Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
http://www.stachanov.com Business Software Development and Consulting 2012 Stachanov Amsterdam, The Netherlands
that forces you to add ten minutes to your daily travel routine in order to reach office on time. As long as the traffic is jammed every morning and the extra commuting time always takes ten minutes, then this can hardly be classified as a risk. You know it will be always there and you have taken this into consideration when estimating your travel time. The same goes for the PD. Whenever a default occurs this can hardly come as a surprise to the bank. Upfront the bank took this in consideration when setting the interest rate and creating the provisions capital buffer. Referring back to our traffic jam scenario, the real risk for the commuter is that the delay due to traffic jam varies wildly and as a result forcing him to start much earlier. Similarly, in the case of the bank, the PD percentage itself does not pose a risk, but the variance of the probability of default contains the risk. If the real defaults exceed the expected defaults this may cause the bank to end up in dire straits. The variance in PD forces the bank to keep an additional financial buffer to absorb these unexpected losses. The buffer to protect the financial institute against expected losses is called the economic capital. Before we can proceed with calculating the economic capital, we have to refine our model by introducing the concepts of loss given default and correlations. Loss Given Default Each time a default occurs, the bank does not loose the complete loan amount. Usually, part of the claim can be recovered by, for instance, selling off the collateral that was pledged as part of the agreement. Now, the average percentage that is actually lost is called the Loss Given Default (LGD). The amount set aside for expected losses is reduced by the LGD. This brings us to the following formula for expected losses: EL = loan amount * PD * LGD Correlations A second correction concerns the correlation between losses. Frequently, defaults are not unrelated, independent events. The banking crises show that trouble rarely comes alone, for instance, shift in the housing market leads to a deluge of mortgage defaults. This correlation increases the uncertainty and partly wipes out the portfolio benefits of spreading the default risk over multiple loans. Clearly, these correlations are a major risk driver and should be included in the model. Economic Capital At this time, we take a closer look at the economic capital model for our bank. The bank at hand has a six billion loan portfolio spread over three economic sectors: IT, Telecoms, and Retail. And for each sector we know the PD and LGD. These numbers can be derived from analyzing historical data the bank has access to, from expert opinion, or can be bought from rating agencies.
http://www.stachanov.com Business Software Development and Consulting 2012 Stachanov Amsterdam, The Netherlands
We start off with calculating the provisions for the expected losses, which is a simple multiplication exercise: we take the loan amount (exposure) * PD * LGD.
In the case of the IT-sector the provisions amount to 1.5 billion * 1.5% * 60% = 13.5 million. Repeating this calculation for each of the economic sector, results in a total of 28.5 million in provisions for our bank. Crystal Ball helps out As indicated before, the real risk for the bank is in the deviations from the average, that is, the higher the variance in the PD, the higher the risk. The Crystal Ball software allows us to model the behavior of the PD IT-Sector input variable by associating this parameter with a probability distributions. In the case of a PD, the lognormal distribution is a good choice since we know the average default rate and the standard deviation (variance and standard deviation are related concepts where the standard deviation is the square root of the variance, and the variance equals the standard deviation cubed). The behavior of the PD is also skewed to the right because on the left-hand side zero provides a boundary, while on the right-hand side maxim could go up to 100%.
Similarly, we can model PDs for other industries and use the same approach for LGDs.
http://www.stachanov.com Business Software Development and Consulting 2012 Stachanov Amsterdam, The Netherlands
With the input parameters modeled, we only need to pinpoint the output variable that we like to calculate. In our case this is the total of provisions. Crystal Ball colors the input variables as green and the output variable as sky blue.
Figure 5: Crystal Ball model with input variables in green and output variable in sky blue
Running simulations Now we are all set to start spinning the roulette wheel. Crystal Ball will generate random numbers for each of the input variables. This is a random process that nonetheless is governed by the distributions associated with these input parameters. As the input variables take on different values, the output parameter and the sum of the provisions will change as well. If we run ten thousand trials in a simulation, we will end up with close to ten thousand different results, while our original Excel model just gave one outcome. We can display these ten thousand results in a histogram with the results grouped in bins. This histogram is basically a frequency diagram with the various amounts for the provisions on the horizontal axis and the rate of occurrence on the right-hand vertical axis. It is important to realize that a higher frequency translates into higher bars in the graph and an increased relative probability.
http://www.stachanov.com Business Software Development and Consulting 2012 Stachanov Amsterdam, The Netherlands
Figure 6: Ten thousand simulation results grouped in bins with the original Excel result as base case
Looking at the histogram, we notice that the majority of the observations stay close to the original Excel amount of 28.5 million. Still, the results are scattered over a wide range which indicates that there is a fair amount of risk. Ideally, the results are clustered in the middle: the more spread out the graph is, the higher the risk. Often, the standard deviation is used as a proxy for risk. Crystal Ball tells us that the standard deviation is close to three million.
