Convopage
See the entire conversation
These aren't backtested results. All real forecasts made in real time, in advance of events.
One think this exercise demonstrates is that Nassim Taleb (@nntaleb) has no idea what he's talking about, but I thought everybody knew that already.
+@NateSilver538 are you backtesting the predictions you made *immediately before* events? If I understand correctly, @nntaleb criticized 538 predictions made too far before events as over-confident due to underconsidering epistemological uncertainty.
557 replies and sub-replies as of Apr 09 2019
Bullshit Nate, my argument has nothing to do with backtesting.
Like it or not, Taleb theories have verified multiple times over the years. Launching a coin > Nate Silver.
That's one smart coin.
Demonstrably untrue tho. Yeah, I mean, launching a coin has less error on it if 538 has a candidate at 50% chance of winning, but a coin's going to look a bit different compared to <25% and >75%.
lol you’re literally saying nothing. It’s just empty words.
severe derp alert
Me Nate , there needs to be mathematical rigor to your argument not “Taleb doesn’t know what he is taking about”.
Read the article
I have read the back and forth since January. Let’s just get to the math and not empty words
Yea, it's not complicated, nate is right.
even less complicated, you are clueless and nate is wrong
Yes, please review the actual substantive paper that Taleb didhttps://arxiv.org/pdf/1703.06351.pdf Please look at the predictions of polling and politics and the uncertainty of the data points. High uncertainty converges predictions to random walk
Except, the empirical data shows taleb was wrong.
A weatherman takes data and uses models to predict the weather, then calibrates those models based on performance against the actual weather that occurred. If a weatherman says theres a 70% chance of rain... then out of all days for that prediction it should rain 70 of the time.
This is exactly what nate has done.
This is not an adequate response to the paper I presented and people fail to comprehend. Weather predictions are not the same as polling or social science. Empirical data is more complex so Taleb is less wrong and Nate is less right. To claim more knowledge than that is foolish
This article gets at the "70%" chance of getting it right. You can't be wrong and right at the same time without mentioning error rates:
Why you should care about the Nate Silver vs. Nassim Taleb Twitter war
How can two data experts disagree so much?
towardsdatascience.com
That’s very mathematical if you. Get a crash course at local college
The issue is roughly that real time polling doesn't actually add any information about who's going to win.
I think there should be a period after 'Nate' instead of a comma.
Getting to the core of the issue!!
Easy there Nassim you want to go after a guy who predicted two out of the last three elections correctly 😂 ? Also, don’t try to argue statistics with him or he’ll start to believe that’s what he does, rather just advanced punditry.
He didn’t predict it correctly; he was just less wrong than most other outlets
2008, 2012 were the ones he got right. He got 2016 completely wrong from the primary onwards. Which is fine. We all make mistakes. But he is a political pundit who thinks he is doing science.
Go back and check his last predictions.
4 out of the 5. Not 2 out of 3. He has only been doing this for 5 elections.
I thought he started in 2008
Yes. Which would be 5 elections.
List then
Nerd fight!
Following you
You truly truly do not know what you’re talking about and need to log off and have some water
The argument is that the roller coaster "chances to win" that Nate shows in the months leading up to an election is nonsense because its all in the very large error bars, correct?
Polls asks ppl "if the election were held today" and ppl answer knowing full well it isn't
He’s also conflating the term prediction with odds. Roll a die—odds are it won’t land on the number 4, but if it does land on 4 it doesn’t mean you were wrong in saying it most likely wouldn’t.
Or that the die could fall off the table and roll under the fridge making a 7th outcome. Elections do not have nice craps table boundaries
Right, anything short of saying the odds are 100 to 0 is not a prediction and the fact that there could be some significant factor that changes in the future is obvious (Comey announces reopening email investigation, Tom Brady breaks his leg practicing for the Super Bowl, etc.)
Isn’t Nate the guy who said trump wouldn’t win the primaries when it was obvious then said Hillary would win Vs trump when it was obvious trump would win?
Not sure how anyone could say it was ever “obvious” Trump would win the general. A few thought there was a better chance than most pundits said, but it was hardly a certainty or even a likelihood in the final weeks. (P.S. I wasn't a Hillary voter if that is important to you.)
People who knew about fake news before the term was coined knew Trump was going to win. If you are indoctrinated and we’re listening to the propaganda machines like @CNN and msNPC then you were shocked on election night 2016.
The only thing Natie backtests is what comes out of his #Democrat friends' asses
Broke: reading black swan and appreciating Taleb's work
Woke: realizing that Taleb's ability to write anything of value was itself a black swan event
That was a good burn.
So true! When he is good, he is very good, when he is bad....
We need a real in person debate to settle this!
Silver is too unsophisticated in probability to get the argument.
If I get your point it’s that his forecast is a point estimate of the outcome, whereas financial markets would price it with uncertainty around it and in the 2016 case even more. So that the digi would always price at 50%.
I guess In Nate’s models a higher vol would lead to estimates converging to 50:50
It is a convexity argument. The upper bound is the poll, so can be 48.
Nassim, you seem to operate under the theory that if nobody understands you, it makes you a misunderstood genius when from a Bayesian standpoint it probably just makes you an intellectual-yet-idiot.
💥🚨NERD BEEF ALERT 🚨💥
Ok, even Taleb would probably have to admit that was a good one...
In the EconTalk version of this conversation, I would then point out that Silver responded to a tweet about backtesting, Taleb then took that to mean that Silver thought Taleb's criticism was about backtesting (which it's not) and we could learn something!
Clearly this is a business opportunity for econtalk... It may not be good evidence of emerging markets if doesn't occur.. in this corner wearing the red trunks the bootlegger ...
Nate Silver, rather Bullshit Nate, my paper went through 4 professional (and throrough) referees who "understood" it. I am not saying "nobody understand me", but YOU don't understand probability.
I do science, you do bullshit.
Does someone want to break the news about peer review...
I can vouch for Taleb. If he'd like to take a little risk on me I have a Black Swan idea, I just need help getting started.
Loooooooolllll now you're falling back on "but my paper was peer-reviewed!", the surest possible sign that you've become an intellectual-yet-idiot.
But anyway, bro, let's debate on @EconTalker's podcast. What you so scared of?
Regression to being mean
outstanding tweet
Did not expect to see esports Twitter in here
did not expect to see dat boi in here
o shi waddup
If I understand correctly he was criticizing your predictions were made too early before the event. He was not stating your forcasts were made after the fact.
I would love this.
I'm with Taleb 90% of the way on this (and I say this as a huge fan of @FiveThirtyEight) but I trust @EconTalker to make sure we get to a better place of shared understanding.
