You are on page 1of 3

Lecture 14 Quantifying Uncertainty and Risk

Some concepts in talking about uncertainty


Expectation and variance may not fully describe the random variables. You
can have two random variables with different payoff and different probabilities,
but with the same expectation and variance. Only normal random variable can
be completely specified by expectation and variance.
Marginal distribution can not describe what the joint distribution could be.
Law of iterated expectation
The Law of Iterated Expectation states that the expected value of a random
variable is equal to the sum of the expected values of that random variable con-
ditioned on a second random variable. E(y) = Ex [E(y|x)] Or can be expressed
as E(y) = p1 E(y|x = c1 )+p2 E(y|x = c2 )+...+pM E(y|x = cM ) E(y) is simply a
weighted average of the E(y|x = cj ), where the weight pj is the probability that
x takes on the value of cj . In another world, a weighted average of averages.
The reason to study iterated expectation is the usage of backward induction
in the form of tree. We need to calculate a value at a node on the tree and
usually the value is the expectation of contigenent event. We express it as the
iterated expectation from back towards front.
The reason to talk about risk and uncertainy is we do live in an uncertain
world. The implication and consequence of uncertainy is hyperbolic discounting.
If everything is certain, we will use exponential discounting. But since things
are uncertain, when we derive D(t), t refers to different maturity, we discover
discounting is done in a hyperbolic way. One day discounting from today and
from one year is totally different.
Hyperbolic discounting: a time inconsistent model of delay discounting. The
exponential discounting is a time consistent model of discouraging while the
1
discounting rate is the same throughout the time. DF = (1+r) According to
hyperbolic discounting, valuations fall relatively rapidly for earlier delay periods
(as in, from now to one week), but then fall more slowly for longer delay periods
(for instance, more than a few days). Say, one year r at year 1 is %5, but one
year r at year 10 is %4. The purpose of hyperbolic discounting is to show the
interest rate at different times are different, implicitly mean the interest rate is
a risk.
Intuitively, hyperbolic discounting make the DF smaller makes sense because
given the equal probability of upper and lower branch of the tree, in terms of
the discounting, the lower branch, i.e., the smaller rate will donimate the upper
branch, the bigger rate, thus in the long term, the rate is actually smaller and
smaller, DF is smaller and smaller.
# use an interest rate tree as an example to demostrate hyperbolic dis-
counting by calculating D(1), D(2), D(3),D(4) and you may find discounting is
smaller in each step.
Lecture 15 Uncertainty and the Rational Expectations Hypothesis: Appli-
cations to Predicting Stock Prices, Default Probabilities, and Hyperbolic Dis-
counting
The professor heavily use backward induction to derive PV. Backward in-

