The test results show that the bank can continue to make profits with 90% confidence level, indicating that the credit risk is small. In extreme cases, some banks will face a loss of 99% confidence level in VaR, but the probability of this extreme situation is very low. This is just an early warning.
The test process is divided into the following steps:
Step 1: Define the model.
Step 2: Estimate the model.
Step 3: Analysis of model estimation results.
Step 4: Design the impact scenario.
Step 5: Construct the frequency distribution.
Step 6: Calculate the mean and VaR.
Step 7: Measure the impact on the profitability of banks.
The process is summarized into seven steps, including the following aspects of calculating profitability. The first is to define this model. There are independent variables and dependent variables in the model and four dependent variables are defined. The dependent variable is that it needs to examine the credit default rate. Its definition of default rate is as follows: loans overdue for more than 3 months and the total amount of loans. I wonder if banks use this data as default rate. This data is also difficult to work out. Usually, the published data is still the non-performing loan ratio, and the definition of default rate is not accurate. Some define that the previous period can be repaid normally and the next period cannot be repaid normally, so there are several definitions of default rate. After all, there were too many policy reasons for the divestiture of commercial banks in previous years. Perhaps there are some irresistible factors in this time series, that is, there are too many fuzziness.
Look at this estimation model. This is the data of retail banks from April 1994 to June 2006. The front is the independent variable, which is the result of historical data estimation, including parameter variables. At the bottom are the observed values and the number of tests.
It can be seen that its sign is consistent, because the default rate is derived from the function Log, so the better the economic environment, the higher the asset quality, in this case, the lower the value of VaR should be. It can be seen that this is positively related to economic growth, real estate prices and interest rates.
At the same time, as mentioned above, there are actually many second-order lag terms in the independent variables, which are obtained by eliminating the first-order lag terms. Many other related variables were not listed, so this is the final simulation result. After simulating this equation, the next step is to set the influence scenario. First, we should design the model, estimate the model, and finally bring new data into our model. Is to bring the first independent variable into the model and make it a new dependent variable. So, what about the new independent variable? For example, after our economic shock, what is our impact? In fact, similar to the economic crisis, there are four impact points. One is the four independent variables I just mentioned, which have an impact on each variable. The first is the change of real GDP in Hong Kong, and the second is the change of real GDP in Chinese mainland, as well as interest rates and real estate. It not only changed the independent variable of the current period, but actually extended the time, turning this influence time into two years. Then, how much should this pressure change after the financial crisis? 1The interest rate in the fourth quarter of 997 was 306 basis points, then decreased in the next two quarters, and increased by 365,438+04 basis points in the fourth quarter. It can be seen that it was more than 300 basis points at first, but it did not change in the next two quarters, and it rose again in the fourth quarter, which was similar to the impact of the Asian financial crisis at that time.
Then, the next step is simulation, because after inputting this data into the model, a new probability distribution can be calculated from the simulated data. Of course, there is also an assumption that there will be no impact after the fourth quarter, and the future default interest rate path is simulated 10000 times in each base period scenario and stress scenario. With the new frequency distribution, we can construct the frequency distribution of our credit loss rate. We have just simulated the frequency distribution of default rate, and our loss percentage data should be the default rate multiplied by loss given default. The data defining loss given default is controversial. How to determine? Without proper statistical data, the value of market information is usually set at 50%. According to BASELII, LGD is 45%, but this figure is not very reasonable. So it is defined as the formula of 2% low point. In this way, loss given default can be multiplied by the default rate we just calculated, so that we can get a frequency distribution data of the credit loss rate. After the shock, we actually moved the frequency to the right. We can see that the frequency of data with high credit loss percentage has increased. These data originally moved outward, so we can see that the frequency of higher credit loss percentage has increased and the frequency of smaller credit loss percentage has decreased.
By calculating the distribution, we can calculate the average value of the percentage of credit losses, and also calculate the probability of losses, so that we can make such a fine judgment. This is the calculated result, and its average value is as follows: First, there is no credit loss percentage in the base period, with an average value of 0.34, GDP shock in the pressure period is 1.59, house price shock is 1.2 1, interest rate shock is 0.7 1, and mainland economic shock is 0.73. 90% of the credit loss percentage in VAR is this data. With the increase of confidence interval, the loss percentage is also increasing. The last one is 99.99%, which is already quite high at this time. The latter two are close to 10%, and the former one has exceeded 10%.
At the 90% confidence level, it can be seen that below 3% is tolerable. In 99% of cases, the value is already relatively high, 3.22 is the lowest value, and the highest value is 5.56, which should be relatively high. This is consistent with the situation after the financial crisis 1 year, so it is more realistic to consider the current period and extension period of the impact when doing stress tests. The calculation here is that before the Asian financial crisis, the bank can calculate the bank loan loss rate as 1.4%, and the loan loss rate rises to 6.0%, but this estimate is based on the estimated LGD of 70%. Then, this raises a question, whether this is reasonable or not, which may need to be considered during the test.
The last step is to measure the impact of the impact on the profitability of banks. Perhaps the bank management thinks, what is the VaR value or probability, how big it may be in 90% confidence period, how big it is in 99%, and how big it is to profitability? How much has the profit dropped? Can you give such data, then you can also work it out through a calculation. If you accept the previous calculation, it is the percentage of loan loss. Through this calculation, the loss must be equal to the loan loss percentage multiplied by the loan balance. It is the change of bank profitability after the impact. First of all, if there is no default, the profit after its future impact should be the same as the current period or base period. If my profit is 3 billion, then there is no default after the impact, then this profit is the same. If it matters, if I fall, how much I fall is a loss.
Suppose there is a bank with a pre-provision profit of 3 billion and a loan balance of HK$ 654.38+03 billion. Assuming that a bank is so big, it can be measured by the above loan loss percentage, and how much its profitability will be affected in different confidence intervals after the impact. This is the result.
The unit is millions, with positive data indicating profit and negative data indicating loss. When the management sees this table, it may be more clear how much losses the bank may have.
For example, within 90%, banks will lose HK$ 8.82 trillion under the impact of Hong Kong's GDP. Therefore, this is 99.99%, that is, the probability of this happening is very high, because the confidence interval is 99.99%, that is, the probability of 0.00 1%, and the loss reaches13.3 billion. In different confidence intervals, the loss is different. Looking back, if there is no simulation, it is a hypothesis. What is the assumed GDP? It has just been put forward. From the front, we can see what the GDP data is and how much it is every quarter. If there is no simulation, we can directly bring this data back to the model and only calculate the data of a loan percentage. Through simulation, we know what its mean value is and what it is in different confidence intervals. In this way, the management may feel more sober. For example, there is no default of 2.554 billion yuan in the base period, which is still quite good. If you do a stress test and give this form to the management, you will know exactly how much the loss is.
Finally, there is a saying that at 90% confidence level, the VaR value is 8.82 million. If the value of VaR is relatively large at the confidence level of 99%, the probability of such an extreme scenario of VaR is 1%.