ECON90033 Quantitative Analysis of Finance I

Lecture 1 - Properties of Financial Data

Barry Rafferty

Semester 2, 2015

1 / 43
Course information

Coordinator: Barry Rafferty

Prerequisite:
ECON20003 Quantitative Methods 2 or equivalent

Contact:
Lecture: One 2-hour lecture per week
One 1-hour tutorial per week, beginning in the second week
Tutor consultation hours (see LMS and subject guide)
Lecturer consultation hours (Wednesday 10am-12pm, FBE
352)
Online tutor (link on the LMS)
Lecture recordings (link on the LMS)

2 / 43
Course information

Main reading:
- Lecture slides

Prescribed text:
- None

Optional reading:
Brooks, C. (2014) Introductory Econometrics for Finance, 3rd
Edition, Cambridge

Background reading
Stock, J.H. and Watson, M.H. (2011) Introduction to
Econometrics, 3rd Edition, Addison-Wesley, Boston.
Hill, R.C., Griffiths, W.E. and Lim, G.C. (2011) Principles of
Econometrics, 4th Edition, Wiley.
3 / 43
Course information

Tutorials:
Start in week 2
Sign up now, using the student portal
Tutorial exercises will involve the use of the computer software
package EViews
AIM: Develop practical skills implementing the techniques
learnt in the lectures to applications in finance and properly
interpreting the results

Software and computers:

- EViews is available:
On campus in the tutorial rooms and the faculty computer labs
Off campus through the Citrix server (see subject guide)

Note that it will be necessary to be able to interpret EViews

output in the final exam
4 / 43
Course information

Assessment:
1-hour mid-semester exam in class in week 7 (worth 15%)
Assignment due in week 9 (worth 25%)
2-hour final exam to be held during the exam period (worth
60%)

5 / 43
Mid-semester assignment

Due Thursday of week 9 (September 24) by 5.00pm

Submit online using the assignment tool
Worth 25%
Group assignment: Group sizes can be a minimum of one and
a maximum of 4. People do NOT have to be in the same
tutorial
AIM: Apply techniques learnt in the course to practical
examples using financial data and EViews, report and
interpret the results
Relevant material: Lectures 1-6 and tutorials 1-6
Tutorials will build up the necessary skills in EViews

6 / 43
Course outline
Lecture Schedule (Tentative)

Week Date Topic

1 28 July Properties of Financial Data

2 4 August Linear Regression in Finance

3 11 August Dynamics in the Mean: ARMA Models

4 18 August Stationarity and Unit Root Testing

5 25 August Forecasting and Forecast Evaluation

6 1 September Volatility Modelling: GARCH Models

7 8 September Mid-Semester Exam (1 hour)

8 15 September VAR Models (1): Concept and Specification

9 22 September VAR Models (2): Interpretation

10 6 October Introduction to Cointegration

11 13 October Cointegration and VECM

12 20 October Review and Exam Information

Lecture Slides

7 / 43
Financial data: An introduction
Financial variables: Stock prices, dividends, earnings, stock
returns, exchange rates, commodity prices, bond yields,
trading volumes, etc

Macro variables: GDP, Investment, Consumption, etc

- How do they relate to financial variables?

We are interested in modelling the paths of these variables

How do they evolve over time?
Can we predict what is going to happen in the future?
- forecasting
Can we explain how the variables are related to each other?
- portfolio and risk management implications

In this course, we will start to build and estimate models to

try and answer these questions
8 / 43
Time series data in finance

Most financial data is time series data

Time series data is a sequence of observations ordered over
time
Typically, time series data is collected at predetermined time
intervals – this is its frequency

Financial data is increasingly available at very high frequencies

Large amounts of interesting data now available

Need models and estimation techniques to help understand it

9 / 43
Time series data in finance

Time series data is distinguished from cross-sectional data in

terms of:
the way the data is collected (over time vs. across units)
the possible sample size at any point in time (one vs. as many
as you can!)
the interpretation of the estimation results (dynamic vs.
comparative static)

The use of time series data presents opportunities, but also

challenges

This course will focus heavily on techniques that have been

developed to analyse time series data

10 / 43
A first look at the data: Prices
Asset prices are very important in finance:
The cost of a financial instrument
Any change in price results in a profit or loss for an investor

We would like to be able to track the price of a financial

instrument by building a model

We might be able to use it to predict future price movements

We might also want to analyse how it relates to the prices of

other assets - portfolio risk management

Let’s start by looking at a United States equity price index

(S&P 500) for the period 1950-2014 and considering some of
its properties
11 / 43
A first look at the data: Prices
The figure gives a monthly plot of the S&P 500 equity index
for the period 1950-2014

