Professional Documents
Culture Documents
Quantitative finance
12/13/21 978-0-7346-1164-2 1
Learning objectives
4
• Asymptotic Single Risk Factor (ASRF) model: the model is defined below
but suggests that a lending portfolio would be affected by a single factor,
with the most obvious being interest rates. This condition is normally
violated by lenders with a large geographical spread.
• The model is as follow: Vit i M t 1 i2 Z it
• V is the value of the assets of lender i at time t
• M is the systematic risk
• Z is the unsystematic risk
• ρ is the risk attached to the systematic risk
5
Concentration risk
• H is the HHI
• s is the proportion of each firm’s loans to the overall portfolio
• n is the number of loans 7
• There have been a number of attempts to modify HHI
to allocate capital but as yet, none have been rigorous.
8
Expected losses
10
Probability of default
• There are many functional forms for probability of default (PD). For
default to occur, either there is no liquidity or equity becomes negative
(e.g. KMV).
• Functional this is:
P(d) = f(lf, ne)
Where:
• P(d) is the probability of default
• Lf is liquidity failure
• Ne is negative equity 11
• These models need to expanded to take into account
not just borrower characteristics but external factors
that affect the ability to repay.
• Thus the model will be as follows:
P(d) = f(micro-factors, macro-factors)
• Micro-factors would be the original model.
12
• The task is to develop models that are rigorous.
However, as most regression methods will give results
that are bounded by infinity. Usually, most models are
logistic to have bounded by 0 and 1.
logp1-p= iβimicro-factors+jβjmacro-factors
13
Loss given default
• Loss given default is difficult to model since the distribution is rarely normal.
Therefore, modeling is often focused on technique rather than variables.
• Variables that are normally used are:
• Type of loan
• Seniority
• Collateral
• Term
• Seniority of mortgage
• Characteristic of company’s liquidity
• Level of interest rates 14
• Most approaches to LGD is to create transformed
distributions that can be used such as:
• Fractional response regressions.
• Inverse Gaussian models
• Decision trees/neural approaches
15
Prepayment risk
16
• The normal approach is to model around the following:
• Monthly prepayment rate = (Refinance incentive) x (Season multiplier) x
(Month multiplier) x (Burnout)
• The interpretation of this formula is:
• Refinance incentive is the current level of interest rates with respect to interest rates
in the portfolio. If current interest rates are higher then there will little incentive to
refinance or prepay loans.
• Season multipliers recognise there are times there are higher than normal
prepayments.
• Month multiplier is similar to the season multiplier
• Burnout recognise that the longer loans exist the more likely they are to be prepaid.
17