Lecture 1

1
G63.2707 - Financial Econometrics and Statistical Arbitrage

Farshid Magami Asl
Lecture 1. 1
Financial Econometrics and Statistical
Arbitrage
Master of Science Program in Mathematical Finance
New York University
Introduction on Time Series Analysis
Building Blocks
Fall 2011
Copyright Protected (Do Not Copy)
Farshid Magami Asl
Lecture 1. 2
Market Microstructur Theory
(Transaction costs and Optimal Control, Algorithmic Trading,)
Risk Management
(Practical Risk Measurement and Management Technics)
Financial Econometrics
(Time Series Review and Volatility modeling)
Strategies and Implementation Process
(Cointegration based pairs trading, Volatility trading, )
What is Statistical Arbitrage?
Statistical Arbitrage covers any trading strategy which
uses statistical tools and time series analysis to identify
approximate arbitrage opportunities while evaluating the
risks inherent in the trades considering the transaction
costs and other practical aspects.
Arbitrage is a riskless profit. Arbitrage Strategy is a
trading strategy that locks in a riskless profit.
2
Farshid Magami Asl
Lecture
Quantitative Trading
Quantitative Trading Strategies
Strong Market Forces Weak Market Forces
Law of One Price
No Arbitrage
Portfolio Replication
Market Identities:
e.g. Put-Call Parity,
Cross Market Arbitrage
Converts
Statistical Relationships
between assets:
Co-integrated Pairs Trading,
Volatility Trading
Mean-Revesion
Factor Models,
Market Anomalies Statistical Arbitrage
Size Effect (Banz, 1981)
Value Effect (Ball, 1978)
January Effect( Roll, 1983)
Momentum and other
Technical Effects
Behavioral Finance
There are three general types of analysis used in finance and trading
1. Fundamental Analysis
2. Technical Analysis
3. Quantitative Analysis
1. 3
Farshid Magami Asl
Lecture
Trading based on Statistical Arbitrage
Strategies are mostly designed based on:
1. Measuring a signal (observable or unobservable) with some properties (
mainly Mean-Reversion )
2. Portfolio effects and Central Limit Theorem.
The Key is to gather many marginally profitable strategies with
low correlations among them
1. 4
Reminder: Central Limit Theorem (CLT)
Let R
1
, R
2
, R
3
, , R
n
be a sequence of n independent and identically distributed (iid) random variables, representing returns on each
asset i (i=1,2,3,,n), each having finite values of expectation (which we like it to be greater than zero) and variance
2
> 0.
The return of a portfolio of these assets is R
P
=R
1
+ R
2
+ + R
n
.
The central limit theorem states that as the sample size n increases the distribution of the sample average of these random
variables approaches the normal distribution with a mean and variance
2
/n irrespective of the shape of the common distribution
of the individual terms R
i
. In other words, the distribution of the portfolio return is N( n , ) n
3
Farshid Magami Asl
Lecture
Trade
$1 Profit
$1 Loss
1 Trade
2 Trades
1000 Trades
-1 0 1 P&L
IR = 0.02
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 P&L
IR = 0.028
-150 -100 -50 0 50 100 150 P&L
IR = 0.632
Trade
$1 Profit
$1 Loss
Trade
$1 Profit
$1 Loss
Trade
$1 Profit
$1 Loss
Trade
$1 Profit
$1 Loss
Trade
$1 Profit
$1 Loss
Trade
$1 Profit
$1 Loss
1. 5
Farshid Magami Asl
Lecture
Effect of Predictive Signals
Number of Trades
0 500 1000 1500 2000 2500 3000
0
0.2
0.4
0.6
0.8
1
1.2
1.4
Information Ratio
Trade
$1 Profit
$1 Loss
Trade
$1 Profit
$1 Loss
Stronger Signals (skills)
0 500 1000 1500 2000 2500 3000
0
1
2
3
4
5
6
Number of Trades
Information Ratio
1. 6
Theoretical
Simulation
4
Farshid Magami Asl
Lecture
Trade
$1 Profit
$1 Loss
Stronger Signals (skills)
Fundamental Elements in Quant Trading:
Number of Trades
Predictive power (Strength) of Signals
Correlation between Signals and hence trades.
Higher Frequency Trading
0 500 1000 1500 2000 2500 3000
0
1
2
3
4
5
6
Number of Trades
Information Ratio
0 500 1000 1500 2000 2500 3000
0
1
2
3
4
5
6
Number of Trades
Information Ratio
10% Correlation
1. 7
Farshid Magami Asl
Lecture Quantitative Trading: Frequency Spectrum
Ultra High
Frequency
Tick Data
Order Book Dynamics
Microstructure Theory
High-Frequency
Seconds - intraday
Statistical Inference
Time Series Analysis
Mid-Frequency
Day - week
Combination of
Statistical and
Some Fundamental Factors
T-S & X-sectional Analysis
Statistics
Low-Frequency
Week, Month
and longer
Price Anomalies/
Asset Pricing Theory/
Maco Forces
Xsection Variations/
EquilibriumMethods
Frequency
C
a
p
a
c
i
t
y
Market Making Market Taking
1. 8
5
Farshid Magami Asl
Lecture Typical Behavior of Financial Assets
The unpredictability inherent in asset prices is the main feature of financial modeling.
Because there is so much randomness, most mathematical models of a financial asset
acknowledge the randomness and have a probabilistic foundation.
Financial assets show Dynamic behavior.
Time in minutes (12/03/2007)
10 randomly chosen stocks in Dow Jones Index
Dow Jones Index
Time in days from1/1/1975 to 07/30/2005
31-Dec-1974 11-Mar-1985 21-May-1995 30-Jul-2005
0
2000
4000
6000
8000
10000
12000
1. 9
Farshid Magami Asl
Lecture 1. 10 Introduction to Financial Modeling
Return in financial assets
By return we mean the percentage growth in the value of an asset,
together with accumulated dividends, over some period:
Change in value of the asset + accumulated cashflows
Original value of the asset
Return =
Denoting the asset value on the i-th day by Si, the return from day i
to day i+1 is given by
i
i i
i
S
S S
R

