Wehrspohn Risk Management

How does risk aggregation work under IDW PS 340 and StaRUG?

Case study

By Dr. Uwe Wehrspohn.

Since the changes to IDW PS 340 and the introduction of StaRUG, the focus is no longer only on individual risks that threaten a company’s existence, but also on small risks that add up to a threat to the company. Using a risk model in Excel, we show how to assess risks gross and net, aggregate them with a Monte Carlo simulation, compare them with the risk-bearing capacity and prepare the results for risk communication.

Companies in Germany have long been obliged to identify and monitor risks that could endanger their continued existence. With the changes to the auditing standard IDW PS 340 and with StaRUG, one question moves to the centre: can many small risks together put the company at risk?

To answer this question, the risks have to be aggregated, that is, combined into an overall risk. Below we show what IDW PS 340 and StaRUG change. Then we go through a risk model in Excel step by step. It builds on risk identification, assesses and aggregates the risks, determines the risk-bearing capacity and management signals, and provides the basis for good risk communication.

Why do small risks now matter too?

As early as 1998, the KonTraG, the German law on control and transparency in business (Gesetz zur Kontrolle und Transparenz im Unternehmensbereich), introduced comparable rules for stock corporations and large limited liability companies (GmbHs). However, experience with this law was not encouraging. Every year, tens of thousands of firms became insolvent, and in hindsight this was often connected with risks that had materialised. Beforehand, however, a risk threatening the company’s existence was almost never documented and reported in these companies.

The Institute of Public Auditors in Germany (IDW) therefore decided to extend the audit of risks that threaten a company’s existence. It now also covers the case in which small risks, none of which is a threat on its own, accumulate and in sum threaten the company.

Since 1 January 2021, StaRUG, the German Act on the Stabilisation and Restructuring Framework for Businesses, has also applied. Under section 1 StaRUG, the management of a legal entity, such as a GmbH or a stock corporation, must continuously monitor developments that could endanger the company’s continued existence. If they identify such developments, they must take suitable countermeasures and report to the supervisory bodies, for example the supervisory board, without delay. The same applies to partnerships in which no natural person has unlimited liability, such as a GmbH & Co. KG. There is no size threshold: the duty of early crisis detection therefore no longer applies only to large companies but to every company with limited liability. Here, too, such a development can arise from many small risks that only become dangerous together.

What changes do IDW PS 340 and StaRUG bring?

Figure 1 summarises the changes introduced by IDW PS 340 and StaRUG. At the top is the goal: the early identification of developments that threaten the company’s existence. Below it are the steps in which this goal is achieved, from risk identification to documentation. We go through them in order.

Figure 1: The changes introduced by IDW PS 340 and StaRUG at a glance. At the top the goal, below it seven steps from risk identification to documentation (German slide).

Risk identification: a complete risk inventory

The first important step is, of course, risk identification. It produces a risk inventory, that is, a list of the company’s risks. What is new is that the risk inventory has to be as complete and holistic as possible. If only a limited set of risks enters the accumulation and aggregation, it is not easily possible to make a credible statement about the company’s overall risk.

Assessment: gross and net

In the second step, the identified risks are assessed, once without and once with the countermeasures already in place. This is called gross and net assessment: gross means before measures, net means after measures. The assessment is usually split into two sides. On the occurrence side, a probability or a frequency of occurrence is determined. On the impact side, a range of losses is determined that can arise if the risk occurs.

Risk aggregation: how do you represent randomness?

Next comes risk aggregation, that is, the analysis of how the risks accumulate into an overall risk. In practice, this is a very difficult step for many companies because it requires new methods. Risks occur randomly and therefore also overlap randomly. This raises the question of how to represent and analyse this randomness. The risk model we present below answers it with a Monte Carlo simulation.

Risk-bearing capacity: risk versus risk coverage potential

Once the risks have been aggregated, you can ask about the risk-bearing capacity, that is, whether the company can bear its risk. To do this, the risk found in the company is compared with a risk coverage potential. The risk coverage potential indicates which losses the company can still bear before it can no longer be continued in its current form.

