C B

rownian Mohou
 旺上

Brownian Motion
standard
Bt is

Àcremeuts
In dependant

20 Bt Bs NCO.t D.tn
B ⼆ 0

Bt continuous as
 Alter 與
Brownian Motion
standard
Bt is

g Gaussian Process

snt
EBFO GCB 則
fo

11

EBBD

Bt continuous as

Note Much easier to use in practice

出
Example Ǚ陷啦 is a BM

i
⼋

i iiiiii.it

t.it
Proof By alternative def

30
⼀
EĒEE 貼 13 0

Gg 盯 ÉGBB
Stt t.tt to S
 B Masbuid

Brownian bridge

ttanos

BM wit drift

Y 只
ogwiǐ
5
put

Xiětttcg iǐmwǐu
的
鬪 o

S.DE

si
dByo.snnvhhtnu
dXttndt oit
器
簽
簽

flmosteeywhg
Btt
isybggga.sn

Unbggg

veore I.nihu
 fnhuiVar
UBI PEIBti
B. a
i 1

csga rB in li to

Ǘ
t.tt tnitn

Tn partition ti tn
 Morey

f is B.li

f f
f
ˊˊ
fisdifntiadea font.cl
dias

eg smooth piecewise

ˊ
irr
so R L integral

f ft dt
t
B.li

RS integral

ffndgn

蜘 斷
So what

SBdB.sn VOTwell defineo

Need new integrals

Ito's Integrals

Why needed
世

g Earnings
⼆
ff dB
Q Is GBM reasonable

in fin modelling

Mt 5Bt
GBM Sie

r
 Implications
Mt 5 是

fze
lnf f
5 只

只
h Sttottn Stzuot
5

hdtx r
Bt
ot r
o ilog

retumstnf.tl
iNG Qinht

Reasonable
log_return vs relative return

Definition

log return h
樂
名
relative return ⼆

舉
Relation

return ⼆ relative return

log
 Proof

log return h
樂
1譽
hl 司

⼆
lnft St
t
⼆ St relativereturn

as
hltfkS fn.ES small
In reality

上
18 FINANCIAL TIME SERIES AND THEIR CHARACTERISTICS

0.2
s-rtn

0.0

−0.2

1940 1960 1980 2000

year

0.3

0.1
log-rtn

−0.1

−0.3
1940 1960 1980 2000
year

Figure 1.2. Time plots of monthly returns of IBM stock from January 1926 to December 2003. The
⼀
upper panel is for simple returns, and the lower panel is for log returns.

0.4

0.2
s-rtn

0.0

−0.2

1940 1960 1980 2000

year

0.2
log-rtn

0.0

−0.2

1940 1960 1980 2000

year

Figure 1.3. Time plots of monthly returns of the value-weighted index from January 1926 to December
2003. The upper panel is for simple returns, and the lower panel is for log returns.
DISTRIBUTIONAL PROPERTIES OF RETURNS 11

o
Table 1.2. Descriptive Statistics for Daily and Monthly Simple and
Log Returns of Selected Indexes and Stocksa

Standard Excess
Security Start Size Mean Deviation Skewness Kurtosis Minimum Maximum

Daily Simple Returns (%)

SP 62/7/3 10446 0.033 0.945 −0.95 25.76 −20.47 9.10

VW 62/7/3 10446 0.045 0.794 −0.76 18.32 −17.14 8.66
EW 62/7/3 10446 0.085 0.726 −0.89 13.42 −10.39 6.95
IBM 62/7/3 10446 0.052 1.648 −0.08 10.21 −22.96 13.16
Intel 72/12/15 7828 0.131 2.998 −0.16 5.85 −29.57 26.38
3M 62/7/3 10446 0.054 1.465 −0.28 12.87 −25.98 11.54
Microsoft 86/3/14 4493 0.157 2.505 −0.25 8.75 −30.12 19.57
Citi-Group 86/10/30 4333 0.110 2.289 −0.10 6.79 −21.74 20.76

Daily Log Returns (%)