I bet @intelligence2 would love to facilitate this!
He also didn't say his article was peer reviewed as why it was superior he just said that the peer reviewer understood it.
A student with a D or an F can claim the paper was graded, not much else
Then silver could say them understanding it doesn't make it true.
You said "nobody" understood it. He gives you examples of people who did. You then ridicule him for relying on peer review. You're just bait and switching, it is rather annoying and doesn't seem intellectually honest.
Hence why "debating" him would literally be feeding a troll as mentioned elsewhere... I gotta say I'm a little disappointed in @NateSilver538 .
I’m guessing he doesn’t want to discuss Taleb’s paper because he doesn’t understand it or he can’t disprove it. So instead he wants a slinging match that will make good “popular” entertainment.
Yeah. Backed w/ more followers he can't risk to lose a debate, but he can try to ridicule. Basic trolling. Not putting one's own hide into game but slinging shit from afar. But it seems Prof Taleb understands trolling and knows not to waste time and energy which is what he wants.
I would like to see this on @EconTalker but if the great Russ Roberts cannot help solve this, the only solution is to put on the gloves and settle it in the ring. Who would be the favorite in 538’s prediction model @NateSilver538?
My money’s on Taleb. TKO within 5 minutes
Watching Taleb run rings around Nate Silver is the best thing on Twitter in weeks!!!!
'What you so scared of?' Yeah, OK Nate...or. I mean...'Bro'. You be straight from 'da hood and shit.
What a dork.
Nate you are an entertainer, Taleb is a scientist.. let it go and continue to entertain
Can I join the debate? I would love to ask some cool probability theory questions to Nassim.
Me too! 🙋🏻♂️🙋🏻♂️
Like: “everyone who disagree with you is dumb?”
What a same, @nntaleb just block me. All I asked was a debate about probability theory. Seems he is not qualified to do so. I wasn't even planing to ask him anything about ergodic theorems.
Already? You didn’t even ask about Bayes Rule...
He is pathetic
*shame, you bastard politically correct auto-corrector.
Só agora eu descobri esse Taleb. Ele parece meio charlatão. As pessoas gostam assim dele? Por que?
O livro do taleb é tratado como uma biblia para certas pessoas:
youtu.be/-jYBVYpUhMU
impressionante. Não tinha ideia disso
O que é Antifrágil? - Os Antifrágeis 1º Episódio - 02/02/2017
Blog Luiz Fernando Roxo - http://blog.luizfernandoroxo.com.br/ Plataforma OpLap: https://www.oplab.com.br/ Conheça meus cursos Curso de Opções Básico - http:...
youtube.com
tem muita gente ganhando dinheiro gracas a ignorância alheia
O livro eu ainda acho interessante. Meu problema é quando ele esnoba a academia e fala “mimimi PhD não tem Skin in the game...”.
Dito isso por um PhD que tem zero relevância acadêmica.
Exato. Gosto doa livros dele. A postura diária, porém, é triste!
Eu li o Cisne Negro e Antifragio, acredito que aprendi alguma coisa. Mas depois desse debate fico pensando quais livros deveria ler. Sou investidor individual, trabalho com comércio exterior e gosto de investir por minha conta. Sugestões?
Seguir @samydana @Tiagogreis @infomoney @terracoecon.
Em termos de livros, The Inteligent Investor, Desafio aos Deuses. Conhece os dois?
The intelligent investor eu li. Desafio aos Deus ainda não.
Obrigado pela dica, vai pra lista.
Obrigado pela lembrança
Devido a ignorância generalizada sobre teoria de finanças no Twitter, vai ter thread de livros.
Mas qual o problema específico com o Taleb, exceto pela versão Olavo de Carvalho americana de xingar todo mundo?
Taleb, o mala-mor.
Hey, Nassim, @rationalexpec just wants to ask you a few probability questions. No need to block him.
yep, I know that @nntaleb probability theory knowledge is limited. We can keep the debate at the undergrad level, I won't go into measure nor ergodic theory.
Let's begin with a easy one: Consider a shuffle a deck of 52 cards containing 4 aces. Then turn up cards from the top until the first ace appears. On the average, how many cards are required to produce the first ace?
Qual a resposta?
2: Choose the largest number from 0 to 10000 in slips of paper. A slip is drawn at random and you have to decide whether that is the largest number or not. If you say yes, you get $ if your are correct or zero if incorrect. If you say no, drawn again. How should you choose?
In this message, it was possible for you, @rationalexpec, to have typed “$ if [you] are correct” instead. ‘Your’ is a possessive determiner; ‘you’ is a pronoun.
I would draw till I get the slip with 10.000 written?
Conheço esse problema 😂. Muito bom e não-trivial. A quem interessar possa (Cap 3, a partir da página 38):
impa.br/wp-content/upl…
vocês dois entregando o ouro para o bandido, assim até o @nntaleb resolve. 🤣
valeu pela dica do livro do Landim
Problema da princesa?
originalmente sim, mas reescrevi para não ficar tão obvio
I'm sure @nntaleb is afraid of forgetting his dentures. Now - will someone please help him find his cane?
I'll buy that for a dollar!
This reminds me of when Andy Kaufman taunted Jerry Lawler to a wrestling match.
I have to say -- I think Nassim is right on this point. If you even assume something simple, like variance of the actual distribution is proportional to the time till the event, you'd arrive much closer to 50% for the forecast far away
I guess his comment could also be interpreted as, there's not even difference between the nowcast and the true "forecast" - and adding a variance term for the time to the actual event would probably stabilize the forecast a lot more
For the Bloomberg forecast, just adding a constant time to election term gave us a great deal more stability and remained reasonably well calibrated - arxiv.org/abs/1704.02664 .. Its a common practice in option trading, and usually quite a win
Accurate Prediction of Electoral Outcomes
We present novel methods for predicting the outcome of large elections. Our first algorithm uses a diffusion process to model the time uncertainty inherent in polls taken with substantial calendar...
arxiv.org
Curious how the constant time to election term impacts the probabilities when you have dark horse third party candidates in the calculations.
enough of a difference between**
WARNING MR WRONG IS A MONSANTO SHILL
This may be correct, but then the debate turns on the utility of a 50% forecast. Should @NateSilver538 wait until the week-of to make a prediction? Or is there something to be gained in a higher-variance, further-out "nowcast"?
Yes, there's something to be gained: readership and ad revenue. Duh.
Ok, but the readership and revenue are driven by an interest in what the data say, even months out. There's a reason candidates conduct polls far ahead of the election.
The data has more utility as a campaign ad than as predictor.