Risk management, risk communication and documentation

Risk management uses the same information that was already used for aggregation and risk-bearing capacity. Management signals are derived from the results to improve the company’s risk situation and, as a consequence, to secure the risk-bearing capacity all the better.

Risk communication is about formulating decision-relevant information in understandable language and feeding it into the company’s reporting lines. That way, all decision makers know about the important matters. Finally, the whole process should be documented and traceable so that it works independently of individual people, for example even when someone leaves the company.

How is the risk model in Excel structured?

In short: Risk Kit is an add-in for Microsoft Excel that lets you build and analyse Monte Carlo simulations directly in the spreadsheet. What we show here applies in general, though: aggregating risks with a Monte Carlo simulation works on the same principle in any simulation tool.

The model we use here has shipped with Risk Kit since version 7.9. You open it in Excel from the Risk Kit toolbar. Under “Beispiele” (examples), the section “Case studies” contains the entry “Enterprise Risk Model”, as Figure 2 shows. The same menu also contains the example “Risk Map” under “Diagrams”, which we come back to when we discuss risk maps.

Figure 2: The “Beispiele” (examples) menu on the Risk Kit toolbar. At the bottom, under “Case studies”, is the “Enterprise Risk Model” (German interface).

The workbook consists of three worksheets. The “Intro” sheet is an introduction with comments on what the model does and which results it produces. The “Modell” sheet contains the actual model. The “Labels” sheet is a helper sheet with labels, for example for the risk maps.

The most interesting part is the model. It also consists of three parts. At the top are the input data with the information on the risks. Below them are the results at the level of the entire risk inventory, once as charts and once as statistics. At the very bottom is a drill-down by risk field, that is, an intermediate aggregate between the individual risk and the entire risk inventory.

When you open the model, every cell that calculates a statistic initially shows the message “Erst nach der Simulation definiert” (defined only after the simulation). This is because Risk Kit evaluates a simulated distribution to calculate the statistics, and that distribution does not exist before the simulation. We therefore run the simulation once so that all results are shown as numbers.

How do you assess individual risks?

Figure 3 shows the upper area of the “Modell” sheet with the input data. It is delimited by the blue bar with the column headers in row 7 and the light yellow and white striped area below it. Each row describes one risk. Rows 8 to 17 contain ten example risks, the rows below are still empty.

Figure 3: The input data in the “Modell” sheet. On the left the master data of the risks, in the middle the assessment before measures, on the right the assessment after measures and the reputational damage (German interface).

Master data: ID, name and risk field

Each risk has an ID in column A as a short label, which is also used in the charts. Column B contains the name of the risk in plain text so that we as humans know what the risk is about. Column C assigns the risk to a risk field, for example market risk, climate risk or operational risk.

A risk field typically groups risks that are assessed and managed in the company with very similar methods. Separating the risk fields therefore has organisational advantages. In addition, we can form an intermediate aggregate for each risk field and thus determine its contribution to the overall risk. This shows us the importance of this type of risk and may allow us to take specific measures.

ID, name and risk field are master data, so to speak. They enter the analysis only as structural information and do not affect the size of the risk.

Gross and net: before and after measures

The quantification of the risk consists of two blocks, once before measures in columns E to H and once after measures in columns I to M. This corresponds to the gross and net assessment. The input data in both blocks are structurally the same. From a business perspective, the only addition after measures is the cost of the measures.

Once or several times?

The first important piece of information for quantification is in column D: can a risk occur only once in a period or several times? The classic approach is that a risk can occur only once. A more modern approach allows one and the same risk to happen more often. In many situations this is simply the case, for example because the company may grow and the frequency of a risk depends only on the size of the company. Or general environmental variables make it clear that a risk is not limited to a single occurrence.

One example is the flooding of the production site by heavy rain or a river flood. It can certainly happen twice in one period. If a flood has happened once, that does not rule out that it happens again. In our example, the same applies to a power outage, to legal disputes and to the failure of projects. For these risks, column D therefore contains a yes.