SP 62/7/3 10446 0.029 0.951 −1.41 36.91 −22.90 8.71

VW 62/7/3 10446 0.041 0.895 −1.06 23.91 −18.80 8.31
EW 62/7/3 10446 0.082 0.728 −1.29 14.70 −10.97 6.72
IBM 62/7/3 10446 0.039 1.649 −0.25 12.60 −26.09 12.37
Intel 72/12/15 7828 0.086 3.013 −0.54 7.54 −35.06 23.41
3M 62/7/3 10446 0.044 1.469 −0.69 20.06 −30.08 10.92
Microsoft 86/3/14 4493 0.126 2.518 −0.73 13.23 −35.83 17.87
Citi-Group 86/10/30 4333 0.084 2.289 −0.21 7.47 −24.51 18.86

Monthly Simple Returns (%)

SP 62/1 936 0.64 5.63 −0.35 9.26 −29.94 42.22

VW 26/1 936 0.95 5.49 −0.18 7.52 −28.98 38.27
EW 26/1 936 1.31 7.49 −1.54 14.46 −31.18 65.51
IBM 26/1 936 1.42 7.11 −0.27 2.15 −26.19 35.38
Intel 73/1 372 2.71 13.42 −0.26 2.43 −44.87 62.50
3M 46/2 695 1.37 6.53 −0.24 0.96 −27.83 25.80
Microsoft 86/4 213 3.37 11.95 −0.53 1.40 −34.35 51.55
Citi-Group 86/11 206 2.20 9.52 −0.18 0.87 −34.48 26.08

Monthly Log Returns (%)

SP 26/1 936 0.48 5.62 −0.50 7.77 −35.58 35.22

VW 26/1 936 0.79 5.48 −0.54 6.72 −34.22 32.41
EW 26/1 936 1.04 7.21 −0.29 8.40 −37.37 50.38
IBM 26/1 936 1.16 7.02 −0.15 2.04 −30.37 30.29
Intel 73/1 372 1.80 13.37 −0.60 2.90 −59.54 48.55
3M 46/2 695 1.16 6.43 −0.06 1.25 −32.61 22.95
Microsoft 86/4 213 2.66 11.48 −0.01 1.19 −42.09 41.58
Citi-Group 86/11 206 1.73 9.55 −0.65 2.08 −42.28 23.18
a Returns are in percentages and the sample period ends on December 31, 2003. The statistics are defined
in eqs. (1.10)–(1.13). VW, EW, and SP denote value-weighted, equal-weighted, and S&P composite
index.
DISTRIBUTIONAL PROPERTIES OF RETURNS 19

f
Table 1.2 provides some descriptive statistics of simple and log returns for
selected U.S. market indexes and individual stocks. The returns are for daily and
monthly sample intervals and are in percentages. The data spans and sample sizes
are also given in the table. From the table, we make the following observations.
(a) Daily returns of the market indexes and individual stocks tend to have high
excess kurtosis. For monthly series, the returns of market indexes have higher
excess kurtosis than individual stocks. (b) The mean of a daily return series is
o
close to zero, whereas that of a monthly return series is slightly larger. (c) Monthly
O
returns have higher standard deviations than daily returns. (d) Among the daily
O
returns, market indexes have smaller standard deviations than individual stocks.
This is in agreement with common sense. (e) The skewness is not a serious prob-
f
lem for both daily and monthly returns. (f) The descriptive statistics show that the
difference between simple and log returns is not substantial.
o
Figure 1.4 shows the empirical density functions of monthly simple and log
returns of IBM stock. Also shown, by a dashed line, in each graph is the nor-
mal probability density function evaluated by using the sample mean and standard
deviation of IBM returns given in Table 1.2. The plots indicate that the normality
assumption is questionable for monthly IBM stock returns. The empirical den-
sity function has a higher peak around its mean, but fatter tails than that of the
corresponding normal distribution. In other words, the empirical density function is
⼀
taller and skinnier, but with a wider support than the corresponding normal density.

0.06 0.06

0.05 0.05

0.04 0.04
density

density

0.03 0.03

0.02 0.02

0.01 0.01

0.0 0.0

−40 −20 0 20 40 −40 −20 0 20 40

simple return log return

o
Figure 1.4. Comparison of empirical and normal densities for the monthly simple and log returns of
IBM stock. The sample period is from January 1926 to December 2003. The left plot is for simple
returns and the right plot for log returns. The normal density, shown by the dashed line, uses the sample
mean and standard deviation given in Table 1.2.
 Characteristics

Non
norm n
i_

A L

high_peak around0
尖峰厚尾
think tails
 returns
log NOT indep.li
But
Manninen
i of

Carey of

Y fndttffsd 是

trend noise
Chapter 2

Brownian Motion

2.1 Why Brownian motion?

Given a financial time series (e.g., stock prices, bond price, interest rates), many di↵erent types
of stochastic processes may be used to model financial time series. In this course, we shall focus
on Brownian motion. In fact, it belongs to every single class listed above. As can be seen later
in the course, it is also adequate.

