group of X participants, after Y binary predictions, there will likely remain X / 2^Y subjects with perfect prediction records. That means for a million participants making binary predictions (the stock market will rise/fall in the coming month), after a run of 16 predictions, 15 subjects are very likely to remain standing with perfect records.

How the designers of this bogus study don't understand this basic statistical fact is beyond me.

It's pseudoscience. If it were not pseudoscience, someone would become richer than Bill Gates or Carlos Slim by making equities calls from his penthouse overlooking the drab world occupied by those who understand logic.

You are completely wrong. I don't know why everyone seems to think the researchers are incompetent. From [here](http://www.economist.com/news/21589145-how-sort-best-rest-wh...):

> The big surprise has been the support for the unabashedly elitist “super-forecaster” hypothesis. The top 2% of forecasters in Year 1 showed that there is more than luck at play. If it were just luck, the “supers” would regress to the mean: yesterday’s champs would be today’s chumps. But they actually got better. When we randomly assigned “supers” into elite teams, they blew the lid off IARPA’s performance goals. They beat the unweighted average (wisdom-of-overall-crowd) by 65%; beat the best algorithms of four competitor institutions by 35-60%; and beat two prediction markets by 20-35%

> You are completely wrong.

Citation needed. There has never been a study of forecasting ability that has stood the test of time -- ever, anywhere.

> I don't know why everyone seems to think the researchers are incompetent.

The null hypothesis, the gold standard of statistical science, offers this as the default explanation, and the researchers have the burden of evidence to contradict it. They will fail, indeed they don't seem to realize they have the burden of evidence.

Have you even considered that the positive results arise from chance? What is the p-factor for their result? Do you know what I am referring to?

Your quotation doesn't have a single reliable word in it, or any content recognizable as science. If it were a scientific paper, it would contain the standard phrase, "Here is our p-factor, the statistical assessment that our result arises from chance."

But it's not science, it's public relations. The thesis is assumed to be correct until incontrovertible evidence proves it false -- the opposite of the scientific outlook, which assumes an idea to be false until incontrovertible evidence supports it.

I gave a citation. The quote is more than enough evidence to disprove your naive "the researchers didn't bother to test for regression to mean" hypothesis. They didn't give a p-value, but the reported effect size is quite huge to be chance, and the study included thousands of people.

> I gave a citation.

You clearly don't understand the meaning of "citation" among scientists. You posted a quote from a press release that didn't even pretend to be a scientific assessment of the work.

> The quote is more than enough evidence

Press releases aren't evidence. Were this not the case, Bigfoot would come into existence on the strength of his press coverage.

> They didn't give a p-value

So, which is it? "The quote is more than enough evidence", or "They didn't give a p-value"? They cannot both be true.

> ... but the reported effect size is quite huge to be chance, and the study included thousands of people.

I can see you have no idea how science works.

> ... to disprove your naive "the researchers didn't bother to test for regression to mean"

You are also not above inventing positions for other people.

I never said anything about "science". Only that your accusations are garbage.

It's one thing to say "let's not take this too seriously because they haven't published anything yet", which I don't even know is true, but whatever. However you went above and beyond that:

* claiming their methodology was "retrospective design"

* claiming they picked statistical outliers and didn't bother to test if they continued to make good predictions: "The described program collects forecasts, forecasts that can be expected to be half right and half wrong for binary questions, throws out those that by chance were wrong, and preserves those that by chance were right."

* various insults: they don't understand logic, statistics, it's pseudoscience, etc.

* claim they twisted language to make failed forecasts look successful.

* claim they didn't do any statistical analysis

The events being tested are not up to chance. There is lots of evidence about which way a situation will go. The test is not predicting a coin toss, but interpreting information about a complex situation.

TL;DR - some things are hard to predict. If you still want to have a go, get some other people's opinions too. Oh, and stubbornness is a vice unsuited to the forecaster.