Conversely, there are risks that can occur only once in a period, for example a macroeconomic event such as the onset of a recession. For operational risks that are formulated very specifically, it is often equally clear that they can occur only once. That a particular bridge collapses can certainly happen only once. That one of hundreds of servers fails, on the other hand, is a generally formulated operational risk that can occur several times. This decision is therefore important to keep the representation of risks realistic.

Probability or frequency of occurrence

Depending on whether a risk can occur once or several times, its occurrence is described differently in column E. If it can occur only once, column E contains a probability of occurrence, for example 0.25 for the recession in row 8. If it can occur several times, column E contains the expected frequency of occurrence, that is, how often the risk occurs on average in one period. For the burden of a small legal dispute in row 15, this is 1.

The impact: minimum, most likely and maximum loss

The three columns F, G and H with the minimum, the most likely and the maximum loss describe the impact side of the risk. They apply to each individual occurrence. For the data protection breach in row 14, for example, the loss per occurrence lies between €100,000 and €750,000, with €250,000 being the most likely.

This representation is a very classic way of modelling. It still follows the scenario analyses that were carried out in the time before Monte Carlo simulations. Risk Kit has several distributions that are described by these three values. They include, of course, the famous triangular distribution, which certainly plays the most important role in industry as a model for the impact of risks, but also the PERT distribution and the modified PERT distribution. The smallest and the largest possible loss delimit the range, and the most likely loss is added.

Applying this modelling to all risks, as we do here, is a simplification. You can generalise it by using a separate distribution for each risk, which then of course also needs other parameters. This is definitely advisable in practice, because for many risks the maximum loss is not easily known. The consequences of a power outage, for example, can be enormous, much higher than in the scenario one had in mind. The possibility of a catastrophic loss is extremely hard to estimate. In practice, it should perhaps be represented by a distribution that is unbounded above. This is easily possible with Risk Kit.

Our example is the entry-level model, so to speak. In industry, many companies use it exclusively in this form: all risks are modelled with the same distribution.

The measures and their costs

The block after measures has exactly the same structure on the occurrence side and for the impacts. In addition, column I contains the fixed total cost of the measures, which is incurred when we take the measures. With this information we can look at the interaction between the cost of measures and the cost of risk. We can ask when implementing measures actually pays off for the company and from which level of risk aversion it becomes interesting to really take measures. Further below we will see that the return on the measures depends on the risk measure we use to calculate it.

Column J again contains the probability or frequency of occurrence, columns K, L and M the minimum, most likely and maximum loss. Depending on the measure, these numbers can differ from the values before measures. If the measure makes the risk occur less often, the probability or expected frequency of occurrence goes down. If the measure contains the loss once the risk has occurred, the three loss values go down. For the flooding of the production site in row 11, the measure affects both: the frequency of occurrence drops from 0.05 to 0.02. The minimum, most likely and maximum loss drop from €100,000, €200,000 and €500,000 to €50,000, €120,000 and €250,000. The measures cost €25,000.

Non-financial impact: reputational damage

Column N contains the reputational damage, a non-financial impact of the risk. Besides the financial loss, the occurrence of a risk can also attract public attention and thereby damage the company’s reputation. In the model, reputational damage is given in the levels “None”, “Minor”, “Moderate”, “Major” and “Severe”.

Qualitative impacts of a risk are not limited to reputational damage. There are others, such as harm to the health of employees injured in accidents, data loss and much more. The method of taking this qualitative information into account in an enterprise risk model is the same in all cases, though. That is why the example is limited to reputational damage.

How is the loss of an individual risk simulated?

The evaluation of the model begins to the right of the input data. Columns P and Q contain the simulated loss amount of each risk before and after measures. These are random numbers that came out of a single simulation run.

The simulation works like this: for each individual risk, the loss amount before and after measures is simulated. The sum of these individual losses is the total loss before and after measures. It is the actual target variable for the company and the result of one simulation run. In the simulation, this procedure is then repeated thousands of times. This produces an overall picture of the range of losses that can occur because of the risks.

Figure 4 shows how the simulated loss amount before measures is calculated for the first risk. Cell P8 is selected; it lies to the right, outside the section shown. Its formula is in the formula bar at the top. Below it we see the input data in columns D to H to which the formula refers.