=
+1
6
Farshid Magami Asl
Assume that the empirical returns are close enough to Normal for this to be a
good approximation.
For start, we write the returns as a random variable drawn from a Normal
distribution with a known, constant, non-zero mean and a known, constant, non-
zero standard deviation:
i
i i
i
S
S S
R

=
+1
= mean + standard deviation x f
2
2
1
2
1
) 1 , 0 (
|
t

= e N
Time in days from 2/1/1975 to 07/29/2005
D
iffe
r
e
n
c
e
o
f
th
e
L
o
g
T
r
a
n
s
fo
r
m
o
f
D
o
w
J
o
n
e
s
In
d
e
x
02-Jan-1975 12-Mar-1985 21-May-1995 29-Jul-2005
-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
0.1
0.15
Farshid Magami Asl
i
i i
i
S
S S
R

=
+1
= mean + standard deviation x f
| oo o + =
=
+ 2 / 1 1
t t
S
S S
R
i
i i
i
Time scale dt
t o
Mean return over
period dt is
2 / 1
t oo
Standard deviation over
period dt is
dX (Wiener Process)
And in the limit dt 0
t
t
t
dX dt
S
dS
R o + = =
7
Farshid Magami Asl
Lecture 1. 13
0
20
40
60
80
100
1
2
3
4
5
0.9
1
1.1
1.2
1.3
x 10
4
Basic Review
STOCHASTIC PROCESS:
A stochastic process is a collection of random variables defined on a
probability space . ( ) P F , , O
{ } t e e t X
t
), (
State 2
State 1
State 3
State 4
State 5
O
e
e For a fixed , a realization of stochastic process is a function of time (t).
) , ( e t X
Farshid Magami Asl
Lecture 1. 14 Simulation of a Stochastic Process
0
20
40
60
80
100
0
50
100
150
200
0.6
0.8
1
1.2
1.4
1.6
x 10
4
8
Farshid Magami Asl
Lecture 1. 15 Definition
Time Series:
A time series, , is a stochastic process where t is a set of
discrete points in time . In other words, it is a discrete time,
continuous state process.
In this course we consider t = { all integers}
X1 X2 X3
Xk
k
0 5 10 15 20 25 30 35 40
-4
-3
-2
-1
0
1
2
3
{ } t e t t X ), (
{ } , 3 , 2 , 1 = t
Farshid Magami Asl
Lecture 1. 16
We want to forecast distributions
Goals of Studying Time Series
1- Forecasting
0
2000
4000
6000
8000
10000
12000
D
o
w
J
o
n
e
s
I
n
d
e
x
2- Understanding the statistical characteristics and building
trading strategies based on them
31-Dec-1974 11-Mar-1985 21-May-1995 30-Jul-2005
9
Farshid Magami Asl
Lecture Modeling a Dynamic system
State Space Model
Mathematical Model
) ( ) (
1 ) (
t u t y
R dt
y dy
A + =
) (
1
) (
1
) ( t u
A
t y
RA
t y + =
-
System
u y
Area (A)
u
Resistance (R)
y
Output
Input
Parameters
From Physical Laws,
) (
1
) (
1
) ( ) 1 ( t u
A
t y
RA
t y t y + = +
) (
1
) ( ) 1
1
( ) 1 (
1 1
t u
A
t y
RA
t y
|
+ + = +

