Basic Problem
When a vital amount of things responds to the same trigger event, those things have to be seen as cluster.
From a risk perspective, clusters are quite nasty as they tend to show an incomprehensive and abrupt behaviour leading to unforeseen and sometimes lethal risk exposures.
The reason is that with pulling the right trigger each of the responses are not independent form each other anymore. And, dependent events show a quite different (risk) behaviour than independent ones do.
Hence, the identification of cluster risks is crucial in risk management operations.
Unfortunately, clusters do not have those clear boundaries in order to make it easy for them to be spotted.
Traditional risk approaches in tagging cluster risks according to e.g. industry, asset class or single markets might fail to comprehend the whole risk picture and may fall short of their purpose.
Data driven risk solutions are able to introduce a more sophisticated risk management approach in order to deal with this problem.
In this we can use a quite wide range of tools, such as statistical modelling, machine learning techniques and network analytics.
But, for the moment we look at the effects of cluster risk in commercial real estate investments.
Cluster Risk - a Basic Example
A loan portfolio is calibrated at a minimum margin level so that the risk of a negative margin contribution is a mere 1 %. The margin level is based on an assumed default ratio derived from earlier default experience. The loan portfolio has therefore the following risk profile:
Though, when cluster risk gets its grip on the portfolio the risk profile changes dramatically. Simulating the unleashing of cluster risk while keeping the structure of the loan portfolio the same shows a significant increase in the risk profile of the portfolio:
In short, the risk of a negative margin contribution rose from 1 % to almost 20 %!!!
How did this work?
Well, the margin calibration was done with a certain default rate assuming that single default cases in the portfolio are independent.
This assumption does not hold in clusters respectively it does not hold in situations where the trigger event (e.g. an extraordinary economic crisis) is so harsh that single default events start to impact one another.
The result is a dramatically worsening risk scenario within a short period of time.
Real Estate Investments and Regional Clusters
In real estate, a way to avoid cluster risk or at least to reduce its impact is to diversify into different real estate asset classes and/ or different real estate markets. Though, sometimes this may not be sufficient enough as our example of market diversification shows.
Suppose, an investor ahs a real estate portfolio comprising of the same asset class but being located in two different markets. In other words, market diversification is the strategy.
However, those two markets show the following empiric development in terms of investment yield:
So far, it seems obvious that both markets are strongly associated and therefore tend to move in the same direction. A market diversification strategy involving both markets might not be such a good idea as exactly those markets seem to build up a regional cluster.
Incorporating this regional cluster into the risk evaluation of the portfolio, possible investment yield scenarios for a 3-years investment period are simulated. Here is a sample of simulated yield scenarios for each of the markets:
At the end of the 3-years term, our investor faces the following possible yield combinations from both markets:
Also the risk simulation reveals the association between both markets. Given current market conditions in those markets (the dotted black lines), the existence of a regional cluster causes a 20 % risk that the real estate portfolio faces a downturn in both markets at the same time.
In terms of investment yield, this is the danger zone (see the red area in the graph) where values really go south.
With respect to market values, this can mean a deterioration between 7 % and even up to 20 % in the overall portfolio which -of course- can have a huge impact on the portfolio performance.
Conclusion
As shown, cluster risk is definitely an issue to be identified and handled with.
Problem is that cluster risk is not always obvious or shows up temporarily triggered by extraordinary events.
Traditional risk management approaches are weak in detecting the hidden patterns of clusters and do not always reflect the consequences of clusters in action.
More sophisticated risk techniques from the area of advanced statistical modelling, machine learning or network analytics offer a great deal to tackle those crucial problems.
For more info on using data driven financial risk solutions please visit d-darks.com.