Figure 7: Standard deviation of three million below and above the mean
Required Economic Capital Let us go back to our original question, that is, how much economic capital this bank should maintain in order to fend off unexpected losses (UL) above the provisions it has set aside to protect itself against the expected losses (EL). Basically, the unexpected losses are occurrences to the right of the mean in the forecast histogram. It goes without saying that if we increase the amount of provisions, more of the possible variance in the outcomes will be covered. If we, for example, increase the provisions from 28.5 to 30 million, we cover over 70% of the surface of the graph and hence of the risk.
http://www.stachanov.com Business Software Development and Consulting 2012 Stachanov Amsterdam, The Netherlands
In this Crystal Ball forecast chart, area in blue is the risk covered and red is the uncovered risk. Risk appetite Now we get to the crux of our model. The economic capital is the difference between the provisions depicted as the base case in the graph and the cut-off point on the horizontal axis for a certain risk level. In our example, the risk covered is 70.73% which corresponds with a cut-off point of 30 million. Since, of these 30 million, 28.5 million were already accounted for in the provisions, this will leave 1.5 million for the economic capital to serve as a buffer for unexpected losses. If we demand more certainty and less risk, we should allocate more economic capital and as a result more of the graphs surface will be in blue. As anticipated with our traffic jam analogy, if the traffic jam duration varies wildly, we have to advance the time we leave the house. How much earlier, depends on the degree of variance of the length of the traffic jam. And, let us not forget as to how important it is for us to be on time. If one runs the risk of losing his job upon reporting in too late for work, wider buffers are called for. Basically, this boils down to the risk one is willing to stomach: the risk appetite. Credit ratings Returning to the bank scenario, the risk appetite is determined by the credit rating of the bank. A triple-A bank will maintain higher capital buffers than a single-A bank. These credit ratings correspond to certainty levels in the Crystal Ball forecast chart as shown in Figure 8, where a higher certainty level calls for more capital. In reality, most banks do not allow for more than a couple of basis points of uncovered risk, which takes the certainty levels for the most common credit ratings over 99%
http://www.stachanov.com Business Software Development and Consulting 2012 Stachanov Amsterdam, The Netherlands
Figure 9: Economic capital for banks with ratings from AAA through CCC
Here we have our answer: the economic capital needed depends on the risk appetite of the bank. In our example, a triple-A bank would need about 13.6 million, while a triple-B bank would get away with just 9.9 million as economic capital. Besides the size of the loan portfolio and the credit rating of the bank itself, one of the main motors of economic capital is the uncertainty surrounding the unexpected losses. We could ask the question as to what causes this uncertainty. Crystal Balls sensitivity analysis identifies the main risk drivers and ranks them by their contribution to the variance of the provisions. Sensitivity analysis is a useful tool in risk mitigation and provides a handy todo-list for further action.
There is one more thing to add to the model: correlation. Crystal Ball allows for the input of correlations on a variable per variable level, but is also capable of reading a correlation matrix.
http://www.stachanov.com Business Software Development and Consulting 2012 Stachanov Amsterdam, The Netherlands
In the economic capital calculation, correlation has an impact on the risk level of the bank and leads to considerable higher levels of economic capital.
In the case of a triple-A bank, we notice that the economic capital jumps from 13.6 million to almost 21 million due to correlation effects. Conclusion We can state that the Crystal Ball is a useful and a handy tool for the modeling of capital adequacy calculations. Ease of use and the interaction with Excel indicate that the banks decision makers can take on more of the model creation themselves and break the artificial barriers between the quants and management. Amsterdam, Jan 23rd, 2012 Stachanov Solutions & Services b.v. Andr Koch Stachanov Group, Amsterdam T +31 20 509 1010 E info@stachanov.com
http://www.stachanov.com Business Software Development and Consulting 2012 Stachanov Amsterdam, The Netherlands
About Stachanov Group Comparatively small, but very fine and well-respected engineering company is serving AAA-customers since 1997. Focus on large database projects for the public sector (e.g. UWV, Ontwikkelingsbedrijf Rotterdam), and (risk-) modelling projects for both public (e.g. Havenbedrijf Rotterdam) and private sectors (e.g. all Dutch banks). Most work is on project basis, but we also have hosting customers (e.g. Euronext). ISO-certified since 2004, we embrace: Innovation, Professionalism & Operational Excellence.
http://www.stachanov.com Business Software Development and Consulting 2012 Stachanov Amsterdam, The Netherlands
10