2.2 Brownian Motion (BM)

Definition 2.1 A stochastic process {Bt , t 0} is a (standard) Brownian Motion (BM) if

(a) Bt has independent increments;

(b) B0 = 0, and Bt Bs ⇠ N (0, t s) for every 0  s  t;
(c) The path of Bt is continuous a.s. (i.e., no jumps.)

BM is related to Gaussian process.

Theorem 2.1 (An alternative definition) {Bt , t 0} is a BM i↵

(a0 ) Bt is a Gaussian process, i.e., (Bt1 , ..., Btn ) are multivariate normal for all t1 , ..., tn .
(b0 ) EBs = 0 and EBs Bt = s ^ t = min{s, t}.
(c0 ) The path of Bt is continuous a.s.

Proof. Suppose that 0 = t0  t1  t2  ...  tn < 1.

4
 • “=)”. Assume that Bt is BM.
(a0 ) holds since (Bt1 Bt0 , ..., Btn Btn 1 ) are multivariate Gaussian, so are its linear
combination (Bt1 , ..., Btn ).
To get (b0 ), suppose that s  t, then EBt = 0, and EBs Bt = EBs2 + EBs [Bt Bs ] =
EBs2 = s = s ^ t.
• “(=” Assume that (a0 )-(c0 ) hold.
From (a0 ), (Bt1 , ..., Btn ) is Gaussian, so is its linear combination (Bt1 Bt0 , ..., Btn Btn 1 ).
The components are uncorrelated (hence independent for Gaussian), since, e.g., E(Bt1
Bt0 )(Bt2 Bt1 ) = t1 t1 t0 + t0 = 0 from (b0 ). This proves (a).
Next, from (b’), we have E(Bt+s Bs ) = 0 and E(Bt+s Bs )2 = EBt+s 2
+ EBs2
2E(Bt+s Bs ) = (t + s) + s 2s = t. So Bt+s Bs ⇠ N (0, t). Finally, taking s = t = 0 in
(b’) gives B0 = 0.

Therefore, any continuous-time Gaussian process {Xt , t 0} with zero-mean and co-
variance Cov(Xs , Xt ) = s ^ t is a BM. This is often useful when checking if a process is a
BM.

Theorem 2.2 {Bt , t 0} is a BM. So are the following

1. {Bt+t0 B t0 , t 0}, 8t0 0; (shifted BM)

2. { Bt , t 0}; (reflected BM)
3. {cBt/c2 , t 0}, 8c 6= 0; (rescaled BM)
4. {tB1/t , t 0}, where tB1/t is taken to be zero when t = 0.
(time reversed BM: 0 ! 1; 1 ! 0)

Proof. We only prove the last one since others are simpler. Write B̃t = tB1/t . (The continuity
at 0 for tB1/t needs to be carefully investigated as done below).

• (B̃t1 , ..., B̃tn ) = (t1 B1/t1 , ..., tn B1/tn ) are clearly Gaussian.

• E(B̃t ) = E(tB1/t ) = tEB1/t = 0, E(B̃s B̃t ) = E(sB1/s tB1/t ) = st(1/s ^ 1/t) = s ^ t.

• Clearly, B̃t is continuous for all t > 0. To show continuity at t = 0, it suffices to show that
P (lims!0 sB(1/s) = 0) = P (limt!1 B(t)/t = 0) = 1. The proof involves some advanced
analysis. We omit it here.

Remark 2.1 BM lies right in the heart of many important processes.

1. BM is a Markov process. To see this, note Bt+s Bs ? Fs and

P (Bt+s  a|Bs = x, Bu , 0  u < s) = P (Bt+s Bs  a x|Bs = x, Bu , 0  u < s)
= P (Bt+s Bs  a x) = P (Bt+s  a|Bs = x).