) ( ) ( ) 1 (
1 1
t u t y t y | + = + ) (t w +
Dynamic System
) (t w
Stochastic
Information
(Signal)
Noise
1. 17
Farshid Magami Asl
Lecture Modeling a Dynamic system
u
y
Output
Input
) ( ) ( ) 1 (
1 1
t u t y t y | + = +
0 50 100 150 200 250 300
-10
-5
0
5
10
0 50 100 150 200 250 300
-10
0
10
20
30
In reality, physical modeling could be difficult or impossible, and we have to work with the observed
data
System
w
y
Output
0 50 100 150 200 250 300
-10
0
10
20
30
In financial systems, we cant measure inputs either
Financial
System
w
1. 18
10
Farshid Magami Asl
Lecture 1. 19 Basic Review
An Example of a Time Series:
0 1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0 6 0 0 0 7 0 0 0
- 1 0
- 8
- 6
- 4
- 2
0
2
4
6
8
1 0
Xk
k
-10 -8 -6 -4 -2 0 2 4 6 8 10
-10
-8
-6
-4
-2
0
2
4
6
8
10
Xk
Xk-1
-10 -8 -6 -4 -2 0 2 4 6 8 10
-10
-8
-6
-4
-2
0
2
4
6
8
10
Xk
Xk-2
-10 -8 -6 -4 -2 0 2 4 6 8 10
-10
-8
-6
-4
-2
0
2
4
6
8
10
Xk
Xk-3
-10 -8 -6 -4 -2 0 2 4 6 8 10
-10
-8
-6
-4
-2
0
2
4
6
8
10
Xk
Xk-10
Xk= j1 Xk-1+ j2 Xk-2++ek
Auto-Regression as a Dynamic System?
We will get back to this
Farshid Magami Asl
Autocovariance Function:
Let {Xt} be a time series. The autocovariance function of process {Xt} for all
integers r and s is:
) , cov( ) , (
s r X
X X s r =
))] ( ))( ( [( ) , (
s s r r X
X E X X E X E s r =
)] ( ) ( ) ( ) ( [ ) , (
s r r s s r s r X
X E X E X E X X E X X X E s r + =
) ( ) ( ) ( ) ( ) ( ) ( ) ( ) , (
s r r s s r s r X
X E X E X E X E X E X E X X E s r + =
) ( ) ( ) ( ) , (
s r s r X
X E X E X X E s r =
0 ) var( ) ( ) ( ) , (
2 2
> = =
r r r X
X X E X E r r Note that
= 0
11
Farshid Magami Asl
Lecture 1. 21 Autocovariance Function
0 2 4 6 8 10 12 14 16 18 20
Lag
Sample Autocovariance Function
) 0 (
X
) 2 (
X
) 1 (
X

Farshid Magami Asl
Stationary Process:
A time series {Xt} is stationary (weakly) if:
) , ( ) , ( . 3
) ( . 2
) ( . 1
2
t s t r s r
X E
X E
X X
t
t
+ + =
=
<