Figure 4: The formula for the simulated loss amount before measures in cell P8. Compound combines the random number of occurrences (Poisson_variate or Bernoulli_variate) with a triangularly distributed loss per occurrence (Triangular_variate) (German interface).

The complete formula reads as follows. WENN and ISTLEER are the German names of Excel’s IF and ISBLANK functions:

=+WENN(ISTLEER(E8);0;Compound(WENN(D8="YES";Poisson_variate(E8);Bernoulli_variate(E8));"Triangular_variate";F8;G8;H8))

It first checks whether the row is filled with a risk at all. If the row is empty, the loss from the non-existent risk is of course zero. If a risk is present, Risk Kit’s Compound function comes into play. It composes the risk from two sources: the number of occurrences, which can itself be random, and the size of the loss in each individual occurrence, which is also random in each individual case.

The number of occurrences

The number of occurrences is determined in the first argument of Compound. If column D contains a yes, that is, if the risk can occur several times, the frequency of occurrence is represented by a Poisson distribution. The function Poisson_variate draws a Poisson-distributed random number, that is, a random number of occurrences. Its only input is the expected frequency of occurrence from column E.

If it contains a no, that is, if the risk can occur only once in a period, the occurrence is drawn as a Bernoulli-distributed random number. Its input is the probability of occurrence, also from column E. A Bernoulli-distributed random number is a coin toss, so to speak, in which heads comes up with the probability of occurrence. It is a very special coin that is adjusted to this probability.

The size of the loss per occurrence

The second argument describes the impact. It calls the function Triangular_variate, that is, a random number from a triangular distribution with the parameters from columns F, G and H: the minimum, the most likely and the maximum loss. For each occurrence determined in the first step, this produces a triangularly distributed random number. The sum over these individual occurrences is the simulated loss amount from the risk before measures. For the loss amount after measures, the formula works in exactly the same way, just with the corresponding other parameters.

How do you measure the risk contribution of an individual risk?

Figure 5 shows the right-hand part of the table with the simulated loss amounts in columns P and Q. In the simulation run captured in the figure, only two risks occurred: the increased cost of capital in row 10 with €13,745 before and €8,421 after measures, and the small legal dispute in row 15 with €31,903 and €17,463. All other risks show €0. Below, in cells M31 and N31, are the totals of this run, a total loss before measures of €45,648 and after measures of €25,884.

Figure 5: The right-hand part of the table with columns M to T. Columns A to L are omitted from the figure. Cell R8 is selected with the formula =ShortfallUp(P8;$P$5), and the confidence level of the risk contributions of 50% is in cell P5 (German interface).

Columns R and S with the risk contribution before and after measures, by contrast, are already statistics that are known only after a complete simulation. A single random number is not enough to determine this quantity. You can see this in row 8: the recession did not occur in this run, and its simulated loss amount is €0. Its risk contribution before measures is nevertheless €65,960, because it is calculated from all simulation runs.

The upper shortfall

In the model we propose to determine the risk contribution of an individual risk by the upper shortfall, in Risk Kit the function ShortfallUp. It has two arguments. The first is the simulated loss amount with its distribution, so P8 in cell R8. The second is the confidence level of the risk contribution, here 50% from cell P5. This number is a parameter you set yourself.

At 50%, the function calculates the mean of the upper half of the distribution. With a higher percentage, the mean is calculated only over the part above the corresponding quantile. The 95% quantile is the value that is not exceeded in 95% of the simulation runs. So if you set 95%, ShortfallUp calculates the mean over the upper 5% of the distribution.

The advantage of this risk measure is that it includes the extreme impacts that result from a risk. Some risks occur only with a very low probability. Most of the time they produce a loss of zero because they do not occur at all. With a small probability, however, they do occur after all and may then have large impacts. It is precisely these interesting upward impacts that this risk measure contains.

The same applies to the risk contribution after measures in column S. Comparing the two columns shows us what effect the measure has on the respective risk.

