Beruflich Dokumente
Kultur Dokumente
Fall 2000
You have 1.5 hours to answer the following three questions. You are allowed both sides of an 11x8.5 in
h
pie
e of paper for notes. GOOD LUCK!
1. (50 points)
Derived Random Variables: Let X and Y be two independent uniformly distributed
random variables on the interval [0; 1. Let Z = X 2 + Y 2 . Please nd the CDF and then
e
the probability density for the random variable Z . Sket
h the density fun
tion.
CDF for A = X 2 is obtained from
Z pa
dx
FA (a) =
0
whi h implies
1
2
fA(a) = a
1 2
on (0; 1) and zero elsewhere. Sin
e X and Y are independent the density on Z is the
onvolution of the densities for X 2 and Y 2 (whi
h we just derived). There are two intervals to
onsider. On 0 z 1 we have
Z
fZ (z ) =
1
4
(z
a)
fZ (z ) =
1 1
(z
4 z 1
a)
= (a) 1=2 da =
1 2
and on 1 z 2 we have
Z
= (a)
1 2
1 2
da =
and that is as far as you
an go using this approa
h unless you have an ex
ellent
ommand of
integral
al
ulus (whi
h I don't), an integral book or a symboli
math program. If you went this
way and formulated it this far, you got almost full
redit. If you went on to get it
ompletely
orre
t you got full
redit and a smile from me for being a
al
ulus wizard.
R
, the integral of p 1 dx is
In any
ase, a
ording to MAPLE
x(
x)
x
=2
)
x
x2
ar tan( p
so that
fZ (z ) = =4
on [0; 1 (whi
h is EXTREMELY
ounterintuitive until you look at the next solution). For
[1; 2 we have
"
1 z=2
1
)
fZ (z ) = ar
tan( p
4
z 1
ar tan( p
(z
(z
1) z=2
)
1)z (z 1)2
Another approa
h is to look at the JOINT probability spa
e (whi
h is a little square in the
(x; y)) plane and
al
ulating the CFD of Z dire
tly by integrating the appropriate region in
the (x; y) plane.
1
For 0 z 1 the region is a 1/4-pie (like the
ir
ular pie you eat) sli
e of radius z . For a
sli
e whi
h subtends angle radians with radius r, the area is 2 r2 This yields
Prob(Z z ) = z
4
Prob(Z z ) = z
1 + z (=4
Dierentiating in z gives
ar tan z
1)
fZ (z ) = =4 ar
tan( z 1)
whi
h (unbelievably enough) is the same result as the rst method.
Note that even if you did not know the derivative of ar
tan, this approa
h allowed you to
sket
h the PDF sin
e you SHOULD know how ar
tan() behaves more or less even if you don't
know the integral.
2. (50 points) A Blast From the Re
ent Past: Suppose an experiment is performed N times
independently and that the probability of event A o
uring is p. We then know that if k is
the number of times event A o
urs in N trials that by the (strong) law of large numbers we
have
k
=p
lim
N !1 N
(a) (25 points) Let X be a dis
rete random variable whi
h
an take
on a nite number of
P
dierent values am , m = 1; 2; :::; M ea
h with probability pm , m pm = 1. Suppose we
have N independently observed values of X , x1 ; x2 ; :::xN . Use the above law to show
that
N
1 X
lim
xi = E (X )
N !1 N i=1
Let km be the number of times event m o
urred in the N trials. Then limN !1 km =N =
pm . Thus we
an form
M
N
M
X
X
k
1 X
am m =
lim
xi = lim
a p = E (X )
N !1 N i=1
N !1 m=1
N m=1 m m
(b) (25 points) If possible, repeat part (a) for a
ontinuous random variable, but one whi
h
is limited to taking on values on a nite interval, say (0,1).