Some constant m for all t
i.e. Cov(Xr,Xs) only depends on r and s and not on t.
) ( ) 0 , ( ) , ( ) , ( s r s r s s s r s r
X X X X
= = =
Note: If {Xt} is stationary, then is a function of
) , cov( ) ( ) (
h t t X X
X X h s r
+
= =
Define h=r-s
Does not depend on t
Astrict (strong) stationary time series
{Xt , t=1,2,,n}
is defined by the condition that realizations
(X1, X2, , Xn) and (X1+h, X2+h, , Xn+h)
have the same joint distributions for all
integers h and n>0.
Note:
) , ( s r
X
) ( s r
12
Farshid Magami Asl
Note:
Strict Stationary
(Strong)
Weak Stationary
(Covariance Stationary)
Not generally true except for the Gaussian processes
Any strictly stationary process which has a mean and a
covariance is also weakly stationary
Farshid Magami Asl
Lecture 1. 24 Stationary Process
Stationary Process and Mean Reversion
We are interested in stationary time series because many models and
tools are developed for stationary processes.
A stationary process can never drift too far from its mean because of
the finite variance. The speed of mean-reversion is determined by the
autocovariance function: Mean-reversion is quick when autocovariances
are small and slow when autocovariances are large.
Trends and periodic components make a time series non-stationary.
13
Farshid Magami Asl
Lecture 1. 25 Stationary Process
0 50 100 150 200 250 300
-20
0
20
40
60
80
0 50 100 150 200 250 300
-6
-4
-2
0
2
4
Stationary Process
Non-Stationary Process
Mean-Reversion
Farshid Magami Asl
Lecture 1. 26 General Approach to Time Series
Time Series Analysis
1. Plot time series and check for trends or sharp changes in behavior
(most of the time non-stationary)
2. Transform into a stationary time series
3. Fit a model
4. Perform diagnostic tests (residual analysis,)
5. Generate forecasts (find predictive distributions) and invert the
transformations performed in 2.
Note for option pricing:
6. Find a risk neutral version of the model
7. Obtain predictive distributions under the risk neutral model
If bad
14
Farshid Magami Asl
Lecture 1. 27 Building Blocks of Financial Models
White Noise Process
=
=
otherwise
s r
s r
X
0
) , (
2
o
0 1000 2000 3000 4000 5000 6000 7000

-5
0
5
W
N
If {Xt} is a sequence of random variables with , and
0 ) ( =
t
X E
) (
2
< o
2
o
{Xt} is called White Noise and it is written as WN(0, )
2 2
) ( o =
t
X E
Note that E[Xt Xs]=0 for t=s Uncorrelated r.v.s
2
o If Xt and Xs independent for t=s IID(0, )
-5 0 5
-5
0
5
Xk
Xk-1
Farshid Magami Asl
0 ) ( =
t
X E
) (
2
< o
=
=
otherwise
s r
s r
X
0
) , (
2
o
White Noise Process (Is it Stationary?)

