Insights from the dairy benchmarking exercise
Four banks – ANZ NZ, ASB, BNZ, and Westpac NZ – are allowed to use their own models to determine their capital requirements, after approval by the Reserve Bank. These capital requirements are determined by the relative riskiness of loans, which is summarised by the risk weights attached to those loans. Risk weights reflect both the probability that the loan will default (PD) and the loss a bank expects to incur if the loan defaults (loss given default, LGD).
The Reserve Bank has conducted an exercise to assess the relative levels of conservatism across the models that these banks use to determine risk weights for their farm lending portfolios.
Banks estimated the relative riskiness of loans…
The exercise required the banks to measure the risk of the same portfolio of loans to 20 hypothetical dairy farms. These farms represented a range of characteristics and varying degrees of risk. Banks were then provided with financial data and descriptive information for each farm, as well as the details of the hypothetical loans.
Banks were asked to estimate the PD and LGD for each loan using their approved models.1 Where appropriate, banks applied overrides to the model estimates to reflect information not captured by the model. This mirrors the approach that banks use in practice.
…and the estimates varied significantly across banks.
The preliminary results of the exercise indicate significant differences in estimates across banks. The highest and lowest average risk weight for the whole hypothetical portfolio differed by 40 percentage points, leading to differences in the hypothetical capital requirement.
Variation in both PD and LGD estimates was significant. Figure C1 shows the range of average PD estimates across five groups, each containing four loans, ranging from the group of loans with the lowest estimated PDs to the group with the highest estimated PDs. Each line represents the estimates of one bank, before overrides. Absolute variation was largest at the mid- to high-risk end of the spectrum, but proportionate variation was large across all levels of risk. The model overrides applied by banks tended to reduce the variation across banks, but it remained significant.
Figure C1 Bank model estimates of loan probability of default (average PD for groups of 4 loans)
LGD estimates also varied, with the highest average LGD estimate for the whole portfolio 18 percentage points higher than the lowest average LGD estimate. Although banks model the LGD of farm loans, the Reserve Bank imposes minimum values based on the LVR of the exposure. These minimums are binding in most cases, but still vary across banks due to differences in the way banks measure LVRs. This is either because banks recognise different kinds of assets as security or because they apply different discounts to ‘book values’ to adjust security values.
The provisional results show significant variation in model outcomes, even for the same level of underlying risk. The Reserve Bank is conducting further analysis of banks’ farm lending portfolios to see if patterns in actual risk estimates are consistent with the results of the hypothetical exercise. This work will help inform the Reserve Bank’s review of bank capital requirements.
1 Banks were also asked to estimate a third parameter – the expected exposure at the time of default. This parameter impacts capital requirements, but is not part of the risk weight calculation.