2,000

1,600

1,200

800

400

0
1950 1960 1970 1980 1990 2000 2010

S&P 50 0 Ind ex Expone ntial Tren d

12 / 43
A first look at the data: Prices
Some things to pay attention to:
Prices (Pt ) are rising over the sample period, from 17.05 in
January 1950 to 1859.45 in February 2014
The time path seems to be trending upwards, with long run
growth. The rate of increase in Pt looks to be increasing
As a first approximation, this general shape could be captured
by an exponential trend:
Pt = Pt−1 exp(r )
where r is viewed as a constant growth rate in Pt
To see why, it is viewed as a growth rate, rearranging gives
Pt − Pt−1
= exp(r ) − 1 ≈ r
Pt−1
since exp(r ) ≈ (1 + r ) for small values of r
13 / 43
A first look at the data: Prices

With a constant growth rate r , the relationship between Pt

and Pt−2 is

Pt = Pt−1 exp(r ) = Pt−2 exp(r ) exp(r ) = Pt−2 exp(2r )

Continuing in this manner, the relationship between Pt and

P0 is
Pt = P0 exp(rt)
This is the exponential function plotted in the figure, where
P0 = 17.05 is the equity price in January 1950 and r=0.0061

14 / 43
A first look at the data: Prices

The exponential function provides a predictive relationship,

based on long-run growth behaviour. Consider an investor in
January 1950 (who knows r=0.0061), who wants to predict
the price of equities in February 2014 (t=770). Using the
exponential model:

P(Feb. 2014) = 17.05 exp(0.0061 × 770) = 1868.983

The actual equity price index in February 2014 was 1859.45 so

that the percentage (forecast) error was

1868.983 − 1859.45
100 × = 0.513%
1859.45

15 / 43
A first look at the data: Prices
We can also analyse the long run time series behaviour of
asset prices, by plotting the logarithm of price over time. The
figure below shows the log of the S&P 500 index
Log Equity Price Index
8

2
1950 1960 1970 1980 1990 2000 2010

log Pt increases at a relatively constant rate

The slope of the line is approximately r 16 / 43
A first look at the data: Prices

To show this, take natural logs of the exponential equation,

Pt = P0 exp(rt)

log Pt = log P0 + rt
pt = p0 + rt

This is a line with intercept p0 and slope r

17 / 43
A first look at the data: Returns

Returns are important in finance, because they provide a

scale-free measure of the payoffs to investing in an asset

Prices changes (∆Pt = Pt − Pt−1 ) are not scale-free, because

they depend on the scale of Pt
For example, how can we compare the payoffs of investing in:
Australian Dollars which now has a price (exchange rate) of
around US$0.73
The S&P 500 index which now has a “price” of around 2050

We can’t compare price changes across assets. Returns

measure proportional (or percentage) changes, so do not
depend on the scale of Pt . Therefore, they can be compared
across assets

18 / 43
A first look at the data: Returns
Discrete returns:
The simple return on an asset between time t − 1 and t is
Pt − Pt−1 Pt
Rt = = −1
Pt−1 Pt−1
The compound return for n periods, Rn,t , is
Pt
Rn,t = −1
Pt−n
Pt Pt−1 Pt−(n−2) Pt−(n−1)
= × × ··· × × −1
Pt−1 Pt−2 Pt−(n−1) Pt−n
= (1 + Rt ) × (1 + Rt−1 ) × · · · × (1 + Rt−(n−1) ) − 1
n−1
Y
= (1 + Rt−j ) − 1
j=0

19 / 43
A first look at the data: Returns

For example, in the case of monthly returns, the annual return

(n = 12) is given by
 
11
Y
R12,t =  (1 + Rt−j ) − 1
j=0

20 / 43
A first look at the data: Returns

Continuous returns:
It is often easier to work with continuous or log returns than
with discrete returns
The log return on an asset is

rt = log Pt − log Pt−1 = log(1 + Rt )

For small values of Rt , the two measures (rt and Rt ) are very
close to each other, since log(1 + Rt ) ≈ Rt
Log returns are also referred to as continuously compounded
returns. This makes compounding returns extremely simple

21 / 43
A first look at the data: Returns

The 2 period compound return is

r2,t = log Pt − log Pt−2

= (log Pt − log Pt−1 ) + (log Pt−1 − log Pt−2 )
= rt + rt−1

The n period compound return is

rn,t = log Pt − log Pt−n

= (log Pt − log Pt−1 ) + (log Pt−1 − log Pt−2 )+
· · · + (log Pt−(n−1) − log Pt−n )
= rt + rt−1 + · · · + rt−(n−1)

