Forums  > Software  > VAR and ES computation via Monte Carlo  
     
Page 1 of 1
Display using:  

Maggette


Total Posts: 1233
Joined: Jun 2007
 
Posted: 2019-12-20 17:22
Just wanted to poll here:
=> is/was lack of speed/long runtime of the calculation of risk measures a problem in your shop?
=> Did you calculate intraday?
=> And if so, is/was it really the simualtion and not the data processing before the simualtion (this I my experience)?
=> how many did work with GPUs?


Thanks


Ich kam hierher und sah dich und deine Leute lächeln, und sagte mir: Maggette, scheiss auf den small talk, lass lieber deine Fäuste sprechen...

nikol


Total Posts: 1126
Joined: Jun 2005
 
Posted: 2019-12-20 18:57
I'm from risk environment, hence, eod processing with minutes calibration time for arma correlated on top of non-gaussian garch11 errors over ~500 risk factors isn't an issue. Calculation of 20k trials takes x10 times more. I'm talking about Matlab powered with 16gb proc. But I'm afraid we compare apples and potatoes.

Hence, specify requirements
- Horizon to get estimation for risk
- Refresh frequency
- number of risk factors
- complexity of models

Use parallel pipelines/data bufferization: while 1st generates trials, the 2d calibrates, the 3d preprocesses data etc.
Use dimentionality reduction techniques. Residuals can be modelled separately as addons.
Preprocessing (cleaning, scenario conversion) can be done in one go (popup idea - is there demand for such "box" if done efficiently?)

Maggette


Total Posts: 1233
Joined: Jun 2007
 
Posted: 2019-12-20 19:27
Hi,
Thanks for the data point.
This is exactly what I am used too and what I am expecting.

I am exactly interested in apples and oranges and what other fruits are around. I am interested in what use case scenarios are out there
Complexity, frequency, number of risk factors.... And if runtinme was considered a problem, and if so, was the MC part the problem.

I experienced a spectrum of cumbersome overnight runs because of bad architecture, lots of legacy systems preparing data, and implementation to relative effective implementations.



Ich kam hierher und sah dich und deine Leute lächeln, und sagte mir: Maggette, scheiss auf den small talk, lass lieber deine Fäuste sprechen...

nikol


Total Posts: 1126
Joined: Jun 2005
 
Posted: 2019-12-20 19:44
Compaire sequential operational chain and parallel.
It is designed parallel for checks (double checks) on top of sequential execution.
You get my point? ;)
Previous Thread :: Next Thread 
Page 1 of 1