The return on the cost of measures

Column T contains the return on the cost of measures. In the first case, the recession in row 8, the cost of measures is zero. Here the return is not defined as such, and the cell shows only a hyphen. The measures are free, so to speak. In many other cases, of course, they cost money. Then we can read off here whether, for this definition of a risk contribution, we actually earn a return on the cost of the measures.

In many cases this is not necessarily so. For several risks the return is negative, for example −9% for the increased cost of capital in row 10 or −60% for the power outage in row 12. This means that our total costs are higher because we take these measures. Nevertheless, a measure can bring an advantage if it cuts off extreme events.

Making a certain net loss on measures is nothing unusual. Think of an insurance policy. The insurance company will usually pass the entire cost of risk on to the company as a premium, plus a processing fee and a profit margin. Insurance is still attractive because it smooths the company’s earnings: extreme swings of the risk disappear and are borne by the insurer.

With this, the modelling at the level of individual risks is complete in principle, and we can look at the results of the risk aggregation.

How are the risks aggregated?

As described above, the core of risk aggregation is to add up the simulated individual losses, once before and once after measures. This is simple business arithmetic. The thousands of simulation runs then produce a distribution of the total loss from which statistics and charts can be generated. They are located in the “Modell” sheet below the input data.

Figure 6 shows the resulting distribution as a histogram. It compares the loss distribution before and after measures: orange is the total loss before measures, blue the total loss after measures. The horizontal axis shows the loss from 0 to 1,200,000. The vertical axis shows the share of simulation runs that fall into each loss range.

Figure 6: Histogram of the loss distributions. Orange is the total loss before measures, blue the total loss after measures (German interface).

You can see very clearly that the measures reduce the extreme part of the distribution. On the right, where the losses are large, there are far fewer realisations after measures. Conversely, there are many more realisations in which no risks or only very small risks occur. That is why the blue bar at the far left is by far the tallest. The histogram is the bird’s-eye view, so to speak, the overall view of the loss distribution in this period.

Which risks have the greatest impact?

From the overall picture we now go into more depth. To do this, we look at the sensitivity of the overall result, that is, how strongly the sum of the risk aggregate responds to the individual input factors. We ask: which risk makes the largest contribution to the total loss? The answer is a tornado chart like the one in Figure 7. Each bar stands for one risk, with its ID from column A on the left. The length of the bar shows the sensitivity of the total loss after measures to this risk. At the top is the risk with the greatest impact, here risk 7, the data protection breach. It is followed by risk 1, the recession, and risk 10, the failure of a major project.

Figure 7: Tornado chart for the total loss after measures. The risks are sorted by their influence, with risk 7, the data protection breach, at the top (German interface).

This gives us a natural priority list. The bars usually get shorter very quickly. If we want to take further measures in addition to the measures already in place today and achieve the greatest effect, we therefore typically start at the top.

How do you draw a risk map when losses have a range?

A very classic way of presenting risks is the risk map. Risk Kit takes a very general approach here: you can plot any two statistics against each other. Figure 8 shows a so-called continuous risk map, whose axes are continuous numerical scales. The statistics chosen here are the probability or frequency of occurrence on the horizontal axis and the risk contribution on the vertical axis. As above, the risk contribution is the mean of an upper part of the distribution, so the extreme events are included here as well. The orange dots are the risks, labelled with their ID.

Figure 8: The continuous risk map “Eintrittshäufigkeit vs. Risikobeitrag” (frequency of occurrence versus risk contribution). Both axes are divided in steps of roughly a factor of 3, and the colour gradient runs from green at the bottom left to red at the top right (German interface).

Which numbers to enter in a risk map is a big problem in practice for many companies. The classic risk map plots a probability of occurrence against an impact, each given by a single number. In a more modern world, however, there are frequencies of occurrence, so risks can also occur several times, and the impact is given not as a single number but as a range. Then the question arises which number to choose, especially for the impact side, but possibly also for the occurrence side if it holds a frequency.

Here this problem is completely solved, because the statistics can be chosen freely. The upper shortfall, also called conditional value at risk, always exists, regardless of whether the risk has a range and whether it has a frequency. It can always be calculated from the simulation results, so the map can always be drawn.