I had assumed campaigns conducted early polls to gather data to use in making campaign decisions.
Are you saying that it is instead, or additionally, being done to feed the appetite for, idk, "data entertainment" by the politics-interested public?
Lol.
Does the model incorporate the model's role in creating the outcome?
We're such a dumb species sometimes. Absorbing millions of data points to conclude we need the event to occur to know what will happen.
Why would you assume that? It seems like there's (at least) two distinct sources of variance - one from changes in the polls and one from reliability of the polls. If latter dominates then you wouldn't expect much impact from time.
What test would you do on his data to falsify?
Debates are for nerds, @NateSilver538 prove everybody wrong by using your models to make money in the markets,sports or election betting etc. Without blowing up. Until then you’re just snake oil salesman
He will just say that you are a charlatan and not worth the time
Fuck the Mcgregor-Diaz trilogy, give me Nassim vs Nate pls.
would get absolutely destroyed by @nntaleb. obviously.
you both are embarrassing
The paper is there. Stop talking and write a response. Are you afraid that you might have to think beyond debate one-liners? Surely you have something other to offer than shallow analysis displayed in pretty charts.
With Trump style branding you must be an intellectual!
"I do science, you do bullshit" What a great fucking line that is.
Who has more F*ck You money?
I think we know the answer to that
NERD FIGHT!!!!
Is this the part of the conversation where someone from outside your profession gets to call one of you a quack?
He's the DJ, I'm the Rapper, I do Science, You do Bullshit
"I do science"? Not saying this is incorrect, but all I have ever seen from @nntaleb are publication in math and philosophy. I suppose the math publications are broadly 'science', but I must admit I never viewed Taleb as a "scientist."
I feel like what this comes down to is, Silver has a product hes selling, so the elegant science is less important than the marketable result, whereas Taleb finds more value in the nuance and complexity of the art. Silver is scoring a Pixar movie, Taleb is conducting a symphony.
You do science? Hahahahahahahahahahahahahahahahahahahahahahahahahaha.
Thread by Dhruv who is the other researcher who got the same result I did. Unlike Bullshit Nate, he doesn't have 3 mil twitter followers. Jut ~ 50.
His paper is on the thread. He is more polite than me with Bullshit Nate, but same result. @EconTalker
I have to say -- I think Nassim is right on this point. If you even assume something simple, like variance of the actual distribution is proportional to the time till the event, you'd arrive much closer to 50% for the forecast far away
Now he has ~51 followers
Now 81📈
Can y’all please debate
now he's got ~100. That said, BS finds far more followers!
Seeing grown men calling each other names...the desperation to not get labelled wrong in front of followers is too damn high.
Nassim, @nntaleb I know your stance on debates (don’t block for this :p)
Many of us want to understand the argument. Please discuss the paper with @NateSilver538 on @EconTalker
No bs “debate”. Genuine conversation on maths and statistics.
Lots of people want to hear that. @naval
Thread by Dhruv who is the other researcher who got the same result I did. Unlike Bullshit Nate, he doesn't have 3 mil twitter followers. Jut ~ 50.
His paper is on the thread. He is more polite than me with Bullshit Nate, but same result. @EconTalker
Dhruv Madeka on Twitter
“@NateSilver538 @nntaleb @NcojoD @EconTalker I have to say -- I think Nassim is right on this point. If you even assume something simple, like variance of the actual distribution is proportional to the time till the event, you'd arrive much closer to 50% for the forecast far away”
twitter.com
Interesting that @NateSilver538 still has no counterargument of substance. Only nitpicking: Complains that Nowcast was used, when all 3 his forecasts suffer from the same principle fault of overreaction months ahead of the election
Yes. The now-cast came later and is BS.
Nassim, it's pathetic that for the past 2 years, you've favorited/retweeted 100s of comments from random Twitter nobodies that take your side of the argument*. Maybe you should save some of your dignity and debate me?
* Not really an argument because you literally have no point.
No blue check mark = "nobodies" in your mind? Just curious what you mean by this.
He means that he thinks he’s better than you because he has a blue check mark and a lot of followers.
"random Twitter nobodies" is an elitist remark. It is possible for someone not in the Twitter cognoscenti to make a good point based on an informed opinion.
Amazon data scientist without blue check mark = "nobody" according to the almighty Nate lol. Ya know, you'd think being mediocre and overconfident would curb his pretentiousness, but nope.
If Trump would have lost you would have taken credit for correctly predicting the outcome but since he won you point to the 30% chance you gave him and scream “you don’t know math!” at anyone who says you got it wrong. You can’t have it both ways
dude your job is to estimate the aggregated opinion of these nobodies
Your noting is noted.
The amazing thing about this fight is that Taleb was first to say “a bunch of bad guys can take over a country and turn it to s**t very quickly” and then he ends up supporting the bad guys.
"random twitter nobodies"
Lol, outstanding argumentation.
Please let Taleb know you are also disappointed in his use of the word "retard" and "unthinking animals" to refer to others.
Good science communication means never calling them “retard” – even if you’re Nassim Taleb
Communicating science to people who aren’t scientists is very hard to do well. Nassim Taleb should be very good at it, based on his enormous book sales and even more enormous opinion of his own ski…
violentmetaphors.com
Taleb is a gasbag looking for coattails to ride to remain relevant.
Silver, you’re a quack. Do not engage me, just broadcast to your followers. 3 million or 3 billion doesn’t make a diff.
I write formal papers.
LOL I can't believe all it took was for someone to stand up to your bullying by trolling you back for you to turn into such a cuck.
To be fair I'd probably be afraid to debate too if I hadn't had an original idea in 10 years.
Here, here
Could some blue checkmark friendly to Nate tell him to dial it back a bit? He's coming off a bit looney and unhinged in these exchanges.
He's doing fine. He's right and he's mad, and he gets to be. Taleb is a bullshit artist who's calling Silver a bullshitter while also refusing to actually debate.
As a Wall Street modeler (for 20 years), I have questions/concerns with how your number is calculated. In particular the forward projected vol used for calculating prob not equaling the realized vol.
An option model is only "right" if the projected vol is correct. That is the bet. A forward election projection is the same (or am I wrong on that?)
I don't know what this means. What is the "volume" that Nate's projections are calculating?
vol = volatility (not volume!) Hopefully that makes things a tiny bit clearer!
It does! Thanks!
God its great to have you back Chris.
One thing recently pointed out to me (cc @AlexGodofsky) is that election-related options would be different from a traditional financial option model in the final day. Suddenly the pace of public information about the underlying fundamentals speeds way up.