22 / 43
A first look at the data: Returns

Useful properties of log returns:

If rt is assumed to be constant over n periods, the n period
return is
rn,t = nrt

Some examples:
If rt is a monthly return, the annual return is

r12,t = rt × 12 = log Pt − log Pt−12

If rt is a daily return, the annual return (based on calendar

days) is
r365,t = rt × 365 = log Pt − log Pt−365

23 / 43
A first look at the data: Returns

The difference between the return on a risky financial asset

and the return on some benchmark (risk-free) asset is known
as an excess return, and is:
Zt = Rt − Rf ,t for discrete returns
zt = rt − rf ,t for log returns
where Rf ,t and rf ,t are the discrete and log returns on the
risk-free asset

The risk-free rate is usually taken to be the return on a

government bond, because the risk of default is usually
considered to be very low

24 / 43
A first look at the data: Returns

The following figure gives the monthly log returns on the S&P
500 index for the U.S. over the period January 1950 to
February 2014

S&P500 Equity Index Returns

.2

.1

.0

-.1

-.2

-.3
1951 1961 1971 1981 1991 2001 2011

25 / 43
A first look at the data: Returns

The returns hover around a mean return value that is close to

zero

A comparison of equity prices and returns shows that:

Prices drift upwards but returns do not
Returns are characterised by relatively high volatility

The different properties of these series suggest that we need

different models for prices and returns. This is something we
will come back to later in the course

26 / 43
A first look at the data: Financial distributions
An important assumption in many theoretical and empirical
models in finance is that returns are normally distributed
Important in portfolio allocation models, in Value-at-Risk
(VaR) calculations, in pricing options, and other applications
An example of an empirical return distribution is given by the
histogram of log returns to the S&P 500 index below
RETURNS

12

10

8
Density

0
-.3 -.2 -.1 .0 .1 .2 .3 27 / 43
A first look at the data: Financial distributions

The red line superimposes a normal distribution over the

histogram, for comparison

The distribution appears to have a large left tail, with a

disproportional weight on extreme negative returns (negative
skewness)

The distribution appears to have a sharp peak in the centre of

the distribution and some evidence of heavy tails (excess
kurtosis)

These features are common in asset returns in several markets

and suggest risks for investors beyond volatility

28 / 43
Importance of return distributions

Investors, when choosing which assets to select in a portfolio,

will be concerned about the probability distribution of the
returns, that is the probability attached to all possible future
realisations of returns

Risk averse investors generally prefer to invest in assets with:

Higher expected returns
Little spread in possible returns (low variance)
Greater chance of extreme positive returns than extreme
negative returns (positive rather than negative skewness)
Small chance of extreme returns (lower kurtosis)

With time series data, we don’t observe the true population

probability distribution, but can get information from the
sample distribution of returns
29 / 43
Summary statistics: Sample mean
When investing in an asset, the first requirement is to know
the expected return E (rt ). An estimated measure of this is
given by the sample mean:
T
1 X
r¯ = rt
T
t=1
Using the previous data on S&P 500 equity index returns, the
sample mean is
r¯ = 0.0061
or returns on average are 0.61% per month. Expressing the
monthly sample mean in annual terms gives
0.0061 × 12 = 0.073212
which shows that average returns over the period 1950 to
2014 are 7.32% per annum
30 / 43
Summary statistics: Sample mean

The sample mean is plotted in the following figure, with rt

.2

.1

.0

-.1

-.2

-.3
1950 1960 1970 1980 1990 2000 2010

Equ tiy Index Retur ns Mea n Re tu rn

As rt hovers around r¯ = 0.0061, the sample mean represents a

summary measure of the central tendency of the data
31 / 43
Summary statistics: Sample mean
An example where the sample mean is an inappropriate
summary measure is where the data are trending
2,000

1,600

1,200

800

400

0
1950 1960 1970 1980 1990 2000 2010

S&P 50 0 Ind ex Mea n Price

In the S&P 500 example, Pt does not hover around

P̄ = 440.966, so the sample mean does not represent a
summary measure
Illustrates different time series properties of Pt and rt
32 / 43
Summary statistics: Sample variance
When investing in an asset, the investor wants to know the
risk of the asset
The risk is often viewed in terms of the variance:
σr2 = E (rt − E (rt ))2
This measures how much returns are spread around the
expected return. An estimated measure of this is given by the
sample variance:
T
1 X
σ̂r2 = (rt − r¯)2
T
t=1