5
 2. BM is a martingale since E(Bt+s |Fs ) Bs = E(Bt+s Bs |Fs ) = 0.
3. BM is a Gaussian process from Theorem 2.1.
4. BM is a Levy process. Xt is a Levy process if it has stationary, independent increments,
and Xt is stochastically continuous (weaker than continuous a.s.).
5. BM is a di↵usion process which is a continuous-time Markov process with almost surely
continuous sample paths, e.g. BM, reflected BM, and O-U process.
6. BM is 0.5-self-similar, i.e., Brownian motion {Bt , t 0} is 0.5-self-similar, i.e.,
(Bct1 , ..., Bctn ) =d (c1/2 Bt1 , ..., c1/2 Btn )
for every c > 0, and any choice of 0  t1  ...  tn < 1 and n 1.
Proof. Both sides are multivariate Gaussian with the same mean functions zero, and the
same covariance functions since
Cov(Bcti , Bctj ) = E(Bcti Bctj ) = ((cti ) ^ (ctj )) = c (ti ^ tj ) ,
Cov(c1/2 Bti , c1/2 Btj ) = cCov(Bti , Btj ) = cE(Bti Btj ) = c (ti ^ tj ) .

2.3 Some properties of Brownian motion

Brownian motion is nowhere di↵erentiable

Theorem 2.3 (Non-di↵erentiability in probability)

!
|Bt+ Bt |
8M > 0 : lim P >M = 1.
&0

⇣ ⌘ ⇣ ⌘
1/2 1/2 1/2
Proof. P (|Bt+ Bt |/ > M ) = P |B |/ >M = P |B1 | > M ! 1.

Theorem 2.4 (Non-di↵erentiability a.s.)

|Bt+ n Bt |
lim sup = 1, a.s.
n &0 n

Proof. Write An = {|Bt+ n Bt |/ n > M }. 8M > 0, it suffices to show:

P (An , i.o.) = 1.
Now
1 [
1
! 1
! 1
!
\ [ [
P (An , i.o.) = P Ak = P lim Ak = lim P Ak
n!1 n!1
n=1 k=n k=n k=n
lim sup P (An ) = lim sup P (|B n / n| > M)
n!1 n!1
⇣ ⌘
1/2
= lim sup P |B1 | > n M
n!1
= 1.

6
Remark 2.2 Weierstrass function below is continuous but nowhere di↵erentiable, where
1
X
f (x) = an cos(bn ⇡x), 0 < a < 1, ab 1.
n=0
P1 n
For example, letting a = 1/2 and b = 5, then f (x) = n=0 2 cos(5n ⇡x).

Brownian motion is unbounded

Theorem 2.5 We have lim supt |Bt | = 1, a.s.

Proof. If suffices to show that 8M > 0, we have P (lim supn |Bn | > M ) = 1. Now

P (lim sup |Bn | > M ) = P (|Bn | > M, i.o.)

n
1 [
1
!
\
=: P {|Bk | > M }
n=1 k=n
1
!
[
= P lim {|Bk | > M }
n!1
k=n
1
!
[
= lim P {|Bk | > M }
n!1
k=n
lim P (|Bn | > M )
n!1
⇣ p ⌘
= lim P |B1 | M/ n
n!1
= 1.

A stronger result holds: it is not bounded above and below a.s.

Theorem 2.6 We have P (supt Bt = +1, inf t Bt = 1) = 1.

Brownian motion has unbounded 1-variation

Let g be a function on [0, t] and ⌧n a sequence of partitions of [0, t]: 0 = t0 < ... < tn = t.
Define

• ig = g(ti ) g(ti 1 ), the ith increment of g,

• i =: it = ti ti 1 , the ith time interval,

• n = max0in i, the mesh of the partition ⌧n ,

Definition 2.2

7
 • The variation of g is
n
X
Vg (t) =: sup |g(ti ) g(ti 1 )|.
⌧n i=1

More generally, the p-variation of g is

n
X
Vg(p) (t) =: sup |g(ti ) g(ti 1 )|p .
⌧n i=1

• Quadratic variation of g is
n
X
[g, g](t) = lim |g(ti ) g(ti 1 )|2 .
n !0
i=1

Remarks

• The two definitions are di↵erent but related. In fact, (see Theorem 2.8)

Vg(2) (t) [g, g](t)

• The variation Vg (t) plays an important role in calculus when defining R-S integrals.

• The quadratic variation [g, g](t) plays an important role in stochastic calculus.

• [g, g](t) is of no use in calculus as [g, g](t) = 0 if g is continuous and of finite variation.

Theorem 2.7

1. [B, B]t = t.