There are ways to handle that (we did them on Wall Street). Time is not linear by day! Bigger issue I would imagine is non hegability. Finance models are based on replicating portifolios. Goal is different
To be more precise the issue is not just nonlinear time it is that there is a setp between the T minus epsilon prediction and the T minus 0 "prediction". The prediction should not converge to 0 or 1 because the polls can be wrong.
Right. Or you need to use a jump model. Again. Jump models are not hedgeable. And Elections are not either...
By contrast with an option the uncertainty decreases towards zero at the end. Essentially there is a lot of "time" in between T-epsilon and T-zero. I believe (but haven't checked) this accounts for the difference btw Nate's prediction and Taleb's view of what it should look like.
Once you incorporate jumps (or discontinuities in underlying), you move to a non-hegable world and results are one-off and only true in aggregate.
I am not a finance persons so it's good to know there was something to my guess.
Your instincts are right. Elections are not hegeable events. They differ from Wall Street. But Wall Street has thought a lot about them. And Nates stuff is good, but I believed flawed
It's kind of like you had an option dated just after the company releases its quarterly earnings announcement. You expect vol to be much larger in the final day, and that's just a normal part of the job.
Another way of framing that is, 88.8% probability on Aug15th seems inconsistent with the subsequent volatility in that number. Or the underlying volatility in the electorate doesn't match the degree of certainty in your model on Aug 15th
The solution is use a higher underlying volatility for the Electorate == which gives a number closer to 50/50 -- and at some higher level starts to wash out any predicatability
Really interested in this. Can you expand on this underlying volatility guiding closer to 50/50 notion in lay terms?
The more variance there is in the path, the less the current position tells you about the likely future position.
I bet one could justify the series using a hierarchical Bayesian model (multiple regimes with different means and volatilities, and some shifting distribution over regimes).
That is true, but from a finance perspective it means the model is useless -- beyond giving you a number that summarizes the Now.
I don't think that means it's useless. Maybe it means you have to be careful how you use it to choose bets.
From a Finance perspective. Again. The goals are different when you can hedge and you approach things from the perspective of a portifolio
I think whether it's useful from a finance perspective depends on what markets exist. (Since prediction markets don't work very well anyhow, the question might be academic.) Might depend on ability to buy options on options or options on options on options.
Why don’t prediction markets work well, in your view?
This latest MCU crossover of Avengers of Quant is EPIC.
Here’s @nntaleb as Thanos.
twitter.com/nntaleb/status…
Silver, you’re a quack. Do not engage me, just broadcast to your followers. 3 million or 3 billion doesn’t make a diff.
I write formal papers.
Regardless. I would debate you on @EconTalker pod. Although I would prefer it to be a discussion since I think issue is one of different goals/models between Wall Street and Political blogs.
The root of the difference is the two diff objective criteria: abnormal $ returns vs public acclaim/notoriety.
You’ll never hear from Nate. His scam is that assigning %s is untestable since we can’t turn back the clock and run the election 9x more to see if Hillary would’ve won 7 of them. Real pollsters assign margins of victory that may be wrong but are testable. Nate predicts nothing.
To the extent that the 88.85 forecast is overconfident, wouldn't that type of error show up in the 538 analysis of their historical projections? They're clearly including all projections, not just the final prediction just before an election
Guess my point is, focusing on presidential election results in looking at a very small sample size. But 538 forecasts hundreds of election results! If they were systematically overconfident in the way you describe, that should show up in the analysis @NateSilver538 is discussing
As long as 538's model does not formally incorporate the decaying time-value of the underdog's option (e.g. via vol & mean-reversion assumptions about the stochastic vote share process between observation date & expiry), NNT has a substantive albeit insufferable argument
He won’t debate because he thinks nobody in attendance is smart enough to understand his points. In other words, he knows he would lose
A good many wouldn’t, tbh.
Ha. Soon we are going to find out who Nate Silver really serves (hint hint....he likes making the right-wing of the Democratic party happy.....why?). To be fair, Taleb has said some batshit crazy things himself throughout the years. But Nate is exposed again.
Nate is obviously a defender/tool of the establishment. He loves the fame and money. Same as all the MSM pundits and reporters.
come on Silver, fly low.
Taleb is nostalgic for some mythical age (roman empire, his Lebanese childhood) when "local" (his term) authorities obeyed his "skin in the game" protocols. His affection for Trump, and his reasons, are part of his simplistic world view.
ORIGIN OF LOCAL
First recorded in 1400–50; late Middle English word from Late Latin word locālis.
Damn, Taleb is old.
You should probably re-read things before replying.
Don't debate. BET. We have an election coming up. @NateSilver538 can post a two way no wider than 5% around his prefered model each week after the primary. @nntaleb has an hour to hit or lift that number. #skininthegame #betnottawk @bryan_caplan
(No comment)
What is an original idea that you’ve had, Nate?
He started a popular blog called 538 that people read and this allows him to earn a living
He has definitely built a valuable business. But I can’t think of even ONE original idea he has had.
Define an "original idea". I've seen Nate apply lots of statistical techniques to things others hadn't thought to. Does that not count? Is it better if someone writes a whole book to say, "Sometimes unexpected things happen"?
You bet on a six-sided die and coming up a 5 or a 6. It did, and you think that rigorously proves something important about statistical models?
So that's why you haven't engaged in a debate in the last 10 years! Thanks for clearing that up.
no amount of time in academia could prepare me for data twitter fights
"cuck" that's gonna be an unfollow. blech
Taleb claims Nate is making a forecasting error, but Nate says his Nowcast is not a forecast of what will happen on election day but instead a prediction of what would happen if the election were hypothetically held *now*
What good is that?
it gives you a sense of where the electorate is *today* and it is volatile enough to be interesting.
IM SO ANGRY ABOUT MATH
Nate finally back on his king shit rn?
The bickering between these two is such good content. The best of twitter, maybe.
Math geek fight! 🍿
Well that escalated quickly.
You just proved his point with this comment.
That's not how that works
Isn’t the simple way to solve this debate for both of you to put your money where your mouth is.
@NateSilver538 offer @nntaleb a contract on your daily odds for the coming election and he can trade them and see if he can make money on his perceived mispricings.
i have no interest in watching a debate but I would like to see an MMA style fight
Silver wouldn't win a fight on an elementary playground. So full of soy.
Dude over there appears to be wearing a santa hat so I'll bet on him. Maybe his maths are novel but he doesn't call anyone cuck, does he? Cuck is a word reserved exclusively for cucks.
I guess we're all cucks then
The “lol” and “cuck” in the reply say it all.