15
Farshid Magami Asl
Random Walk Process
0 1000 2000 3000 4000 5000 6000 7000
-100
0
100
R
a
n
d
o
m
W
a
l
k
If {Xt} be a sequence of random variables , a sequence {St}
with S0=0 and
Is called a Random Walk.
2
o IID(0, )
=
=
t
j
j t
X S
1
(Integrated Process)
-10 -5 0 5 10
-10
-5
0
5
10
Sk
Sk-1
Farshid Magami Asl
Random Walk Process (Is it Stationary?)
2
o
IID(0, )
=
=
t
j
j t
X S
1
{Xt}
16
Farshid Magami Asl
Moving Average Process
Let {Xt} be WN(0, ), and consider the process
Where q could be any constant. This time series model is called a first-
order moving average process, denoted MA(1).
The term Moving Average comes from the fact that Yt is constructed
from a weighted sum of the two most recent values of Xt.
1
+ =
t t t
X X Y u
2
o
Yk
-4 -2 0 2 4
-4
-2
0
2
4
Yk-1
0 1000 2000 3000 4000 5000 6000
q =0.5
Farshid Magami Asl
Moving Average Process (Is it Stationary?)
{Xt} is WN(0, )
1
+ =
t t t
X X Y u
2
o
17
Farshid Magami Asl
Autoregressive Process
Let {Zt } be WN(0, ), and consider the process
Where |f |<1 and Zt is uncorrelated with Xs for each s<t. This time series
model is called a first-order Autoregressive process, denoted AR(1).
2
o
t t t
Z X X + =
1
|
It is easy to show that E(Xt)=0
0 100 200 300 400 500 600 700
-5
0
5
f=0.7
-10 -5 0 5 10
-10
-5
0
5
10
Xk
Xk-1
Farshid Magami Asl
Autoregressive Process (Is it Stationary?)
{Zt } is WN(0, ), and
Where |f |<1 and Zt is uncorrelated with Xs for each s<t.
2
o
t t t
Z X X + =
1
|
We will see this later
18
Farshid Magami Asl
t t t
Z X X + =
1
|
0 50 100 150 200 250 300
-10
0
10
20
30
-10 0 10 20 30
-10
0
10
20
30
f = 1
Random Walk
0 50 100 150 200 250 300
-10
-5
0
5
10
-10 -5 0 5 10
-10
-5
0
5
10
f = 0.9
AR(1)
0 50 100 150 200 250 300
-4
-2
0
2
4
-4 -2 0 2 4
-4
-2
0
2
4
f = 0.1
AR(1)
Farshid Magami Asl
Lecture 1. 36 Transforming a Non-Stationary Process to a Stationary Process
Classical Decomposition
t t t t
Y S m X + + =
Original
Time series
(Nonstationary)
Trend
Seasonal
component
Stationary
Time series
(zero-mean)
=
=
d
j
j
S
1
0
Seasonal component St satisfies
St+d=St where d= period of seasonality
Also for mathematical convenience assume
Most observed time series are non-stationary but they can be
transformed to stationary processes.
19
Farshid Magami Asl
Classical Decomposition
t t t t
Y S m X + + =
t t
t t
S m X X
^ ^
*
+ =
Idea of transformation is to estimate mt and St by mt and St, then work
with the stationary process:
Assume there is no seasonal component (St=0)
t t t
Y m X + =
2
2 1 0
^
t a t a a mt + + =
Consider a parametric form for mt e.g.
2
1
^
) (
=

n
t
t
t
m X
Using observed data X1, X2, Xn, choose a0, a1, a2 to minimize
Farshid Magami Asl
0 1000 2000 3000 4000 5000 6000 7000 8000 9000
0
2000
4000
6000
8000
10000
12000
D
o
w
J
o
n
e
s
I
n
d
e
x
20
Farshid Magami Asl
0 1000 2000 3000 4000 5000 6000 7000 8000 9000
5
5.5
6
6.5
7
7.5
8
8.5
9
9.5
10
L
o
g
T
r
a
n
s
f
o
r
m
o
f
D
o
w
J
o
n
e
s
I
n
d
e
x
Farshid Magami Asl
0 1000 2000 3000 4000 5000 6000 7000 8000 9000
5
5.5
6
6.5
7
7.5
8
8.5
9
9.5
10
L
o
g
T
r
a
n
s
f
o
r
m
o
f
D
o
w
J
o
n
e
s
I
n
d
e
x
t m
t
0004 . 0 1513 . 6 + =
21
Farshid Magami Asl
0 1000 2000 3000 4000 5000 6000 7000 8000 9000
-0.3
-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
0.1
D
i
f
f
e
r
e
n
c
e
o
f
t
h
e
L
o
g
T
r
a
n
s
f
o
r
m
o
f
D
o
w
J
o
n
e
s
I
n
d
e
x
Farshid Magami Asl
Lecture 1. 42
Forecast
Transforming a Non-Stationary Process to a Stationary Process
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
-0.3
-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
0.1
0.15
F
o
r
e
c
a
s
t
o
f
t
h
e
m
o
d
e
l
22
Farshid Magami Asl
Lecture 1. 43
Forecast
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
6
6.5
7
7.5
8
8.5
9
9.5
10
C
o
n
v
e
r
t
b
a
c
k
t
h
e
d
i
f
f
e
r
e
n
c
e
i
n
t
h
e
F
o
r
e
c
a
s
t
o
f
t
h
e
m
o
d
e
l
Farshid Magami Asl
Lecture 1. 44
Forecast
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
0
5000
10000
15000
c
o
n
v
e
r
t
b
a
c
k
t
h
e
L
o
g
o
f
t
h
e
F
o
r
e
c
a
s
t
o
f
t
h
e
m
o
d
e
l
23
Farshid Magami Asl
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
x 10
4
M
o
n
t
e
C
a
r
l
o
S
i
m
u
l
a
t
i
o
n
o
f
t
h
e
F
o
r
e
c
a
s
t
Farshid Magami Asl
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
x 10
4
M
o
n
t
e
C
a
r
l
o
S
i
m
u
l
a
t
i
o
n
o
f
t
h
e
F
o
r
e
c
a
s
t
24
Farshid Magami Asl
0 10 20 30 40 50 60 70 80 90 100
0.8
0.9
1
1.1
1.2
1.3
x 10
4
0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6
x 10
4
0
20
40
60
Farshid Magami Asl
Trend Elimination by Differencing
Definition: Differencing Operator
V
1
= V
t t t
X X X
1
=
t t
X BX
Definition: Backshift Operator B
Therefore
t t t t
X B X X X ) 1 (
1
= = V

