What's Hot

"Risk Dashboards should serve the stakeholder" | Advanced Risk Dashboards

Thursday, May 15, 2014

Big challenges in risk

Only the other day, someone asked me what are presently the biggest challenges in risk management and while risk culture is probably always going to feature, Risk Information Technology or Risk IT has a lot of issues. When I refer to Risk IT here, this has nothing to do with computers or network connectivity but instead the analytics on these machines. Interestingly, what holds the risk community back is not the technology aspect per se but in some respects narrow mindsets. Let's take a look ...

What holds us back?
There is an old saying and I am nearly sure that any risk manager that has been around for a while has probably heard this, well at least once: "If you want to manage something like risk, you need to scope it and evaluate it first. So be it, but to evaluate it will inevitably lead you down the path of measurement." ~ This all sounds obvious enough, so what is the problem?

There are a couple of roadblocks we hit which are people related more so than technology, although it hasn't always been this way. We only have to stretch back a decade or so ago and it was very difficult, complex and expensive to enable risk models at an applied level on an electronic device. I am not talking about theory here, no formulas specifically but applied, usable risk technology, extensible and configurable yet ready to go. This constraint has dissipated somewhat today not because processing power has improved but due to the rich enablement of risk theory in software.

Risk IT Discussion | In the workshop

We also need to fix the people problem which seems to come in two pervasive flavors. The first hurdle are those that detest risk models period and there are a lot of these type of 'risk-beings' out there. They kind of talk the talk, well some of them do but that is about as far as they can go.

Surprisingly, I do have a lot of sympathy for risk managers that won't or can't model, especially when I cast my thoughts to the years gone past. I have vivid memories of sitting numb in a classroom, gazing at a blackboard (there were no such things as whiteboards when I went to school) full of calculus while the teacher rambled on and the long hot summer afternoons slowly passed by. Nonetheless, if you end up in this 'profession' as I have, one needs to put their game face on and skill up. Failing to do so just leaves you handicapped when you need to measure risk and then you are behind the eight ball when you go about managing it.

The second kind of risk analyst out there is just that, they are complete model junkies, hardcore geeks that try to model everything and they are equally dangerous to the fraternity of risk management because they often fail to appreciate the fallibility within their models. All models have scope limitations and all models generate results with at least standard error.

Worse, measured risk should not result in a deterministic number but a range or trajectory of positions with a set of confidence levels and as obvious as this is, few rarely perceive risk to be evaluated in this manner. To the non-modeler or novice, risk is often depicted as high, medium, low. In the world of risk pricing, risk has a specific currency value or a premium if you prefer but whichever method you choose, both these perspectives of risk are deeply misleading and consequently feeble measures of exposure.

To save us from this rock and a hard place this industry desperately needs to invest in Risk IT people and lots of them. Risk is akin to engineering, some see it as a science but we can nearly be sure that it isn't a pure art. Either way, we have too many risk managers that just don't do models and a handful of quants that talk in actuary tongues and never the twain shall meet.

Model Enabled
If the risk community is on-board, it's going to take time for a good idea to catch on but if we want to coherently measure risk, we need to accept that Microsoft Excel is probably never going to cut it. Spreadsheets are great, I use them often but they aren't the right environment for administering risk or statistical data because they are designed to work with formulas around data. In the short of it, they just don't treat data as vectors that can be lifted easily through functional models and they are slow with iterative calculations. More critically, stock standard Microsoft Excel as a model enabled environment is just too fundamentally basic and that leaves a lot of model coding work to churn through.

Whichever path you take, R always seems to win | In the workshop

Some risk managers simply bite the bullet as the saying goes and spend big to buy directly off the shelf. These pre-programmed risk tools have improved modeling functionality these days and there are so many products out there to choose from. A cursory glance at Bobsguide [LINK] will definitely confirm this and it is probably one of the most comprehensive websites tracking risk management systems today.

Another alternative would be to chance a tool designed uniquely for the job and the two ideal candidates that come to mind which really nail this space are Python [LINK] and R-Project [LINK]. Personally I favor the latter, R-Project and for many reasons which I will list below.

[1] The R-Project Community
The R-Project modeling community is broad, diverse, active and mature. There are stacks of fanatics pulling models together and sharing them without limitation. This community can certainly bring risk analysts up to speed quickly on the R-Project modeling environment and I recommend taking a look at the R-Bloggers forum [LINK].

[2] The Library Set
The next point is important. Designing, programming and testing risk models is time consuming, and risk analysts are often stretched across a lot of different business agendas. However, R-Project comes with an extensive set of libraries that are pre-designed and ready to go. Everything from portfolio analytics and Value at Risk to Bayesian trees has been enabled in R-Project through various additional packages. In fact there are over five thousand libraries available for you to choose from [LINK], go knock yourself out with extensible risk models. 

[3] Connectivity
Risk modelling is generally predicated on data and often there is a lot of it to administer. R-Project is able to load data from a huge array of sources including CSV files, ODBC data sources, the windows clipboard, you can connect R-Project up to a Bloomberg account, you can even hook up R to one of the humongous free data repositories out there and for those interested; try taking a peep at Quandl [LINK], you'll be surprised with what you can find in Quandl.

[4Operating System Compatibility
R-Project is available on Unix systems, Mac OS and pretty much all versions of Microsoft Windows. Additionally, the system inter-operates on all these different computing platforms in an identical manner.  There is even a version of R-Project available for those analysts using nothing more than a web browser and I have run R-Project on an ipad through the StatAce provider [LINK].

[5] The tool is free !!!
For some people money talks but when you can use R-Project for free, it's going to be pretty hard to turn it down ~ that is unless you don't do modeling. Something free, so powerful is an absolute must for risk analysts and if you are that keen, you can even download the source code for the software as well. There will always be those risk analysts that can't model, won't model or don't want to learn but they probably stopped reading this piece somewhere around the second paragraph.

Okay then, R-IT for Risk IT and in the next blog I'll take a look at some of the really neat risk libraries that have been written specifically for R-Project.

No comments:

Post a Comment