Really Different

What if the future is really different from the past? What does that do to the whole approach of quantitative risk management? When do you give up on the models that just do not help?

Here is a scenario of a Really Different Future

Farrell suggests that the economy will go through chaos for 10 years until things get so bad that we decide to actually do something about it.

We talk about stress testing for our companies.  What I am trying to suggest here is stress testing our risk management approach.

How well does your risk management system hold up to the scenario described in that article?

I am not asking for a top of the head answer.  I am suggesting that you walk through 10 years of economic bad times, alternating with uncertain times like the past 18 months.

Does your ERM system give good advice throughout?  Or do your models continually give bad signals as they very slowly incorporate the emerging world mess?  And then when things come back in 10 years, will the models be wrong again on the low side for another 10 years or more as you incorporate better experience?

So is there another choice?  I think so.  The choice is multiple risk models of different regimes.  You need a model of high volatility and low drift, a model of high drift and low volatility, a model of moderate volatility and moderate drift and a model of negative drift and low volatility.

Think about it.  If those four models reflect states of the world, is there any point in using a model that combines all four sets of experience?  It will always be wrong.  A Bayesian model that is constantly updating for experience assumes a stable underlying distribution.  Otherwise it is just wrong all the time.

Think about it.  If the next 10 years will be years of high volatility and low drift interspersed with periods of negative drift with low volatility, what good is a model with moderate volatility and moderate drift?  Or the combined all regime model of slightly higher volatility with slightly lower drift.

Advertisements
Explore posts in the same categories: Cultural Theory of Risk, Risk Management System

Tags:

You can comment below, or link to this permanent URL from your own site.

2 Comments on “Really Different”

  1. Max Rudolph Says:

    This is another way of measuring exposure. Rather than looking at a zillion scenarios and looking at the average, take several sets of regime scenarios and measure the exposure for each one. It will tell you what type of scenarios will blow you up and will help you make better decisions.

  2. Robert Arvanitis Says:

    That is the classic actuarial trade-off of credibility (sufficient volume of data to be believable) versus relevance (sufficiently recent data to reflect present conditions).

    Very hard to get both. So don’t.

    Proprietary new method – two systems, running in tandem. One with long data (credible but slow), the second with current data (revelant but twitchy). The two systems “vote.”

    Inspired by the early NASA practice of having say 5 computers. In case one or more failed, the “vote” still went to the most likely answer for any calculation.

    Also similar to the brain. Computers currently set one bit at high voltage difference between “0” and “1” in order to avoid erroneous setting if volatage fluctuates. Biological system has several neurons per bit at low voltage, and the signals add (“vote”). That’s how the human brain can be held to tolerable energy consumption.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s


%d bloggers like this: