3
$\begingroup$

Is there any known probability distribution that behaves like a rock climber doing lead belay climbing (where they're clipping the rope in at various points as the climb up a rock) - where there are usually small gains (like how a rock climber usually makes slow, steady progress up a rock), but less frequently, there are really big drops (like when a rock climber falls)? So maybe it would look something like this:

rock climbing distribution

(So usually things go pretty good, but when they go bad, they are really bad. This would also be a good model for the modeling markets prone to developing bubbles).

I suspect that such a "distribution" would be a good model for things like the "capacity utilization" in this graph:

capacity utilization graph

(Note how typically, there are small gains, but periodically, there are big drops)

So my question is: is there a known distribution that would model things like this already?

The poisson reminds me of it, but is a bit different. Perhaps, the answer is a model which is 2 separate normal distributions - one for normal small gains and one for massive losses (and of course, some probability distribution that defines which of the two is drawn from in a given sample).

Context:

Recently, I was modeling some mutual funds with a Normal distribution, and a thought occurred to me: digesting historic returns of a mutual fund and outputting 2 parameters, mean and standard deviation, is a pretty big information loss. Obviously, trying to build a model is a "lossy" "compression" technique, but I'm concerned there could be some valuable information being lost in a Normal Distribution model we could maintain in a better model like the one I describe above.

$\endgroup$
2
$\begingroup$

You seem to have probability distribution functions (PDFs) confused with time series. A PDF has integral 1, so it can't have arbitrarily high bumps arbitrarily far along the real line. A time series is typically modeled not as a single random variable (and hence a single distribution) but as a function mapping from time to random variables; that is, a random process.

$\endgroup$
  • 1
    $\begingroup$ I disagree about the confusion. Given bounded falls, bounded distances out the real line, I can easily imagine a proper PDF for this, but I do not know what this would be. I could simulate it given the probability that the next move was a fall, or a rise, and some distribution for their respective magnitudes. There is lots of work on mixture distributions, but I know little more than that. $\endgroup$ – astaines Sep 27 '16 at 10:07
  • $\begingroup$ Sorry, I should have labeled my axes, but the idea is this is a PDF, with an integral of 1, representing the probability of getting each specific return rate (in this case, as a percentage). $\endgroup$ – caleb Sep 27 '16 at 14:00
  • $\begingroup$ @caleb I see. In that case, you probably want to construct a mixture distribution, such as a mixture of two normal distributions. $\endgroup$ – Kodiologist Sep 27 '16 at 15:34

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.