Taleb has blocked so many mathematicians/physicists/logicians/etc. (for calling out his crank nonsense) that neither me nor any of my colleagues can see his tweets to watch this interaction unfold lol.
If you disagree with Taleb, go write a comment, submit to same journal. Otherwise Zach Kim you are the crank. Cranks are those who do not use refree system.
This would individuate anyone mildly educated in the domain as a crank, which seems a bit undesirable. The only people who wouldn’t be cranks would be those who are, in fact, uneducated. For you, there might be another pragmatic usage of “crank” operative here?
It remains that if you disgree with Taleb's technical work, trolling him on twitter is not the solution. He is justified to block you.
I finally blocked Nate Silver.
hell yeah dude
Arbitrarily expressing false propositions is a respectable hobby
I feel like the better option would be to stop tweeting bullshit about him.
What are you responding to, btw.? The tweet is unavailable for me.
Taleb said he finally blocked Nate
I’m not sure that a group of experts on P/NP shitting on his (not at all technical) conspiracy wrt. it consists in “trolling.” Else much of maths/physics just is trolling. Your account’s only function is to troll, though, so this is a weird flex?
And I’m not saying that it’s normatively bad that he blocked a bunch of experts who debunked his nonsense (no one takes him seriously anyways). I was just noting that we couldn’t follow the exchange!
"no one takes him seriously anyways" is BS. I should be blocking you.
True, cranks do take him seriously; I should have qualified that proposition it more carefully. And oh no, not the block!
Taleb has 18000 Google citations. By scholars, that is. How many do you have?
Uh, yes, this doesn’t preclude him from being a crank in another domain? JP also has plenty of citations, but is a crank in multiple fields (philosophy, econ, physics, causal inference, etc.).
How many citations do you have, please.
The only way I can tell you are not a crank is whether you have published work taken seriously to cite.
While I’m not going to doxx myself to a crank, this isn’t how we individuate cranks (and it’s not the extension of ‘crank’ in English). An obvious reductio: all physicists are cranks until they receive their first citation. Then, magically, they become experts.
So you are some anonymous person claiming to be an expert on twitter and attacking Nassim. And you believe I should believe you?
Huh? You don’t have to believe me, nor do I believe (or care) that you ought to?
How many citations does your work have?
Someone posted this, I found it a good summary of the Twitter argument.
Nassim Taleb’s Case Against Nate Silver Is Bad Math
Nassim Nicholas Taleb has overplayed his hand this time and is left looking, well, klueless.Photograph by Salzburg Global Seminar…
nautil.us
Generally, it's not a serious article if in it's first para it declares someone right and wrong. twitter.com/crimsonceo/sta…
twitter.com/DhruvMadeka/st…
twitter.com/__anoop/status…
A better overview here towardsdatascience.com/why-you-should…
Of course, in the end one believes what one likes to believes.
That’s a rather silly heuristic. E.g., was someone to descend to Taleb’s level for his being a crank wrt. P/NP, it would likely be most desirable to state immediately that he’s horribly confused and wrong, and that mathematicians don’t follow similarly.
The crux of the matter is: Talab might be rude/unproportionately harsh but silver has not yet refuted his central point, instead behaving like a juvenile with ad hominems like 'cuck' etc.. The thing is silver has too big a reputation/ego to acknowledge his faults.
I personally don’t think that anyone (especially not anyone educated in the domain) should waste their time responding to Taleb, so I don’t mind this too much.
Well, if Silver really didn't care about it he wouldn't engage in this twitter spat so your point doesn't apply to him, i think even if you hate someone you should respond to a peer reviewed scientific paper if you think they're wrong.
It does apply to him (it’s just a general normative prescription), he just doesn’t adopt it.
As for the second part, sure! But Taleb doesn’t even follow this principle when he enters the domains he’s a crank in (e.g., pure math, genetics, medicine, and so on).
You see here's the problem: If you don't want to engage with someone just don't do it altogether. But if you choose to, reply with substance rather than rhetoric. Also, your view that experts shouldn't engage with Taleb is your opinion, many experts care about his ideas.
I (and quite a few people with the corresponding views in social epistemology and the like) disagree, given that there’s a rather strong non-rational component to our interactions. We engage with flat-earthers and anti-vaxxers in ways that don’t reduce to a lit rec.
And yes, again, people take him seriously insofar as he confines himself to his humble domain of expertise. But this fails to obtain when he leaves it, and becomes a crank who isn’t taken seriously by anyone. This isn’t at all uncommon.
So, using the previous example, exactly zero experts care about his silly beliefs about P/NP (especially his insane attempts to use it to assist his anti-GMO conspiracies).
I blocked this anonymous liar. I am publishing in ALL the fields where I take positions. Includes genetics.
I don't understand why people make this complaint. Are they really such unsophisticated computer users that they don't understand that all Taleb's tweets are visible to anyone not logged in to twitter or who view from an incognito window?
It’s not so much a complaint, and I think the nuance is captured by its obvious implicature. That is, no one really cares that they’re blocked (and they probably already have looked at the exchange through other means). It’s just funny to note that most of us have been blocked.
Being blocked is usually taken as a good thing—at least, in this context. It’s just an indirect use of language!
You do know you can simply log out right? Or use a private browsing window which won't have your login context.
Ye, I’d already seen the exchange before tweeting this
I hate when my dads fight!!!
(No comment)
I am blocked by one of these people
(No comment)
Get his ass
Nice insult, bro. You’re coming off really well in this thread. Zero substantial points, but you do say bro, cuck, and LOL a lot.
BOOM. mic drop.
Pulling out “cuck” is an automatic loss buddy
Getting called a cuck by the nerdiest guy on Twitter is hilarious
Nate go on Chapo
Super happy about the number of fellow amused leftists in this thread
Nate had some lunch beers, everyone
Excuse me???
Save the "cuck" for 4chan, Nate.
Did you call someone a “cuck”? There goes a lot of respect I had for you.
Nate! No!
You sound like a child. Grow up.
I AM HERE FOR THIS
ya know, i don't think you are really loling if you got pissy enough to become a Debate Me / Cuck Guy that fast
Bronze.
Lmao Bullshit Nate
Gonna file this one under ‘signal’
Ohhhh now I see why people don’t like you
In b4 the deletion.
Having a normal one I see
please do not kink shame. i thought u were better than thsi
Dude I can tell you're a huge badass.
Way 2 use mysogonistic slurs. Looks good m8
Now finish it off and call him a little bitcg fgggt
Lol damn dude you look exactly like the kind of person that uses “cuck” unironically. Would you also like to call him a Chad while you’re at it? How many more years until you’re using your stats website to try and prove race science
Been hanging out on 4chan, dude?