Being actively open-minded. I wonder why some people are like that and others not. I've met people like that from different cultures and backgrounds and I've similarly met their opposites. It doesn't seem like cultural and educational background is a factor. Even the article contrasts two such personalities with similar backgrounds but that's just good storytelling.

I wonder if it is possible to train yourself to be open-minded, or even to spot if you aren't open-minded. After all, while it is easy to spot open or closed-mindedness in others, I expect that almost everyone thinks that they are open-minded.

If you're close-minded, you're not going to be conscious of it - that's the nature of the beast.

You can't force someone's mind open, as all you're doing then is teaching them to be close-minded with a different set of axioms that you've behooved upon them.

As to yourself - it's hard to say, as having any objective view of self is challenging to say the least.

I like to think I'm extremely open-minded, but I might be wrong.

I'm close-minded, emotionally, I'm wired to enjoy knowing in advance (anxious personality). Until I realize that I was very wrong and could enjoy discovering things that were oblivious to me. I'm still close-minded but headed in the opposite direction now.

Research seems to suggest that psilocybin (the drug that makes magic mushrooms "magic") permanently increases open-mindedness [1][2], so that's one "easy" way to do it.

There's also a lot you can do consciously in daily life. By deciding to be more open-minded through constantly working to really understand why the people you disagree with (no matter if it's serious topics like religion, politics, morals, ideology or more trivial topics like tastes and interests) think the way they do it's almost impossible to not end up more open-minded.

[1] http://www.psychologytoday.com/blog/beautiful-minds/201110/p...

[2] http://www.theverge.com/2014/7/3/5869465/scientists-figured-...

I think cultural and educational background is a factor, but not in the fashion of "persons who had an education of X type in culture Y have superior prescience", but rather "persons who had education experiences of X, Y, Z, and developed in cultures A, B, C, D, Q, R have superior prescience", as having a broader gamut of experiences to draw from allows you to eliminate many biases from which you'd otherwise suffer.

Nobody seems to talk about the advantages of being stubborn and close-minded. The sad truth is we live in an environment where being right is less advantageous than cheerleading those in power. When the right answer contradicts group think or whatever your biggest bully (boss, religious leader, local warlord) is saying, you will be punished. Hopefully it will only be shaming as "not a team player" or "cynical." God forbid you are actually right - then you could be in real trouble.

If people are rewarded for open-minded and accurate thinking then they will start doing it. I can't see that ever happening.

Quote: "But ultimately one might expect the same basic finding as always: that forecasting events is basically impossible. Wrong. To connoisseurs of the frailties of futurology, the results of the Good Judgment Project are quite astonishing. Forecasting is possible, and some people – call them “superforecasters”– can predict geopolitical events with an accuracy far outstripping chance. The superforecasters have been able to sustain and even improve their performance."

This is such perfect nonsense. The described program collects forecasts, forecasts that can be expected to be half right and half wrong for binary questions, throws out those that by chance were wrong, and preserves those that by chance were right.

In the next round, using only subjects that were right already, the same procedure is used -- throw out the bad forecasts, preserve the good ones. Each experimental cycle eliminates 1/2 the participants. After 8 repetitions, 1/256th of the original participants remain, all with perfect records. These are the "superforcecasters" the article describes. A computer could do as well, with an equal amount of logic -- none at all.

There is a financial scam called "Miracle Man" that uses the same scheme. It seems very convincing until you think about it:

http://arachnoid.com/equities_myths/index.html#Miracle_Man

If this were't the most absurd kind of pseudoscience, if real forecasting ability were being modeled, it could be tested using a prospective model (the gold standard for real science) instead of the useless retrospective design presently being used.

The authors of this study need to learn statistics, probability, the undermining influence of confirmation bias, the importance of the null hypothesis in result analysis, and a bit of common sense wouldn't hurt.

It has always seemed to me that 'predicting' the future is useless from an actionable perspective.

Even if you were able to give me a perfect prediction of the future, I don't need that. I need to model how my actions to take advantage of that future will change the outcome from what you predicted.

