Loki-can you make sense of this

[DemocratSinceBirth]

https://twitter.com/nntaleb/status/771372006914846720

[Loki Liesmith]

1. For the future: Instructions for summoning Loki

1) Tell the biggest lie you can
2) Pick up a hammer and start verbally abusing it.


In any event, yes, I can make SOME sense of it. I know enough stochastic calculus to understand what Taleb is doing. I am by no means an expert. Now information theory...well...we can do that later.


He is arguing that the mathematical underpinnings of guessing the future state of a binary variable is not a stochastic process as Nate (And everyone else) models it, where

Probability(time t + 1) = K*Factors + Probability(t).

where K is a matrix of coefficients and Factors is polls and whatever else you throw into the model.

Instead he's arguing that if Nate's measure really measured probability, it would look different (like the blue line in his plot).

So whatever Nate is modeling, it's not probability of a win. It's something else.

Very interesting. Taleb is super smart. Also an asshole, but still super smart.

[Loki Liesmith]

2. What Nate is measuring is his model's confidence

Not really probability. I have agonized about this. What does it mean to say that the probability of a win today is 56%? How can we test that? We can't. IF the model calls hundreds of elections correctly then we can say the model is good. But we still can't test the instantaneous probability of a win at the moment the model evaluates the data.

[MyNameIsKhan]

3. He is basically trashing Nates formula of aggregating polls + weights

In simple terms, He is saying this becomes a simple Heat transfer equation, how to measure heat as it travels via any media, which finally has logarithmic error function... Combining heat from various sources without ensuring all sources provide same amount will result in invariance. Nate is already correcting his model by providing weights so all sources are equal...

I can look more tomorrow, late right now and answer.

[Imperialism Inc.]

4. Basically he is saying that one's prediction

should hover around 50-50 right up until the actual election. The reason being that polls can go back and forth. So basically he's saying polls are meaningless.

That's not how actual voters behave though and empirically his graph is way off. Put it this way, if elections were held more regularly you could make a ton of money betting against Taleb's model. Every September if one candidate's chances, based on the current polls, got to about 66%, based on any of the models out there, get Taleb to give you an even money bet (when he should be asking for 2 to 1 odds). You will make out like a bandit because it is an empirical fact September polls are predictive.