Nate still around after the forecasting hoax of World Cup 2014 & election 2016?
Really?
Better forecast baseball Nate...where probabilities live in Mediocristan. Much easier
The@other attempts where just a good example of model fragility
Did you just unironically use the word "cuck"?
I gotta admit, I’m getting a little horny imagining Nate’s entire body trembling as he typed out the word ‘cuck’
The two biggest probability nerds getting into a standoff. This is the type of content I’m here for.
Nate, dude, “cuck” is such an alt-right thing to say
(Possibly sensitive)
when your extremely not mad online
wtf i love nate silver now
(Serious people don't use "cuck" this way, just sayin'.)
Jesus Christ Nate Silver is a bad ass... LOL
When probability nerds fight....
bruh snapped
lololololol wut happened here
did you just say cuck
I love it when soyboys think they know what “cuck” means.
This whole exchange shows how Twitter halves your IQ.
It would be worth considering the contemporary meaning of the word "cuck" at this point. I don't understand its usage most of the time anymore, but it's typically used in right wing circles. Why is Silver using it here and what does that tell us about his discourse?
Is this a Ben Shapiro burner? JEEEEEZ
Nate said cuck, indicating not mad at all
This foolishness makes me glad I stopped reading 538 in 2016. People will mostly likely continue to pretend 538 is scientific and not just standard nonsense punditry.
madman just used the c-word
Whoooooaaaaa there Nate dial it back.
You think this makes you look tough but it actually sort of just makes you look way too online. Time to log off my man
Cuck? Dude. Do you know where this word comes from and what it means?
Nate, what the fuck is this?
in addition to pretending his forecasts don't suffer from massive volatility, now-cast of an election doesnt make any fucking sense
there is simply NO GROUND TRUTH in a "now-cast" for an event that doesnt happen -- utterly meaningless
The only purpose for a "now cast" that is months in advance of the election is to try to influence public opinion. I watched their live tweets of both party's conventions, and their bias was painfully obvious
It's marketing tool for the journal he writes for. Fancy graphs wrapped in fake analysis. Basically he's a poll marketer selling his poll data for $100 dollars when one can buy it for $10. It's also basically useless to anyone but the campaigns and news orgs.
Thank you for showing me that Nate is a waste of time. You helped me to see who Stephan Moleyneux really is, and now you've showed me about this Nate guy. Bravo.
Same here
Same here. Loving it too
I'm surprised you haven't blocked him yet
Silver is like one of those people who thinks he'd make a few hundred million dollars on Wall Street because of his paper trading prowess.
While I personally do not understand the math, any critique of your position without utilizing math is most likely BS. If Silver had the math to back him up, he would show where you are wrong. This has not happened. So I assume he is full of crap.
Asymmetry of info is strong.
Jaffer's bullshit detection is strong.
NNT community's BS detector is strong.
Most authentic in world right now, literally No.1
The sum total of original ideas is useless if every one of them is bullshit. The secondary effects might even prove costly. Mr.Silver’s analytical skills appear to be more than worthy of a Goldsmith Prize.
I write formal papers 😂😂😂😂😂😂
Taleb, you've turned into a quack.
You used to be well-respected.
No, stick with "bullshit nate", that's a good look. very trumpy
Has anyone actually read the paper and *not* taken Taleb's side of the argument?
I found one person in Nate's corner who claims to have read the paper, but I'm skeptical since (s)he blatantly conflates Taleb's critique with De Finetti's no Dutch Book argument.
Nassim Taleb’s Case Against Nate Silver Is Bad Math
Nassim Nicholas Taleb has overplayed his hand this time and is left looking, well, klueless.Photograph by Salzburg Global Seminar…
nautil.us
De Finetti's argument is that no-arbitrage prices need to satisfy equations of probability. By same reasoning, no-arbitrage means there's an equivalent measure making prices martingales. This is the source of Taleb's critique, but he misses that the converse is trivial.
You're just repeating what I criticized you for. Your argument is literally
"Nate uses a model where probabilities sum to 1, so there's no arbitrage opportunity." Taleb specifies a distinct no-arbitrage condition and explains in the paper how it differs from that of De Finetti.
No. Probabilities that respect Bayesian updating are automatically martingales by the tower property. Therefore, there is already an equivalent martingale measure -- the physical measure -- hence no arbitrage by the converse direction of FTAP. Read this:
Response to Taleb.pdf
(no description)
drive.google.com
Also, Taleb does not actually articulate a no arbitrage condition; he just says 538 forecasts "violated arbitrage boundaries" because they fluctuate too much. But a binary option on a BM halfway toward expiry has the uniform distribution on [0,1]. So where's the boundary?
His basic mistake is that increasing volatility also increases realized deviation between the spot and strike price; this cancels out the effect of higher implied vol on the option price. All points I made in the blog post, in simplified form.
I would be interested in hearing @nntaleb's take on whether with increased volatility in the price of the underlying asset option prices will fail to converge on 0.5 due to concomitant deviation between spot and strike price. Sounds plausible to me, but I'm not an options trader.
What *doesn't* seem plausible is that @nntaleb would be wrong about a foundational relationship in the specific domain (option pricing) where he made his fortune and reputation.
Thanks for providing a link to a formal paper, Aubrey. I didn't realize you'd written one. I look forward to checking it out, and withdraw my criticisms until then.
All the math and papers come to something, and from all that, here and now, he is just making us dumber with the insults. I know real-time is real-life which is full of uncertainty. Demographic tendencies, and the like, can be accounted for when modeling but life is uncertain.
why so condescending Nate, random twitter nobodies can still make a valid point and debating is beside the point since you can publish a peer-reviewed rebuttal if you do care about the substance of *your* methodology. A debate about math is as illuminating as smelling a proof.
Nobodyism.
Ah, so the more followers one has, the more right they are?
Ha, I used to look up to @NateSilver538. The line "...random Twitter nobodies..." says a lot about how full of himself he is.
Or, perhaps Twitter following is correlated with IQ.
An actual examination of that argument might prove the opposite - a negative correlation
Nate Silver just called you a Basket of Deplorables.
How much do you lift?
I am a Twitter nobody and proud of it. 👍
Hear hear brother.
Stick to Sabremetrics.
Solidified your persona as an Entertainer, not expert, with that "Twitter nobodies" comment.
Nate, the only reason why you want to debate Nassim is to elevate your brand from 3rd rate pundit to academic. (Without doing the work)
If you have an argument you can just make it here, or write a blog post, as Nassim has done it repeatedly.
Hey, what makes someone a "nobody"?