The second characteristic feature of risk maps is the colour gradient. There is a green corner with, for example, small probabilities of occurrence and small risk contributions. Then come a yellow and an orange area with increasingly critical risks and finally a red area with extremely problematic risks. We position the risks in this field and see what the field actually looks like and what we have to prepare for. In Figure 8, for example, risk 1, the recession, has the largest risk contribution and is the highest.

The categorical risk map for reputational damage

Reputational damage is included in the analysis in exactly the same way. For this, Risk Kit has the so-called categorical risk map, in which both dimensions are divided into fields. For reputational damage these are the levels none, small, medium, large and very large, in the model “None”, “Minor”, “Moderate”, “Major” and “Severe”. For the financial loss, intervals are formed, which are made clear by the axis labels. The risks are then positioned and plotted in the individual segments. As with the continuous risk map, there is a colour gradient, here from green through yellow, orange and light red to dark red. Figure 9 shows the result.

Figure 9: The categorical risk map “Reputationsschaden vs. Finanzieller Schaden” (reputational damage versus financial loss). Risk 7, the data protection breach, lies in one of the dark red fields (German interface).

The horizontal axis shows the reputational damage, the vertical axis the financial loss. Risk 1, the recession, for example, has no reputational damage but is at the very top for financial loss. Risk 7 has very large reputational damage (“Severe”) and lies on the right in one of the dark red fields. The cluster of dots at the bottom left comes from the empty rows of the template.

What the risk maps need as inputs

The functions for risk maps are relatively complex to use. For the continuous risk map you have to give the system two groups of information, for the categorical one even three. The first group is the risks themselves. For the continuous risk map, each individual risk is specified with its two values, the probability or frequency of occurrence from column U and the risk contribution from column S. You can see this in the formula of the continuous risk map:

=ContinuousRiskMap(U8:U27;S8:S27;Labels!A7:A13;Labels!B7:B12;A8:A27;;;"Eintrittshäufigkeit vs. Risikobeitrag";"Eintrittswahrscheinlichkeit / -häufigkeit";"Risikobeitrag";K33;480;480)

Column U repeats the probabilities and frequencies of occurrence because empty data fields cannot be passed to the function. It always needs a number, so a zero has been added for the missing rows. The dots of the empty rows therefore lie on top of each other at the very bottom left of Figure 8. After the data ranges, the formula contains the axis labels from the “Labels” sheet, the risk IDs from A8:A27 for labelling the dots, and the titles of the chart and the axes.

The second group is the axis labels. They are on the “Labels” helper sheet in columns A and B, shown in Figure 10. Rows 7 to 13 contain the divisions for the continuous risk map: the frequency with the values 0, 0.03, 0.1, 0.3, 1, 3 and 10, and the size of loss with €0, €3,000, €10,000, €30,000, €100,000 and €300,000. This scale is chosen so that each step is approximately a multiplication by 3. This gives what is effectively a logarithmic scale.

Figure 10: The “Labels” helper sheet with the axis labels of the continuous risk map (rows 5 to 13), the levels of the colour gradient (rows 16 to 21) and the matrix of tolerance levels for the categorical risk map (rows 23 to 30) (German interface).

In principle you are free to choose the divisions and can draw the axes as you like. Risk Kit takes care of the scaling automatically in the background, even if the scale changes very abruptly.

For the categorical risk map it works the same way in principle. Here too, the two divisions and then the labels are specified. In addition, however, the map needs the colour gradient. That is why the “Labels” sheet contains, under the heading “Axis labels categorical risk map”, the levels “Tolerance”, “Low”, “Medium”, “High” and “Non-tolerance” and, below them, the matrix of tolerance levels. It assigns a level to each combination of financial loss (rows 25 to 29) and reputational damage (row 30). We see these levels in Figure 9 as the shading of the fields.

If you want to familiarise yourself with using risk maps, you will find examples for charts among the examples on the Risk Kit toolbar, including a separate example for risk maps, the entry “Risk Map” in Figure 2. It shows once more exactly how to proceed.

