Are simulated probabilities a meme?
In particular, if we want to calculate the probability that that price will be equal or lower than today’s (Jan 20th 2018), we’d just have to integrate the shadowed area in the following plot:
And what’s that probability? 9.84%.
I concur
Ah yes
We are either going to be poor or rich
Great insight
that's just mcmc, it's a bayesian statistical technique. like all statistical techniques, the results are only as good as the model itself.
without clear published backtesting data, at least one day lookahead data squared, the model could be wrong as fuck and they're just trying to convince you that the "simulatedness" of the model makes it righter than a "nonsimulated" model.
basically- just because it uses mcmc doesn't mean it's right.
looks like they were all wrong Top fucking kek
I like the one that goes to 1 million. Mhmm yea
No actually its more nuanced than that. Depending on how I behave in a system which exhibits the statistics properties descibed by the model I may do better or worst.
it's not MCMC. there is no markov chain in there.
Just a bunch of lines.
If there was one that resembled 2018 maybe id be interested
>random walk on a log scale
what the literal fuck
The yellow one near the bottom is kinda close
These simulations are used in Finance to estimate the price of an option.
Using it to predict the future is the dumbest thing ever.
That isn't even remotely an McMc.
Don't use phrases you don't understand.
This is retarded, in the past years bitcoin was trending upwards on average pretty steadily (yes there were drops/crashes but if you take a random candle from anywhere between 2010 to 2018 it's pretty likely that you hit a green one).
If you now produce a random walk from these samples it's clear that the random walk also will be trending upwards in the long term.
HOLY SHIT
This is a fucking joke, you can only predict future movements using IV, and there is no volume on btc options.
please explain topic or at least post introductory link, senpai
"An Introduction to MCMC for Machine Learning" by andrieu et al
tldr: you can use the law of large numbers to approximate expectations (for example the mean or some probability). people have invented methods to efficiently generate samples that allow you to do the same approximations as sampling directly from the true distribution can be difficult. mcmc is one of those methods where you start with one sample and then iteratively generate a new sample based on previous one, that is, you use a markov chain to generate the samples. what OP does is estimating the distribution of the (log?) returns and then sampling each sample from that estimated distribution. It is actually a special case of mcmc, but calling it mcmc would be quite misleading as he doesn't use the previous sample at all.
thanks, man. appreciate it.
yes
>Just a bunch of lines.
this
yes, stochastic processing is a meme