Says the man who is slowing turning into a @CillizzaCNN-style clickbait pundit
Nate Silver is still upset from his hilarious prediction of 2016 election
Nate Silver's only claim to fame is that he knew Obama was going to win 🤣 gutsy call
Nate you are getting so owned it’s painful to watch , can’t wait for the finale
Hey Nate, what does the statistical model look like for nobodies that fall outside of twitter
Is there any real-world interpretation possible to multiple (non-identical) forecasts of a binary event such as a presidential election, except as a repudiation of all previous forecasts? (And hence, ∀ days prior to the final day, the current forecast is presumed repudiated.)
Just to be sure I'm on the same page with Mr. Taleb's argument, there's a contradiction between the high level of certainty that 538 had months before the election and their huge swings in probability in response to new polls, right?
If he just faked the data starting in mid-September, he'd probably end up being more accurate.
He missed the point: the Silver models have error risk associated with Silver's hand-tuning of the model parameters based on his current intuition. Taleb treats this as the "assessor" in his paper.
Have Nate's model(s) successfully called any election with accuracy??
538 has successfully called many elections because polls tell you a lot. 538 has failed when Silver's political bias fails to accurately assess systemic polling error; spectacularly in the 2016 election (i.e. the shy Trump voter).
538 made this call prior to elections that looked to swing in favour of conventional candidates?
That is not the point. Imagine you make your election forecast by flipping a coin 50 times. Will your forecast be just as accurate than Nate's?
Sadly, the answer is Yes.
That’s blatantly false, though his models do have issues when looking too far ahead (they model “if the election were held today” vs “odds on Election Day”, which causes them to be way too confident months away from the election)
So, at what point should Nate start using the predictive powers of 538?
Should there be a predilection for only periods when available data confirms public sentiment/bias for a candidate?
Wouldn't this render the model ineffective??🤷♂️
Well, what he should do is include the MOE in all his graphs and percentages, which would portray a much more accurate picture (ie, candidate x has a huge lead right now, but since the election is still 3 months away there's a lot of uncertainty)
I think they're pretty accurate when you take their last prediction before an election, but honestly I'm not sure what value they provide, at least with election predictions
I mean other than "oh this is kind of a neat thing that will make almost every layperson mad if someone we gave a 30% chance of winning actually wins"
If you say this, then it means 538 can ascribe a certain percentage point to Trump ahead of 2020.
But we both know this can never happen. If it were, then he'd effectively be the end of Las Vegas...& Macau 🤔
...what?
I mean sure they CAN, but this far out the margin of error would be huge.
As for their sports predictions, if they have an edge on Vegas, one of two things will happen, either Vegas will lose money for a short time and then quickly self-correct...
...or the # of ppl getting an advantage will be small enough as to be meaningless to them
I was a trader for the world's 3rd-largest soccer betting syndicate for the last 5ys. @NateSilver538's soccer predictions were a laughing stock in our firm. He'd be bleeding money if he bet based on the predictions his outfit puts out. That's why he makes his money as a pundit
Thing is, I've never heard of some widely applied use of his predictive models. Kinda like he pops up every 4 years or so. If he were that "good" he'd be in the nine-zero earning range by now...maybe I might be wrong.
He made money playing poker during the poker bubble times, and he probably could have made money betting sports about a decade ago. For current times his modelling is too unsophisticated for betting purposes. Did well for himself as a quack pundit though, he is good at marketing.
Ummm I don’t understand. The Donald winning was a low probability event. He may have had only a 20% if winning and still won. That doesn’t mean the model is inaccurate he just means he got lucky. What’s the problem?
Except it wasn't 20%. And nothing you can offer can prove otherwise, because in hindsight he had a 100% chance of winning and these forecasts simply don't know what variables to track in order to reach that conclusion.
Don’t know and *cant* know that far in advance.
Couldn't you make this same argument about a coin flip? It had a 100% chance of hitting "heads" based on the side it was on when I picked it up and how high I lift it and how hard I flip?
He didn’t have a 100% chance of winning until he won. I’m looking for @nntaleb philosophical viewpoint on frequentism?
It also doesn't mean the model is right. How do you not get that?
This is a totally valid criticism, even if it can be hard to tell what point Taleb is making sometimes (although seeing data from one election is only suggestive). If the day-to-day changes in your estimated prob. are this high, your estimated prob. should be much closer to 50%.
And of course Taleb's point that you could arbitrage the shit out of this is also correct. So yeah, despite all the noise this is a really valid point - and I think show that 538 is not so much excellent at modellng as the only game in town.
Or alternately, they are optimizing for clicks by creating a volatile model.
I dont see how showcasing 3 very good polls of that election is proving a point...against Nate Silver? Like thanks for showing us how his polls were the cream of the crop I guess?
If you want to know who is going to win with highest probability just ask your local cab driver.
Bullshit Nate 😂
You argue in bad faith. You think being curmudgeonly means you have a better chance of being right.. #badstats
So far there is no technical answer from Nate, IN TECHNICAL DOMAIN.
You do deadlifts too!
Ok, you guys have to debate now
Sam Harris will ref
Give it up Nate. It’s over.
Nassim doesn’t even understand the term Bayesian, so...
Only one way to solve this Twitter Battle: Thunderdome
"You know the law: Two men enter, one man leaves."
NATE! NATE! NATE! NATE! NATE!
(No comment)
This is the single greatest statistical burn ever.
Here for the ratio
Bingo
Sorry Nate, I think perhaps I didn't understand what you said. As everyone knows, my goal is furthering knowledge through respectful discussion. Let's try again!
Agree with you @NateSilver538
You guys get a room please.
This should have been booked for #WrestleMania
I wouldn’t worry, Nate. Soon enough, like most others who can’t appreciate his stable genius, you’ll be blocked too.
His sports predictions if bet vs market lines lose even more than money than his newsite
Nassim is so awed by his own genius, he could not foresee any situation he could be wrong.
Backtest this Nate: fivethirtyeight.com/features/berni…. Why someone listens to an economic major who claims to be a statistician is beyond me.
Bernie Sanders Could Win Iowa And New Hampshire. Then Lose Everywhere Else.
Hillary Clinton’s campaign is now telling reporters that she is at risk of losing Iowa to Bernie Sanders in the February caucuses. One ought to view these stori…
fivethirtyeight.com
Very Trumpian, Numb-Nuts Nassim.
Nassim, not sure if it's a publicity thing for you or your actual personality, but man you come off as kind of a jerk. Always interested in smart people sharing insights, but bruh...seriously?
If I understand correctly: Nate is assuming he is sampling an underlying trend, Nassim assumes each observation is a stochastic dynamic that is representative.