Also
t t t
X B B X B X ) 2 1 ( ) 1 (
2 2 2
+ = = V
2 1
2
2

+ = V
t t t t
X X X X
25
Farshid Magami Asl
Definition: Integrated Process of order n. Time series yt I(n) is
integrated of order n if it is non-stationary, but it becomes stationary after
differencing a minimum of n times.
Example: A stationary process is I(0)
Example: Random Walk is I(1)
t t
Y t X + + =
1 0
| |
Note: Difference removes linear trends as well.
Suppose
1 1 0 1 0
) 1 (

+ + = V
t t t
Y t Y t X | | | |
1 1
+ =
t t
Y Y |
Stationary Process
with mean zero
Constant
Note: Difference twice removes quadratic trends.
Warning: Dont difference too much. Error will be magnified in forecasting
Farshid Magami Asl
Differencing when the seasonal component is present
Definition: Lagged Differencing Operator
d
V
d t t t d
X X X

= V
t
d
X B ) 1 ( =
Note:
t
d
t d
X X V = V
t
d
X B ) 1 (
t
d
X B ) 1 (
26
Farshid Magami Asl
Suppose
d t d t t t t d
Y S d t Y S t X

+ + + = V ) (
1 0 1 0
| | | |
t t t t
Y S m X + + =
d t t
S S
+
=
t m
t 1 0
| | + =
Usually d is known
Stationary Process
with mean zero
Constant
) (
1 d t t t d
Y Y d X

+ = V |
Farshid Magami Asl
Lecture 1. 52
Important for
financial models
Other transformations are used to transform a non-stationary process to a
stationary process. Sometimes trends are multiplicative or exponential
instead of additive and random variations are non-Gaussian.
Box and Cox (1964) proposed a general class of transformations:
Box-Cox / Log Transformation
=
>
=
0 ) log(
0
) 1 (
) (
x
x
x f
27
Farshid Magami Asl
0 1000 2000 3000 4000 5000 6000 7000 8000 9000
0
2000
4000
6000
8000
10000
12000
Dow Jones Index
0 1000 2000 3000 4000 5000 6000 7000 8000 9000
5
5.5
6
6.5
7
7.5
8
8.5
9
9.5
10
Log Transform of Dow Jones Index
Farshid Magami Asl
Lecture 2. 54 Properties of Autocovariance Function
Autocovariance Function:
For a stationary time series {Xt}
) , cov( ) (
h t t X
X X h
+
=
Does not depend on t
Properties:
0 ) var( ) 0 ( > =
t X
X
1
h h
X X
s ) 0 ( | ) ( | 2
) ( ) , cov( ) , cov( ) ( h X X X X h
X t h t h t t X
= = =
+ +
3
Symmetric
| ) ( | ) 0 (
)] ( [ ) 0 ( ) 0 (
)] ( [ ) ( ) (
2
2 2 2
h
h
X X E X E X E
X X
X X X
h t t h t t

>
>
>

2
2 2
) ) ( ) ( ( ) ( ) ( Inequality s Schwartz' dx x g x f dx x g dx x f
} } }
>
28
Farshid Magami Asl
Lecture 2. 55 Autocovariance Function
0 2 4 6 8 10 12 14 16 18 20
Lag
Sample Autocovariance Function
) 0 (
X
) 2 (
X
) 1 (
X

