Michaelmas term 2021
FM405E
Fixed Income Securities and Credit Markets
Summative work, counting towards 30% of the final mark All the summative work
is for individual assignment, not for group work! The deadline for SW1 is 3 November, while
the deadline for SW2 (which will be released on the 3 November) is 25 November. It is plenty
of time, but do not leave it for the last minute!
© LSE 2021/FM405E Page 1
SW2 Summative work 2: Monte Carlo Simulation [15%] Build and calibrate a quanti-
tative mortgage valuation model as described below by writing an appropriate code in
Matlab or R or Excel (Please use only general toolboxes, without relying on ready-made
financial routines, e.g. a routine calibrating Ho-Lee trees.) If you use Excel, please use
T = 5 years and N = 1000 simulations at all points applicable below. If you are using
a coding language, please use T = 10 and N = 100000. Otherwise, the choice of the
software is yours. Please organize your assignment as follows:
There should be a detailed write up, which includes figures and procedures to get
your answers, for example, some pseudocode. This should be self-contained and
well-explained. In principal, your mark is based on this document. The other files
are for us to replicate how you got your results.
We also expect a zip folder with the code and the data files or Excel sheets. Ideally,
after expanding your folder, we can run your code/ see your Excel sheet generating
all the figures and output on which the write-up is based on.
It is January 2019 and you are working in the back-office of the University Building
Society (UBS) providing mortgages to university students and employees all over UK
and issuing mortgage backed securities based on those mortgages. UBS has just decided
to market a novel product, a T -year fixed rate mortgage. Your boss asks you to develop
a model which helps to come up with the right interest rate for this product. If the
interest rate is too high, no one will buy the product. If it is too low, UBS will lose on
the mortgages. (Hint: These problems build on the example in 13.6 in Veronesi (2010)
with longer maturity and a different yield curve. Perhaps a good way to start is to write
the code using the original data in that example, so you can verify whether your code
gives the same numbers. Then you can modify the input according to the problem below.)
(a) You decide to work with the binomial tree methodology. For this, you have to
start with a binomial tree model of risk-neutral short rates. For robustness and
simplicity, you decide to experiment both with the Ho-Lee model (HL) and the
Black-Derman-Toy (BDT) model, with semi-annual (∆ = 0.5), continuously com-
pounded short-rates. You estimate that the historical volatility of the level of
short interest rates is σ = 1.73% while the volatility of the log of interest rates is
σ = 21.42%. You should calibrate both type of trees to the current yield curve,
which you can find on Moodle (as provided by the Bank of England) in excel sheet
[Link]. (Treat the yields as continuously compounded.)
(b) As a benchmark, assume that the buyers of the mortgage will prepay optimally
and will not default. Following our method of valuing mortgages in Chapter 4 of
© LSE 2021/FM405E Page 2
the lecture notes, find the fixed rates for which the value of the T-year mortgage
is (approximately) par under the Ho-Lee model and under the Black-Derman-Toy
model. Which one is higher? Do you have any intuition for that?
(c) Now you switch to the Monte Carlo methodology. Using N simulations plot a
histogram of the simulated interest rates in year T under each of the short rate
models. Can you comment on the difference? (Hint for Excel users: You might
want to use the function INDEX() to help locate the simulated path on the value
tree, so as to avoid repeating calculations. Look in the Help section of Excel for
instructions. )
(d) Let us proceed with BDT from here. Use the corresponding mortgage rate you have
found in problem (b). As a check on your calculations, you decide to revalue the
mortgage contract using a Monte Carlo methodology. What is your point estimate
and your confidence interval using N simulations? Is that consistent with your
answer in (b)?
(e) You decide to incorporate additional assumptions on prepayment. (You continue
with the BDT model and Monte Carlo).
You realize that sometimes mortgages are prepaid even if it is not optimal.
This typically happens when people move. You also realize that people tend
to move more often during the summer months and tend to repay less often
at the beginning of the contract. To have some guidance on the probability pi
that a mortgage is prepaid in period i even when it is not optimal, you assume
that pi is related to the so-called PSA measure of prepayment speed, discussed
in Section 8.3.1 of Chapter 8 in Veronesi (2010). Under the PSA experience
the conditional prepayment rate in period i (CP Ri ) follows the rule of
CP Ri = min(12i × ∆ × 0.2, 6)(%)
where the first period is i = 1. (It starts at 0.2% in the first month, increases
by 0.2% in each months then levels of at 6%.) In particular, you use the 50%
of the PSA probability and assume that it doubles during the period including
summer. (As we start in January, this is every even period.) To be more
specific, you model pi as
pi = season index × (1 − (1 − 0.5 × CP Ri )∆ )
where ”season index” is 1 or 2, depending whether the period is odd or even.
You also realize that for various reasons (e.g. inattention, financial or psycho-
logical cost of the refinancing decision) borrowers might not prepay, in periods
© LSE 2021/FM405E Page 3
when it would be optimal. They prepay with larger probability when interest
rates are low, but this probability might not be 1. For instance, it makes sense
to assume with that in each period when it would be optimal to prepay, they
do so only with probability
qi = a × e−b×ri
where 0 < a ≤ 1 and b ≥ 0 are calibrated parameters. Smaller a implies less
prepayment and larger b implies a larger sensitivity to interest rates. According
to your calibration, a = 0.8, b = 20. (Do not let the typo on page 486 of
Veronesi (2010) confuse you. The last expression in the one but last paragraph
should be RAN D() < qis .)
What is your point estimate and your confidence interval using N simulations on
the value of the mortgage under these additional assumptions using the interest rate
you have found in (b)? Comment on the effect of these assumptions. Under these
additional assumptions, how would you change the mortgage rate you recommend
to UBS?
(f) As the new product is very successful, UBS decides to issue a pass-through secu-
rity, an interest rate only security and a principal only security backed by these
mortgages. The maturity of these MBSs are the same. However, to recover the cost
of securitization, UBS decides to offer them with an interest rate 50 bp lower than
the mortgage rate you recommend in problem (e).(If your recommended mortgage
rate is smaller than 50 bp, you can use 1% as the interest rate here.) Using the
assumptions of problem (b) and a Monte Carlo with N simulation, what is your
point estimate and confidence interval for value of the three MBSs?
© LSE 2021/FM405E Page 4