Let the
ontinuous random variable be X with probability density fX (x). Form bins of
size 1=M and essentially quantize the sample values xi to values am = m=M , m =
1; 2; ; M . Call the derived (dis
rete) random variable X~i . We then dene N = QM
for an integer Q and the dieren
e Vi = Xi X~i . We then have
pm = Prob((m
1)=M X~ i m=M ) =
m=M
=M
1)
fX (x)dx
lim
x~i =
Q!1 QM i=1
m=1
am pm = E (X~ )
We then write
X
1 QM
X
1 QM
lim
x~ = lim
(X
Q!1 QM i=1 i Q!1 QM i=1 i
and note
p that we must have jVij
Q = M we have
lim
Vi ) = lim
X
1 QM
Q!1 QM i=1
1=M .
3=2
MX
M !1 M 3=2 i=1
Vi
Xi
X
1 QM
lim
V
Q!1 QM i=1 i
Mlim
1=M = 0
!1
am pm = E [X
lim
M !1 m=1
and we're done.
3. (50 points)
Rutgera and Wall Street: Rutgera Universa, the world famous Rutgers University Graduate student has a part time job on Wall Street with an investment bank. Rutgera's job is
to nd promising investment strategies and make money for her bosses.
Rutgera is approa
hed one day with the followed s
heme.
You must help Rutgera de
ide whether this is a good or bad investment s
heme.
(a) (10 points) Assuming Rutgera puts C (1) = Q dollars in at the beginning of day 1,
arefully sket
h the probability mass fun
tion for P (1) and
al
ulate its expe
ted value.
Cal
ulate its varian
e.
Prob(P (1) = 2Q) = 1=2, Prob(P (1) = Q=2) = 1=4 and Prob(P (1) = Q=4) = 1=4.
The expe
ted value is Q(2(1=2) + (1=2)(1=4) + (1=4)(1=4)) = (19=16)Q. The varian
e is
Q2 [(13=16)2 (1=2) + (11=16)2 (1=4) + (15=16)2 (1=4) = 684Q2 =1024 = 0:668Q2
(b) (10 points) Ea
h day Rutgera reinvests the whole portfolio C (n) = P (n 1). Let n
be the s
ale fa
tor for day n. Find an expression for P (n) in terms of C (1) and the i .
What is the expe
ted value and varian
e of P (n) assuming C (1) = Q.
Q
P (n) = Qn ni=1 i . Sin
e the i are independent E (P (n)) = Q[E ()n = Q(19=16)n .
The varian e is
Q2n E [(
i=1
"
i (19=16)n )2 = Q2n
i=1
E [2i 2(19=16)2
i=1
E [i + (19=16)2n
whi
h simplies to
h
(19=16)2n = Q2 (133=64)n
(19=16)2n
(
) (10 points) Dene L(n) = log P (n). Cal
ulate the mean and varian
e of L(n).
Q
P (n) = Qn ni=1 i . Sin
e the i are independent E [log(P (n)) = log(Q) + nE log() =
n log(Q) n=4. The varian
e is the sum of varian
es of the individual i . Sin
e they
2
2
2
2
2
log(
P (n)) = n = n[(5=4) =2 + ( 3=4) =4 + ( 7=4) =4 = (108=64)n = (27=16)n
(d) (20 points) Is this a good or bad investment s
heme? HINT: This is an appli
ation of
the law of large numbers.
It looks like a good investment s
heme at rst be
ause the expe
ted value of the portfolio
on day n is Qn n where = 19=16 > 1. That is, the expe
ted value of the portfolio
goes to innity in n. However, the fa
t that the varian
es also go to innity is
ause for
on
ern.
So we note that the log portfolio is a sum of independent identi
ally distributed random
variables. That means that the sample mean goes to the expe
ted value WITH PROBABILITY 1!!!!! The expe
ted value is -1/4 whi
h means that log(P (n))=n ! 1=4.
Thus, all the money is lost with probability 1. Ergo, the investment s
heme SUCKS!SUCKS!SUCKS!
Note: this problem was slightly dierent from previous versions on previous exams in
that you did not need the
entral limit theorem (be
ause you don't know that theorem
yet). If you used the CLT, you did not get
redit.