1
duction is like calculating bond price through discounting. You can deduce
the default probability through comparing default free bond (US treasury) and
default-able bond like Argentina bond. You can also derive it through CDS
spread. # Follow up: there is a hot topic about bond cds basis, I am not sure
what it is related.
Generally, high volatility of the interest rate will support the situation of
hyperbolic discounting. # Follow up: read the corresponding chapters in fixed
income book and there are certain tree examples explaining that.
Lecture 16 Backward Induction and Optimal Stopping Times
Relationship between default probability and spread
Backward induction and optimal stopping time. The beauty of backward
induction is on the boundary, the value is easy to calculate. In the middle of
tree, for each node, the value is the result of a more complex min/max function
You know the boundary value and you know the branch probability, through
expectation calculation, you can derive value at each step backward and if the
value is positive, you make the choice, otherwise you stop.
Lecture 17 Callable Bonds and the Mortgage Prepayment Option
The general idea to value a callable bond and mortgage is to do backward
induction. I am not writing down the details because these are simple usage of
min/max function. Just put things in perspective, using mortgage as an exam-
ple, 30 years mortgage , 7% mortgage rate, 6% interest rate, if no prepayment,
the mortgage is valued at $109. If the borrow prepay, the mortgage is valued at
$95. The borrower is borrowing $100 at the beginning. So you can see whether
prepay makes a big difference between the bank and borrower.
Why hedge fund buy mortgage? They know what fraction of people will
prepay, so they can value the mortgage better. From the above example, the
true value of the mortgage is between $109 and $95. If you know prepayment
better, you will have an edge. Say you know there are more prepay, the mortgage
is only worth $96, but you can sell it at $101, you can make a profit.
How to make the trinomial tree and binomial tree compatible? It means
that the standard deviation and the expectation of the interest rate ought to
be the same. Thus, in the transformation process, say from binomial tree to
trinomial tree, if you know the probability in each trinomial branch, you are
solving for the new sigma in the trinomial tree. This transformation/conversion
also shows binomial tree is not a special case for valuation purpose.
Lecture 18 Modeling Mortgage Prepayments and Valuing Mortgages
The professor proposed a way to show where the prepayment can occur in
the interest rate tree. Usually, in each node of the tree, we can calculate the
PV after prepayment or non-preayment. We can’t tell from PV itself whether
it is a result of prepayment or non-prepayment. Say, $95 at one node.
But if we normalize the balance of each node to $1. If at some node, the PV
is $1, we know at this node, we are pre pay. Here is how we set up:
(( coupon B5 ∗ W6p ) ∗ 0.5 + ( B5 B5 ∗ W6d ) ∗ 0.5)]
1 coupon
W5 = min[1, 1+r 4 B5 + B6 + B6
When you have prepayment model, you can incorporate the prepayment
forecast in your valuation. Therefore, the mortgage value at each step will
equal to:

2
B5 + B5 ∗ W6p ) ∗ 0.5 + ( B5 + B5 ∗ W6d ) ∗ 0.5)
( coupon
1 B6 coupon B6
W5 = prepayment + 1+r 4
Here, you don’t need $1, you replace it with some prepayment your forecast
and the second half of the formula represents the normal payment.
Lecture 19 History of the Mortgage Market: A Personal Narrative
Requirement: understand the motivation behind securitization and differ-
ence between products. These difference reflects on people’s thoughts about
how to securitize mortgages.
Mortgage pass throughs: put them together in gigantic pools called pass
through polls. They have to meet strict criteria.
Securitization adds great liquidity to the lender or buyer of the security. The
lender can off load the loan from their balance sheet. The buyer/investor can
buy or sell anytime they want.
Securitization is used to be done by government agency, but now can be
done by any financial intermediates.
Collateralized mortgage obligation, CMO. Different through pass through,
it has different tranches with different characteristics of risks and rewrds, tar-
getting to different investors. The investment bank can buy a pool but sell the
CMO pieces.
Collaterialzed debt obligation, CDO. The underlying is NOT mortgage but
bonds on the mortgage. Take the BBB bonds from different deals (CMOs), put
them into a pool and cut the pool further into different pieces. Some of the
pieces can be collateral to issue AAA bonds.
# During the securization process, how does the collateral gets used and
stretched? If the overall leverage ratio is 1:10, 1 collaterial, 10 overall credit
issued, how can the collateral be stretched more than that regardless whether
it is used and resued many times? Follow up with the ”The Leverage Cycle”
article published by the professor.
Hedging is important because if you know how to hedge away the risks you
don’t want, then you can bid for the risks you can bear and if you have an edge,
you can make money on it. If you can’t hedge, you may not be willing to invest
and take on so many risks.
Lecture 20 Dynamic Hedging
First, need to understand hedging and how one step hedging evoles to multi
step hedging and dynamic hedging.
One step hedging, need to tell the difference between betting trade and
hedging trade.
Here are the betting tree and hedging tree.
Expected payoff: 0.6 ∗ 100 + 0.4 ∗ (−100) = 20. Hedge to secure the payoff
regardless which state the game ends up with. 0.6 ∗ (−80) + 0.4 ∗ 120 = 0 which
is a fair game. Therefore, to make money, you have to know how to hedge and
can find a market readily available to hedge.

You might also like