Farshid Magami Asl
Lecture 2. 56 Autocorrelation Function
Autocorrelation Function is the normalized version of the autocovariance
function:
) , (
) 0 (
) (
) (
h t t
X
X
X
X X corr
h
h
+
= =
From property : h h
X X
s ) 0 ( | ) ( | 2
1 ) ( s h
X
) 1 ) 0 ( ( =
X
Correlogram is the graph of autocorrelation function which is the scaled version of

the autocovariance graph.
29
Farshid Magami Asl
Lecture 2. 57 Correlogram
0 2 4 6 8 10 12 14 16 18 20
Lag
Sample Autocorrelation Function
1 ) 0 ( =
X
) 2 (
X
) 1 (
X
1
Farshid Magami Asl
Lecture 2. 58 Sample Autocovariance Function
Autocovariance function can be obtained from the time series models.
In practical problems, we do not start with a model, but with the observed (or
realized) data {X1 , X2 , , Xn }.
Time Series Model
) (h
X
) ( h
X
Observed Time Series

Data {X1 , X2 , , Xn }
Sample Autocovariance
Function
Autocovariance
Function
Can be obtained from
30
Farshid Magami Asl
Let {X1 , X2 , , Xn } be observations of a time series.
Sample Mean of the observations is:
Sample Autocovariance of the observations is:
. ) ( ) (
1
) (
| |
1
| |
n h n x x x x
n
h
i
h n
i
h i
< < =

=
+
Sample Autocorrelation of the observations is:

.
) 0 (
) (
) ( n h n
h
h < < =
Note: If you observe n data points, you can only calculate up to

h=n-1.
) (h
X
=
=
n
i
i
x
n
x
1
1
Farshid Magami Asl
0 100 200 300 400 500 600 700
-4
-2
0
2
4
0 2 4 6 8 10 12 14 16 18 20
-0.5
0
0.5
1
Lag
S
a
m
p
le
A
u
t
o
c
o
r
r
e
la
t
io
n
Sample Autocorrelation Function (ACF)
White Noise
0 2 4 6 8 10 12 14 16 18 20
-0.5
0
0.5
1
Lag
S
a
m
p
le
A
u
t
o
c
o
r
r
e
la
t
io
n
0 100 200 300 400 500 600 700
-20
0
20
40
60
Random Walk
31
Farshid Magami Asl
0 100 200 300 400 500 600 700
-10
-5
0
5
10
0 2 4 6 8 10 12 14 16 18 20
-0.5
0
0.5
1
Lag
S
a
m
p
le
A
u
t
o
c
o
r
r
e
la
t
io
n
AR(1) f = 0.9
0 100 200 300 400 500 600 700
-4
-2
0
2
4
6
0 2 4 6 8 10 12 14 16 18 20
-0.5
0
0.5
1
Lag
S
a
m
p
le
A
u
t
o
c
o
r
r
e
la
t
io
n
MA(1) q = 0.5
Farshid Magami Asl
0 2 4 6 8 10 12 14 16 18 20
-1
-0.5
0
0.5
1
Lag
S
a
m
p
le
A
u
t
o
c
o
r
r
e
la
t
io
n
0 100 200 300 400 500 600 700
-10
-5
0
5
10
AR(1) f = -0.9
0 100 200 300 400 500 600 700
-6
-4
-2
0
2
4
0 2 4 6 8 10 12 14 16 18 20
-0.5
0
0.5
1
S
a
m
p
le
A
u
t
o
c
o
r
r
e
la
t
io
n
MA(1) q = -0.5

Lecture 1

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Lecture 1

Uploaded by

Copyright:

Available Formats

1

G63.2707 - Financial Econometrics and Statistical Arbitrage

G63.2707 - Financial Econometrics and Statistical Arbitrage

0 1000 2000 3000 4000 5000 6000 7000

White Noise Process (Is it Stationary?)

G63.2707 - Financial Econometrics and Statistical Arbitrage

Correlogram is the graph of autocorrelation function which is the scaled version of

Observed Time Series

Sample Autocorrelation of the observations is:

Note: If you observe n data points, you can only calculate up to

You might also like