Random variables number obey a normal distribution

A

Andrzej

Joined
5/11/16
Messages
3
Points
11
I was asked about generating a random number normally distributed based on uniform distribution. I told about Box-Muller transform, inverse CDF, Central Limit Theorem... So my question is, which methods is widely used by the Quants and which one is only theoretical curiosity?
 
At the university we studied Box-Muller.
As I started my career one gave me a task to rewrite an implementation of Box-Muller from C++ to Java; it was a test on a Schnuppertag (trainee test day).

Generally, quants just blindly use reputable out-of-box libraries, unless pretty the performance is really critical.
 
yetanotherquant said:
At the university we studied Box-Muller.
As I started my career one gave me a task to rewrite an implementation of Box-Muller from C++ to Java; it was a test on a Schnuppertag (trainee test day).

Generally, quants just blindly use reputable out-of-box libraries, unless pretty the performance is really critical.
Click to expand...

That's about 3 lines of code... 5 minutes work :D

BTW, did you stumble on the nasty Neave side effects with Box-Muller? It's not very good. It's a show-stopper. Besides, a more efficient variant is PolarMarsaglia (no trigonometric functions).

And it's not true that quants just blindly use rngs. That's a simplification and one helluva generalization.

here is one example
sitmo | High Quality C++ Parallel Random Number Generator
 
Last edited:
Daniel Duffy said:
That's about 3 lines of code... 5 minutes work :D
Click to expand...
As a fresh graduate I needed a little bit more, but probably they wanted to check whether there are programming skills at all (at the Uni-Ulm the curriculum for Master Students of quantitative Finance does not contain any programming courses but clever students like me attended additional courses; in Germany you can take as many courses as you want free of charge).

Daniel Duffy said:
And it's not true that quants just blindly use rngs
Click to expand...
It is true in the majority of use cases. Moreover, the modeling quants often blindly use the whole software (and not just small components like RND), which is written by in-house IT or third-party providers.

Daniel Duffy said:
Just out of curiosity .. is this the standard RNG in Java or are there also specialised libraries?
Click to expand...
Have a look at what they use at OpenGamma.
 
Ok, I understand. But, I am still wonder which one is commonly used in a build-in software components. After quick glace of C++ environment, I often meet with inverse CDF...
 
Andrzej said:
Ok, I understand. But, I am still wonder which one is commonly used in a build-in software components. After quick glace of C++ environment, I often meet with inverse CDF...
Click to expand...
C++11 <random> and Boost C++ Math Toolkit (see quantile function).

There are several kinds of algorithms for computing CDF. Glasserman's book on Monte Carlo.

===
Nice link to the quantile

Non-Member Properties - 1.60.0
 
Last edited:
Back
Top Bottom