Embedding charts in the workbook

All charts are embedded in the workbook as images. To do this, you specify at the end of the formula the cell for the top left corner where the chart is to be embedded, followed by the size of the chart in pixels. The formula of the categorical risk map, for example, ends with K58;480;480. So the chart is anchored at cell K58 and is 480 by 480 pixels. This procedure is the same for all charts, including the continuous risk map, the sensitivities and the histograms: always the top left corner and then the size of the image as width by height.

How is the risk-bearing capacity determined?

Expected loss and value at risk

Besides the charts, the model provides statistics at the level of the entire risk inventory that describe the risk aggregate in individual values. Figure 11 shows them at the top left in the block “Risikoaggregate” (risk aggregates). For the total loss before and after measures, it lists the expected loss, that is, the mean over all simulation runs, and various quantiles that explicitly show individual points of the loss distribution. The quantiles are labelled as value at risk (VaR) with their confidence level, from the 50% VaR to the 99.9% VaR. The 90% VaR, for example, is the loss that is not exceeded in 90% of the simulation runs.

Figure 11: Risk aggregates and risk-bearing capacity. At the top the expected loss and the value at risk of the total loss before and after measures, next to them risk capital and capital used. At the bottom left the total return on the cost of measures, at the bottom right the total cost of measures of €116,000 (German interface).

The expected total loss before measures is around €150,000 (€150,708), after measures still around €70,000 (€70,409). So it is around €80,000 lower. The 50% quantile is €104,500 before and €23,774 after measures. It is the median of the distribution, so to speak, that is, the value that half of the simulation runs do not exceed. The median is considerably lower than the mean. The mean is evidently pulled up by extremely large risk events.

Further down the table follow quantiles that are exceeded less and less often. The 99.0% VaR is exceeded in only 1% of cases, the 99.9% VaR in as few as 0.1% of cases.

Risk capital as a limit

These quantiles can be the basis for defining a risk-bearing capacity. To do this, we compare a risk with an amount of risk capital that we are prepared to put at risk with a certain probability. In this way we express our risk tolerance or risk appetite in monetary amounts. They can be measured, for example, by the available assets, liquidity or free debt capacity. These amounts have to be worked out in a separate calculation. They enter the model as ready-made risk capital amounts, in Figure 11 in the column “Risikokapital” (risk capital).

This capital is a limit that the total loss after measures must not exceed. This is exactly the approach of IDW PS 340 as well: we formulate our risk coverage potential as risk capital, compare this risk capital with the risk and regard it as a limit that must not be exceeded.

The right-hand column “Verwendetes Kapital” (capital used) shows what part of the risk capital is used. At the 50% VaR, €23,774 of the €120,000 capital is used, which is about 20%. The cell is deep green, and we are far from the limit. With the higher quantiles and the rarer risk events, the capital we are prepared to put at risk also rises, because it is lost less and less often. At the same time, however, the risks rise as well, so the capital used at the respective confidence level also increases. At the 99.9% VaR, a total loss after measures of €596,687 faces a risk capital of €700,000, and the capital used is 85%. One limit applies throughout: the capital used should never go above 100%.

IDW PS 340 is about the continuation of the company. You can therefore first choose your risk preference and decide with which confidence level you want to be able to continue the company: whether, for example, you want to be 97% or 99% sure that you can continue the company. At that value, the capital must then be higher than the risk.

Do the measures pay off?

When discussing the input data, we said that we can compare the cost of the measures with the savings in the cost of risk. The measures reduce the expected loss from around €150,000 to around €70,000. In terms of expected value, we therefore save about €80,000. However, the total cost of the measures is €116,000, as the field “Gesamte Maßnahmenkosten” (total cost of measures) in Figure 11 shows. In terms of expected value, we are therefore considerably worse off with the measures than if we simply let the risk run.

This does not apply to all quantiles, however. The table “Gesamtertrag auf die Maßnahmenkosten” (total return on the cost of measures) at the bottom left of Figure 11 shows the return the measures achieve at the individual values at risk. The more extreme the events we look at, the better the return on the measures. In terms of expected value, as we have just seen, we lose money: the return is −31%. At the 50% VaR we also lose money, here −30%. But already at the 90% VaR we have a 60% return on the measures. For the very conservative risk measures, which represent very catastrophic events, the measures then protect very well, and the return becomes enormous: at the 99.9% VaR it is 195%.