2. VBt (!) ([0, t]) = 1 a.s.

Pn 2
Proof. Denote Qn = i=1 ( i B) . Note iB = B ti B ti 1
⇠ N (0, i ), 1  i  n, are
independent.

p
1. Hence i B/ ⇠ N (0, 1) and ( i B)2 /
i i ⇠ 2
1, i.e., ( i B)
2
⇠ ( 2
i) 1. Hence,
E( i B)2 = i and V ar( i B)2 = 2 2i .
n
X n
X
2
EQn = E( i B) = i = t,
i=1 i=1
n
X n
X n
X
2 2 2
E(Qn t) = V ar[Qn ] = var[( i B) ]=2 i 2 n i = 2t n ! 0.
i=1 i=1 i=1

Thus Qn (!) ! t in L2 (hence in probability).

8
 P liihEP
B. I 1
凾

問 三 Editio
Modes of Convergence in Bob

xn ixx MO
pllxn XPD
limPCIXn y.sk
nes
Xn X

P 蕊 X_X o

KǛX
lim El X_x N 0
n_n

particular Eonr
t

Ehxi
x_x
Ep ⼀
腿
t ll

ii
 X r
Xn
2. Suppose that VBt (!) ([0, t]) < 1 for a given !. Thus, F
n
X n
X
2
Qn (!) =: ( i B(!))  max | i B(!)| | i B(!)|
1in
i=1 i=1
 max | i B(!)| VBt (!) ([0, T ]).
1in

Since Bt is continuous a.s., so it is uniformly continuous a.s. on [0, 1], which in combination
iii
with mesh(⌧n ) ! 0, implies that maxi=1,...,n | i B(!)| ! 0 a.s. From the assumption that
VBt (!) ([0, T ]) < 1, we get

ru Qn (!) ! 0.

On the other hand, from the first part, Qn (!) !p t, hence Qnk (!) ! t a.s. for some
(3.1)

subsequence {nk , k 1}. This contradicts with (3.1). The theorem is proved.

of
R1
0
bounded
tdB ,
t 0
R1
B
TSt R
Remark 2.3 R-S integrals of the type E f (x)dg(x) assumes that the integrand, g(x), is
variation (B.V.). However, since BM is NOT of B.V. a.s., integrals such as
t dBt are not defined. New integrals must be introduced.

tmi
2.4
SLLN
Processes derived from Brownian motion

A Brownian bridge is Xt =d (Bt |{B0 = 0, B1 = 0}), t 2 [0, 1].

x
• Xt is a Gaussian process with Xt ⇠ N (0, t(1
with zero uncertainty at the nodes.
t)). So the middle section is most uncertain

Furthermore, EXt = 0, Cov(Xs , Xt ) = s(1 t) for s < t. The increments in a Brownian

bridge are not independent.

• A simple way to construct a Brownian bridge is from a BM:

CLT Xt = B t tB1 , 0  t  1.

A Brownian motion with (linear) drift µ is

dT
Xt = µt + Bt , t 0.

EXt = µt, sNGD

Clearly, Xt is a Gaussian process with mean and covariance functions:

Cov(Xs , Xt ) = 2
min(s, t), s, t 0.

Bachelier (1900) first used this to model stock prices, however, it can be negative.

DT
A geometric Brownian motion is

Xt = X0 eµt+
ft 9
Bt⾼ , t 0.

9
 然
Lo
 Bt is
moment generating function is
continuous
Clearly Xt is NOT a Gaussian process as the d.f. of Xt is on

Z Z
0 B1 ⇠ N (0, 1), its
log-normal. Since

1 2
t
MB1 ( ) = Ee B1 = e x (x)dx = e x p e x /2 dx
2⇡
Z
2 1 2
= e /2 p e (x ) /2 dx
2⇡
2 /2
= e .
i
By 0.5-self-similarity of BM, Bt =d t1/2 B1 , we have
t1/2 B1 it 2 t/2 2 )t
EXt = Eeµt+ Bt
= eµt Ee = eµt MB1 ( t1/2 ) = eµt e = e(µ+0.5 ,
and, for s  t,
Cov(Xs , Xt ) = E(Xs Xt ) (EXs )(EXt )
µ(t+s)+ (Bt +Bs ) (µ+0.5 2
)(t+s)
= Ee e
2
= eµ(t+s) Ee (Bt Bs +2Bs ) e(µ+0.5 )(t+s)
2
= eµ(t+s) Ee (Bt Bs ) Ee2 Bs e(µ+0.5 )(t+s)
2
= eµ(t+s) Ee Bt s Ee2 Bs e(µ+0.5 )(t+s)
2 2 2
= eµ(t+s) e(0.5 )(t s) e(0.5(2 ) )s e(µ+0.5 )(t+s)

