Normal Distribution Problems
Uploaded by
Ajitesh KumarNormal Distribution Problems
Uploaded by
Ajitesh KumarNormal Distribution Problems
1. A bank is considering an application for loan by a new upcoming business. In order to
determine the appropriate interest rate for the loan, the bank needs to assess the expected
loss from the transaction in the event of a future default on the loan. A specific parameter
of interest in making this assessment is the loss given default (LGD) defined as the
percentage of loan amount that goes unrecovered. The bank manager thinks the business
has potential to succeed but has limited experience from the past data that can provide an
idea on LGD. After a careful look at past deals, the chief risk manager with the help of his
team identifies 5 historical deals that had gone into default in the past. The LGDs for these
deals were 5%, 8%, 10%, 12% and 5%. This suggests an average LGD of 8%. Given the
limited past experience, the chief risk manager also sought opinion from three experienced
risk professionals on his team and asked them to suggest a reasonable LGD based on their
experience on similar deals in their career. The three of them generally agreed that such
deals are more risky than what is apparent from the 5 historical deals that the company has
pulled together. The LGDs suggested by the three of them were 40%, 50% and 60%.
a) Assuming a normal model for observed LGDs (i.e. ( , ) ) with unknown mean and
known sd of = 2%, and a discrete uniform prior for the three estimates from experts,
obtain the posterior estimate for on such deals.
b) Construct a normal distribution prior for , using expert information. Obtain the
posterior distribution, 95% credible intervals. Compute the posterior mean and variance.
c) Compare the posterior and prior pdfs. What happens if n (number of past deals) were to
increase in support of the 8% average?. What happens if the prior variance decreases?
d) What is the predictive distribution for LGD on the new deal?
e) What is the probability that LGD on the new deal will be greater than 50% based on the
predictive distribution?
f)
Now assume a normal-gamma prior on ( ,
). i.e.
distribution for
with shape=10 and rate=40.
compute the posterior distribution for .
~ (10,
) and Inverse gamma
Numerically, using simulations,
Picture of I-Beam, and I-beam supporting roof of a house.
(https://en.wikipedia.org/wiki/I-beam)
2.
An engineer takes a sample of 5 steel-I beams from a batch and measures the amount they
sag under a standard load. The amounts in mm are 5.19, 4.72, 4.81, 4.87 and 4.88. It is
known that the sag is ( , ).
a) Suppose SD is fixed, i.e.
.25. Use the prior
~ .5, . 5 and obtain the
posterior distribution for .
b) For a batch of steel-I beams to be acceptable, the mean sag under the standard load
must be less than 5.20. What is the probability that is less than 5.20 under the
posterior distribution?
c) Suppose now that the mean is fixed, i.e.
.5 but we have a prior on . More
specifically, assume an inverse gamma prior for
with shape=20 and rate=80. Using
the data 5.19, 4.72, 4.81, 4.87 and 4.88, find the posterior distribution for .
d) Find the predictive distribution for the amount of sag experienced by a new item from
batch.