This effect of the measures can be visualised very nicely. For risk communication within the company, which is also about justifying the decision to take measures, this is a very big advantage.

How do you communicate the results within the company?

The whole cockpit of charts and tables is built with risk communication in mind. The histogram in Figure 6 shows the risk profile before and after. You can see straight away that the measures clearly reduce the tail of the distribution, that is, the right-hand end with the rare, large losses. The tornado chart in Figure 7 shows the priorities by which we can take additional measures. This presentation of the sensitivities is a great advantage for risk communication.

The same applies to positioning the risks on the risk map in Figure 8. Risk maps are also extremely widespread, so this visualisation brings a high degree of familiarity. In addition, there is the inclusion of reputational damage in Figure 9. There we may find risks that are very well managed and small from a financial point of view but have larger impacts in terms of reputation. Or we find risks that are problematic by several criteria and should therefore be given double priority in risk management.

Which risk fields does the risk come from?

For risk communication, but also for risk management itself, it is often very helpful to pin down better which sources the risk actually enters the company from. To do this, you look not only at the individual risk but at intermediate aggregates that contain a whole group of risks. You can form them, for example, by risk field or, in exactly the same way, by organisational unit. This shows us which types of risk or, in a second step, which organisational units bring the risk into the company.

The aggregation by risk field is carried out at the very bottom of the “Modell” sheet, as Figure 12 shows. The table “Drill-down nach Risikofeldern” (drill-down by risk field) contains, for each risk field, the expected loss and the value at risk at confidence levels from 50% to 99.9%, in each case for the total loss after measures. It shows which area contributes which expected loss and which quantiles, and at which confidence levels the contributions are large. To the right is the tornado chart for the risk fields.

Figure 12: The total loss after measures, broken down by risk field. On the left the expected loss and the value at risk per risk field, on the right the tornado chart with cyber risk in first place (German interface).

Market risk starts very early. With €36,404 it has a relatively high expected loss and materialises quickly: already at the 50% VaR it shows an amount, namely €15,082. Further up, however, it does not rise very much, reaching €144,968 at the 99.9% VaR. Conversely, there are risks such as cyber risk that remain inconspicuous for a long time. From the 50% VaR up to the 95% VaR it shows €0, so with 95% probability nothing happens at all. When these risks do materialise, however, they have very large impacts: at the 99.9% VaR the figure is €439,526, the largest value in the table.

This is a very characteristic feature of cyber risks: they are inconspicuous for a long time, and when they happen, it gets very unpleasant. This is quite typical of one or another type of risk, and this presentation shows it very nicely. Here too, the sensitivities can be displayed. In the tornado chart, cyber risks even come before market risks.

So we have two different perspectives here. One is a risk that already has a certain magnitude in day-to-day business and therefore naturally calls for management. The other is a type of risk that has dramatic effects in a crisis and potentially endangers the company.

How can the model be extended?

There are many ways to generalise this model further. Four large groups are the following:

  • Individual loss distributions: a separate loss distribution is assumed for each individual risk, as we already mentioned above.
  • Several periods: instead of a single period as here, you look at several periods.
  • Complete and incomplete periods: especially if you want to look at the effects of a risk on the profit and loss statement, it may be interesting for the current financial year to look at a shorter period than a whole year. Perhaps there are only six months left until the end of the year. Then the probabilities and frequencies of occurrence have to be adjusted to this shorter period.
  • Dependencies: individual risks can influence each other, trigger each other or, conversely, make each other impossible. These effects are not included in the simple example model.

For small and medium-sized companies, this model is nevertheless a very good starting point for addressing IDW PS 340. Looking at several years and at dependencies is certainly only the second step for them. The first step is to identify and assess the risk inventory and to have a language for what the risk aggregation looks like, in what order of magnitude it is measured and how it is visualised and communicated.

Only when this whole process is fully established, when the company has seen an overall risk for the first time and has carried out a risk analysis, can you take further steps and add more complexity.