uuifg
2 2
= eµ(t+s) e(0.5 )(t+3s) e(µ+0.5 )(t+s)
2 2 2 )(t+s)
= e(µ+0.5 )(t+s) e⇣ s e(µ+0.5 ⌘
2 2
= e(µ+0.5 )(t+s) e s 1 .
2 )t
⇣ 2t
⌘
In particular, var(Xt ) = e(2µ+ e 1 .

2.5 Bti
Appendix:Bti KE
Variation of a function
伶 as long as

Theorem 2.8 For p11

Pn
2 (0, 1], we have Vg(p) ([a, b]) = lim n !0 i=1 |g(tni ) g(tni 1 )|p . (The result
may not hold for p > 1.)
B
Proof. For p 2 (0, 1] and c 2 [a, b], by the triangle inequality:
|a b|p  |a c|p + |c b|p . (5.2)
Pn
the sums Vg(p) ([a, b]) = sup⌧n i=1 |g(tni ) g(tni 1 )|p increase as new points are added to the
partitions. The rest of proof follows easily.

The above theorem may not hold true for p > 1. Take p = 2 for example. Note that
(a b)2 = (a c)2 + (c b)2 + 2(a c)(c b), hence, we may NOT have a similar triangular
inequality: |a b|2  |a c|2 + |c b|2 .

Remark 2.4 To show (5.2), it suffices to show that |a b|1/m  |a c|1/m + |c b|1/m for
integer m > 0, which follows from |a b|  |a c| + |c b|  (|a c|1/m + |c b|1/m )m .

Remark 2.5 In real analysis, the most interesting variation is the 1-variation. In stochas-
tic analysis, 1-variation is usually 1, so one uses quadratic variation instead.

10
Some examples of variations

1. Monotone functions
If g(t) is increasing, Vg (t) = g(t) g(0).

2. Continuously di↵erentiable functions

Rt Rt
If g 0 (t) is continuous, and 0 |g 0 (s)|ds < 1, then Vg (t) = 0 |g 0 (s)|ds.
Proof. By Theorem 2.8 and the mean value theorem, we have
n
X n
X Z t
Vg (t) = lim |g(tni ) g(tni 1 )| = lim |g (⇠˜i )|(tni
0
tni 1 ) = |g 0 (s)|ds.
n !0 n !0 0
i=1 i=1

3. Pure jump functions

If g is a regular right-continuous (càdlàg) function or regular left-continuous (càglàd), and
P
changes only by jumps (i.e., step functions) g(t) = 0st g(s), then it is easy to see
from the definition that X
Vg (t) = | g(s)|.
0st

4. A function of infinite variation

The function g(t) = t sin(1/t) for t > 0, and g(0) = 0 is continuous on [0, 1], di↵erentiable
at all points except zero, but has infinite variation on any interval that includes zero.
Take the partition 1/(2⇡k + ⇡/2), 1/(2⇡k ⇡/2), k = 1, 2, ....

Necessary and sufficient conditions for a function to have finite variation are

Theorem 2.9 (Jordan Decomposition) If g : [0, 1) ! R is of finite variation, then g(t) =

g1 (t) g2 (t), where g1 and g2 are increasing functions.

11
2.6 Exercises
1. Compute E[Bt1 Bt2 Bt3 ] for t1 < t2 < t3 .

2. Let Zt be a Brownian bridge process, i.e. Zt = Bt tB1 . Show that if

Xt = (t + 1)Zt/(t+1) ,

then {Xt , t 0} is a Brownian motion process.

3. • A necessary and sufficient condition for a Gaussian process Xt to be strongly sta-

tionary is that EXt = c and Cov(Xs , Xt ) depends only on t s, s  t.
• Show that Vt = e ↵t/2 B(e↵t ) for t 0 is a stationary Gaussian process. It is called
the Ornstein-Uhlenbeck process. [Hint: Use (a).]
R1 Pn
4. Find the d.f. of Y = 0 Bt dt. (Hint: Y = limn i=1 B ti i, take ti = i/n.)

12