In the case of the returns data, the sample variance is

σ̂r2 = 0.00176
33 / 43
Summary statistics: Sample standard deviation

In finance, the sample standard deviation is a a more common

measure of risk (also known as the volatility)
v
u
u1 X T
σ̂r = t (rt − r¯)2
T
t=1

In the case of the returns data, the sample standard deviation

is
σ̂r = 0.042

34 / 43
Summary statistics: Sample skewness

When investing in an asset, an investor wants to know

whether the extreme returns are more likely to be above or
below the expected return
If the extreme returns are positive (negative) returns, the
distribution of rt is positively (negatively) skewed
A measure of skewness in the sample is
T 
1 X rt − r¯ 3

ˆ
SK =
T σ̂r
t=1

This is an estimate of the population skewness:

rt − E (rt ) 3
 
SK = E
σr
35 / 43
Summary statistics: Sample skewness
RETURNS

12

10

Density 6

0
-.3 -.2 -.1 .0 .1 .2 .3

S &P 500 Returns Norma l

The histogram of S&P 500 log returns shows a larger

concentration of returns below the sample mean of r¯ = 0.0061
(left tail) than there is for returns above the sample mean
(right tail). The sample skewness is
ˆ = −0.665
SK
36 / 43
Summary statistics: Sample kurtosis
A fourth requirement for an investor to know is whether there
are extreme returns (either positive or negative), as investors
prefer returns closer to expected returns
If there are extreme returns relative to a benchmark
distribution (usually the normal distribution), the distribution
of rt exhibits excess kurtosis
A measure of kurtosis in the sample is
T 
1 X rt − r¯ 4

ˆ
KT =
T σ̂r
t=1

This is an estimate of the population kurtosis:

rt − E (rt ) 4
 
KT = E
σr
37 / 43
Summary statistics: Sample kurtosis

Comparing this value to KT = 3 (the kurtosis value of a

normal distribution) gives the excess kurtosis:
T  4
ˆ )= 1 X r t − ¯
r
EXCESS(KT −3
T σ̂r
t=1

In the case of the S&P 500 index log returns:

ˆ = 5.418
KT

Since this value is greater than 3, there are more extreme

returns in the data than would have been predicted by the
normal distribution

38 / 43
Summary statistics: Sample covariance
It is often important to understand the interrelationships
between two or more financial time series
For example, in constructing diversified portfolios, the aim is
to include assets that are not perfectly correlated
A measure of the co-movements between the returns on two
assets, rit and rjt , is the sample covariance:
T
1 X
σ̂ij = (rit − r¯i )(rjt − r¯j )
T
t=1

where r¯i and r¯j are the sample means of the log returns on
assets i and j
This is an estimate of the population covariance:
σij = E {[rit − E (rit )][rjt − E (rjt )]}
39 / 43
Summary statistics: Sample correlation

Another measure of association is the sample correlation

coefficient:
σ̂ij σ̂ij
ρ̂ij = q =
σ̂i2 σ̂j2 σ̂i σ̂j

where
T T
1 X 1 X
σ̂i2 = (rit − r¯i )2 , σ̂j2 = (rjt − r¯j )2
T T
t=1 t=1

represent the respective sample variances of the log returns to

assets i and j
The correlation coefficient has the same sign as the
covariance, as well as the property that

−1 ≤ ρ̂ij ≤ 1
40 / 43
The efficient markets hypothesis: Return predictability

The efficient markets hypothesis theorises that all available

information concerning the value of a risky asset is factored
into the current price of the asset

A corollary of the efficient markets hypothesis is that the

current price provides no information on the direction of the
future price

One way to examine this proposition is to test the asset

returns for autocorrelation

41 / 43
The efficient markets hypothesis: Return predictability
The autocorrelation statistic measures the strength of
co-movements between current returns (rt ) and returns on the
same asset k periods earlier (rt−k )
1 PT
ˆ (k) = T t=k+1 (rt − r¯)(rt−k − r¯)
acf 1 PT 2
T t=1 (rt − r¯)
The numerator represents the autocovariance of returns k
periods apart and the denominator represents the variance of
returns
If returns exhibit no autocorrelation, then future returns are
unpredictable
If returns exhibit positive (negative) autocorrelation, then
successive values of returns tend to have the same (opposite)
sign and this pattern can be exploited in predicting the future
behaviour of returns
42 / 43
The efficient markets hypothesis: Return predictability

The following table gives the first 5 sample autocorrelations of

the S&P 500 index log returns
Lag Autocorrelation
1 0.051
2 -0.040
3 0.040
4 0.047
5 0.077

The autocorrelations are reasonable close to zero. However,

some of them are large enough to suggest that there may be a
little predictability. We would need a test to tell conclusively

43 / 43