Dear Quantrix staff,
planning the way you describe is very static. Of course, a kind of dynamics comes into play, once you work with an assumption matrix or you do some sensitivity analysis or what if analysis.
But sometimes a situation is much more dynamic like the current economic sitution clearly shows. Non linear effects, accelerating effects, delays and self enforcing complex cause and effect relationships between not only two or three variables (which are traditionally the limit in a planning matrix), but a complex network of different (planning relevant) variables.
As we all know, this is the original area of tools such as Powersim, Vensim, Stella, Consideo and so on to deal with. These tools are based on System Dynamcis Theory and used for special probelms.
I am discovering more and more interest in the market to get true simulation capabilities in which complex variable networks can be modeled and afterwards be simulated.
Any plans form your side to cover such kind of powerful simulation capabilities in future in Quantrix? Also interesting for Risk Management (based on Monte Carlo simulation functionality).
Hi, I think those tools you mention are rather specialised to a specific problem domain, whereas I would say Quantrix is designed to be a generalist tool (much as any spreadsheet is, though obviously Quantrix is a lot more sophisticated).
However, the multi-dimensional nature of Quantrix and its API does allow it to easily model quite complex situations. For example, a typical model might place its inputs in an “Assumptions” matrix. By adding a “Trial” category to your matrices, and providing sample inputs to the Assumptions matrix along the Trial dimension, you turn any model into a simple Monte Carlo simulation.
Quantrix provides a rand() function as well as a number of inverse CDFs so you can get your inputs through inverse transform sampling. Or, you could add new matrices where you calculate samples from the distribution of your choice. I have done this for actuarial modelling — the only issues are building in dependence (correlation etc.) and speed/memory. Plug-ins can help here, e.g. adding multivariate distributions to obtain copulae, or by taking a different MC approach — instead of a Trials dimension, sample from your PDF in plug-in code, enter the data into the input cells, recalculate the model, then store the output cells of interest.
I found Quantrix useful for building a quick prototype model to check code then written in something more high-performance (Python with the numpy and scipy wrappers around C and Fortran code).