No, cher ami.
Explain the role of sampling: I.e the actual versus small sample statistic
Consider an index
This is not sampling of a time series but sampling of a population
Prof... So results and reality are wrong not Silver models 😂
Nate is trying to make the best of small samples of a large population (polls). Nassim is considering volatility of a stock price and its prediction.
Then it is as simple as the mean is not the true mean....Volatility approach is the right away to go about it.
Both of them can go to predictit and create accounts for betting on 2020. The winner of the debate should be the one who has more money predictit.org/markets/detail… @nntaleb @NateSilver538
There's no upside for either of you in these exchanges.
You're half right. 😁
Careful you'll only make his persecution complex worse
Whose? Silver or Taleb's?
I find this amusing.
I always use this phrase when I don't know what I am talking about.
"... epistemological uncertainty"
epistemological - relating to the theory of knowledge, especially with regard to its methods, validity, and scope, and the distinction between justified belief and opinion.
Misconstrues @nntaleb criticisms. Model you use fails to account for epistemic uncertainty of uncountable known and unknown variables leading up to elections rendering your numerical percentage predictions of little meaning. Fair enough for entertainment, but sold as statistics.
can't you say that about any model of any phenomenon?
Problem isn't epistemic uncertainty itself. Problem rather that despite it @NateSilver538's model still produces single neat (but fallacious) "odds"
which ignores uncertainty of aggregated possibilities as if this was linear world without complex second (etc.) order effects.
By second order effects, you mean something like Nate Silver's predictions themselves alter the probabilities of an event happening?
If he says something will happen 70% of the time and that something does happen 70% of the time, I don't see how that has no meaning. You're criticizing the methods which achieve correct results.
Firstly, that's not the case so this is false premise. Secondly, one must prove that the predictions are more than random chance. Thirdly, psychics and astrologers are right sometimes too, but that doesn't make it science as opposed to entertainment.
First do this with election predictions over time-won't look good. Second backtesting can only be used to reject premise not prove it because historical data can't account for future fat-tail events. Third not legit cross-validation (lacks parameter estimation/vulnerability).
Not qualified to contribute to this debate; just an aside: I've always found Taleb's books hilariously self-contradictory. On one hand, they warn about deep and inevitable uncertainties; on the other, he's absurdly certain of opinions he holds about areas he knows little about
Pero no sabe de todo?
Lo sabe todo de todo y el resto de los mortales somos unos "fragilistas"
mmm I read the exact opposite. Only have confidence in things that have been tested over generations e.g. in nature. Otherwise, you need a convex payoff as you will be wrong far more often than you are right i.e. make the payoff worth all the losing bets.
Sure, I remember that from the books. However, I'd argue that we should be very wary about nature-testing being a reliable proxy for goodness
Can you give an example?
I'd need to go back to the examples he gives in the books, but I vaguely remember his giving dietary advice based on that principle, for instance
Med/Jap diet is definitely not harmful. Maybe processed fat/sugar/carb & alcohol is also not harmful and something else is killing obese Americans & Brits, but A) Quite a high bar to proving that and (crucially) B) There is no downside in doing something definitely not harmful.
This seems to be a theme with statisticians. I feel like it deserves a name such as the central-irony-theorem. As you gather more knowledge about uncertainty you converge to certainty about your opinions on everything.
It may happen with statisticians, I don't know, but not with this designer/journalist/whatever: the more I learn the more insecure I feel about almost *any* opinion...
That is the curse of science.
It’s a theme among humans. Maligning a whole spectrum of people who happen to share a career just makes you seem judgemental!
You’re punching down.
Nerd fight!!
The interesting thing about that statement is it implies that you do know what your talking about . The gravity of both statements is weighted by the consequences of your actions: @nntaleb trades and risks $$ @NateSilver538 speculates on elections consequence 0.
Actually Silver's professional reputation is built on his election forecasts. So both are risking $$ though I don't know if Taleb is risking dollars on his predictions for "society", which is what we're talking about.
How has he risked anything in 2016 election forecast . He is still gainfully employed and I think he sold his co for 20 million. As is the punditocracy it is paid to support the status quo view point.
I love your two hate-hate relationship. Best on Twitter
It sounds like Taleb doesn't have any skin in the game, it's easy to play Monday morning quarterback when he doesn't have any elections predictions to judge him by. If we could get his predictions on the upcoming 2020 elections, we'll see who the best man is.
WARNING MR WRONG IS A MONSANTO SHILL
Your inability to keep a level head during this has negatively affected my opinion of your work. You've resorted to as hominems instead of statistical arguments. Im unfollowing and won't be visiting 538 anymore.
I haven't been reading @nntaleb's portion here (he blocked me years ago lol) but it doesn't look like you answered @KaufmanMic's question. When you have a model of "Cavaliers have X% chance to win championship" where X changes over time, how do you bucket that? Do you sample it?
I should have said “grading,” not “backtesting.” My mistake. What I am wondering is whether your results showed increasing overconfidence the earlier the 538 prediction. Believe @nntaleb would posit yes. Did the results agree?
Truth is Nate let personal bias infiltrate the numbers well before the number crunching began.
We often mistake “he has no idea what he is saying” with “he has no idea what I am saying”
Please, both of you do this :)
As far as I can discern amongst the Twitter theater, this whole debate would be avoided if 538 simply added the phrase 'if an election were held today' in fron of their forecast.
Ok i think i understand its not just the uncertainty created by time that at issue, its the unknown unknowns and their unknown behavior that make the 538 bounded model useless. ok to estimate dice rolls or even perhaps "rules bounded" sports/games, but not an election.
No, it can't be an argument about what goes into the model because Taleb doesn't know that. His point is that if the day-to-day changes in the predicted probability are high, then your predicted probability should be close to 50% because of path dependencies.
Ok so the argument is the 538 model is not making a probalistic forecast at any point of time as the variance in forecasts between points of time is so high. So the forecast of 70 % chance on day 1 can not be a true probability at that point of time, if on day 2 it's 60% etc?
Yeah, or not a good forecast - and you could jump from 70% to 10% on some days or during some elections, it's just that if it happens frequently it means the model is bad.
... Because if there is a reasonably high probability you will be at 10% tomorrow then you can't be at 70% today.
Ok thanks got it.
His main point is valid though @NateSilver538 and I think you probably know that. If the day-to-day variability of a model's predicted prob is super high, the model should be predicting something closer to 50% - a path from 70% to 20% to 70% etc must be a low probability event.
I mean...I predicted Trump would win when he announced his candidacy and I'm just some dude withiut access to a polling company.