Most smart people can predict future trends, the value is knowing the right methodology to interact with that data to take meaningful action.

I see nothing here that relates to that.

Or as the Cowboy Junkies put it: you can always see it comin', but you can never stop it. (Bea's Song)

http://www.cowboylyrics.com/tabs/cowboy-junkies/beas-song--r...

1. Find 20,000 people to participate in Coin Predicting.

2. Flip a coin. Sort out all participants that were wrong.

3. Repeat step 2 until there are 10 people left. These 10 people now have a track record of predicting the future. Some of them even will continue to predict the future for a number of times. Eventually, all will be wrong.

The thing is, you don't know who will be the expert forecaster in the first place and the same is true for the Good Judgement Project.

The Good Judgement Project does encourage forecasters to make informed predictions. In fact there is an RSS feed of relevant news stories on the topic up for trade.

Your coin example is a game of pure chance while trading on current events have much more context to inform predictions. It's really quite a difference.

> Your coin example is a game of pure chance while trading on current events have much more context to inform predictions. It's really quite a difference.

The only difference is that, in a coin toss, the statistics are much clearer. In real event forecasting, the perversities of language can make a failed forecast seem successful. Here's an example:

A seer in the Greek city of Troy issues a warning: "Beware of Greeks bearing gifts." The citizens don't know what to make of this saying, so they proceed as before.

After the now-famous Trojan Horse incident, the seer repairs her reputation by revising her earlier warning -- "I actually meant to say, 'Beware of gifts bearing Greeks'."

The bottom line is that, whether flipping coins or evaluating real-world predictions, one can separate successes from failures with equal efficiency, and an equal lack of meaning.

In a prediction marketplace there are predictions that are more successful or less successful, resulting in more or less profit for the predictor. While the result may be binary, the forecast intelligence is useful in that it contains trend data that provides better predictions than any one forecaster can provide. Further, the outcomes of many individual prediction questions are multifaceted rather than binary, with four, five or more possible outcomes.

Yes, and without rigorous scientific controls, predictions can be made to look much, much better than they really are. The study being discussed didn't even perform a statistical analysis based on the null hypothesis or produce a p-factor -- the possibility that the result came about because of chance. These are basic to science, and in their absence, the result is anecdote, not science.

http://en.wikipedia.org/wiki/P-value

Quote: "In statistical significance testing, the p-value is the probability of obtaining a test statistic result at least as extreme or as close to the one that was actually observed, assuming that the null hypothesis is true."

Chance occurrences can be made to look very convincing. Consider the "Miracle Man" scheme used in investing -- on first hearing about it, many people assume it's not possible:

http://www.arachnoid.com/equities_myths/index.html#Miracle_M...

> The thing is, you don't know who will be the expert forecaster in the first place and the same is true for the Good Judgement Project.

They should turn it on itself - make anonymous profiles of the participants, and see if they can predict who will make a good forecaster!

I don't know why everyone seems to think the researchers are incompetent. From here (http://www.economist.com/news/21589145-how-sort-best-rest-wh...):

Given that it's a random event, a good coin flip forecaster would be predicting a 0.5 probability for each side rather than a 1 for either heads or tails. Consequently, of the 20,000 coin flippers, you can tell who'll be a good predictor before a single coin has been thrown - the good ones are the ones who think it through and are open-minded enough to say their prediction is "I don't know."

If you insist on a heads or a tails then your experiment is broken by design and tells you nothing.

I don't know if your assessment of the study is valid (it seems a little simplistic to think that they wouldn't account for this to be honest), but what you describe is classic Survivorship Bias.

Here's an entertaining video from Derren Brown that shows how to use this technique to convince people you have psychic powers: https://www.youtube.com/watch?v=9R5OWh7luL4

Edit: also see the comment about the Miracle Man scam in this thread.

If I wanted to know how to see into the future, I'd ask a futurologist with a good track record, not a publication whose MO is writing about things that had just happened (hours or days ago.)

Reply on news.ycombinator.com