Transcript power point slides for lecture #8 (ppt file)

Lecture 8:
Review: Forecasting
methods
Understanding Markets and
Industry Changes
2
The full model
 The model with seasonality, quadratic trend, and ARMA
components can be written:
y t  b1D1,t  ... bsDs,t a1 t a 2 t 2  ut ,
ut  1ut1   2 ut2  ...  p ut p  ...
et  1et1  ...  q etq
 Dummy variables, Djt control for seasonality.

 The variable t controls for trend if your data appears to grow or fall at a
linear rate.
 If the “trend” doesn’t look like a straight line, you may consider taking the
natural log of your original series, or including the quadratic trend part, t 2.
3
The full model continued…
 The model with seasonality, quadratic trend, and ARMA
components can be written:
y t  b1D1,t  ... bsDs,t a1 t a 2 t 2  ut ,
ut  1ut1   2 ut2  ...  p ut p  ...
et  1et1  ...  q etq
 If your data is related to past observations of itself, include autoregressive
components (expect processes having autoregressive components to have
slowly decaying autocorrelations).

 If your data is related to past errors made in fitting the data, include moving
average components (expect processes having moving average components to
have slowly decaying partial autocorrelations).
4
Model selection
 An important statistic that can used in choosing a model
is the Schwarz Bayesian Information Criteria. It rewards
models that reduce the sum of squared errors, while
penalizing models with too many regressors.
 SIC=log(SSE/T)+(k/T)log(T), where k is the number of
regressors.
 The first part is our reward for reducing the sum of
squared errors. The second part is our penalty for
adding regressors. We prefer smaller numbers to larger
number (-17 is smaller than -10).
5
Important commands in EViews
 ar(1): Includes a single autoregressive lag
 ar(2): Includes a second autoregressive lag
 Note, if you include only ar(2), EViews will not include a
first order autoregressive lag
 ma(1): Includes a first order moving average term. This
is not the same as forecasting using an average of
recent values
Selecting an appropriate time
series model, concluded
 Determine if trend/seasonality is important
 If it is, include it in your model
 Estimate the model with necessary trend/seasonal
components. Look at the correlogram of the residuals:
 From the equation dialogue box:
 View => Residual Tests => Correlogram Q-statistics
 If ACs decay slowly with abrupt cutoff in PAC, this is indicative
of AR components. If the PAC doesn’t cutoff, you may need to
include MA components as well.
 Re-estimate the full model with trend/seasonality
included with necessary ARMA components. You will
likely have several models to choose from.
6
Selecting an appropriate model,
cont.
 After you estimate each model, record SIC/AIC values
 Use the SIC/AIC values to select an appropriate model.
 Finally, investigate the final set of residuals. There
should be no correlation in your residuals.
 Evidenced by individual correlation coefficients within 95%
confidence intervals about zero.
 Ljung-Box Q-statistics should be small with probability
values typically in excess of 0.05.
7
8
Exponential smoothing
 Eviews provides five options when you ask it, no tell it,
to provide exponential smoothing:
 Single: (no seasonality/no trend)
 Double: (trend – value of a=b).
 Holt-Winters – No seasonal (Trend, a and b are not equal,
but are estimated in the data).
 Holt-Winters – Additive (Trend and Seasonality. The
seasonal component is estimated with an additive filter).
 Holt-Winters - Multiplicative
Moving average methods
 Two-sided (for some arbitrary m):
m
1
 y t  i , with m  2 :
2m  1 i  m
1
( y t  2  y t 1  y t  y t 1  y t  2 )
5
 One-sided:

m
1
y t  i , with m  2 :

m  1 i 0
1
( y t  y t 1  y t  2 )
3

To calculate a moving average:
 Limit your sample to the period the average is being
constructed under.
 Create a variable that is simply an average of the
variable you are looking for, for the restricted period.
 For both the exponential smoothing model and moving
average model, EViews will not give you the mean
squared error. You will need to calculate it your self.
You can do so in EViews or in Excel.
11
Breaks?
 Uh oh? My data appears to have a break.
 The developed time series methods assume the
black box generating the data is constant.
 Not necessarily true:
 Learning curves may cause cost curves to
decrease
 Acquisition of companies or new technologies
may alter sales/costs
12
Dealing with breaks?
 Solutions:
 Limit the sample to the post break period
 Sometimes taking logs and/or differencing can help
mitigate the effects of breaks/outliers.
 Include variables that help identify the breaks
 Model the breaks directly:
 The most obvious way is to include a break in mean and/or a
break in trend.
 We should make sure that the included break is modeled in a
sensible way
 A negative linear trend, for example, will imply the data
may eventually turn negative.
13
Break in mean
100
90
80
70
y
60
50
40
30
20
10
0
0
20
40
60
80
100
time
120
140
160
180
200
14
Break in trend
90
80
70
60
50
40
30
20
10
0
0
20
40
60
80
100
120
140
160
180
200
Statistics useful in comparing the
out of sample forecasting accuracy
 Mean squared error: For an h-step extrapolation
forecast:
(YˆT 1|T  YT 1 ) 2  (YˆT 2|T  YT 2 ) 2  ... (YˆT h|T  YT h ) 2
h
 Root mean squared error is the square root of this
number.
 absolute error
 Mean
| YˆT 1|T  YT 1 |  | YˆT 2|T  YT 2 | ... | YˆT h|T  YT h |
h
15
16
In Eviews:
 If you have a forecasted series, say xf, and an original
series x, you can calculate the mean squared error as:
 genr mse=@sum((x-xf)^2)/h
 To calculate the moving average forecasts:
 Suppose you use the most recent four periods
 Limit your data set to include only the last four observations
 A variable called “maf_4” is calculated by:
 genr maf_4=@mean(x)
Lecture 8: Part 2
SHIFTS IN SUPPLY AND DEMAND
UNDERSTANDING INDUSTRY
CHANGES
– Summary of main points
• A market has a product, geographic, and time
dimension. Define the market before using supply–
demand analysis.
• Market demand describes buyer behavior; market
supply describes seller behavior in a competitive
market.
• If price changes, quantity demanded increases or
decreases (represented by a movement along the
demand curve).
• If a factor other than price (like income) changes, we
say that demand curve increases or decreases (a shift
of demand curve).
Lecture 8 – Summary (cont.)
• Supply curves describe the behavior of sellers and tell you how
much will be sold at a given price.
• Market equilibrium is the price at which quantity supplied
equals quantity demanded. If price is above the equilibrium
price, there are too many sellers, forcing price down, and vice
versa.
• Currency depreciation in a country increases demand for exports
(supply to another country) and decreases demand for imports
(demand for another country’s products).
• Prices are a primary way that market participants
communicate with one another.
• Making a market is costly, and competition between market
makers forces the bid–ask spread down to the costs of making a
market. If the costs of making a market are large, then the
equilibrium price may be better viewed as a spread rather than a
single price.
Anecdote: Y2K and generator sales
• From 1990-98, sales of portable generators grew 2%
yearly.
• In 1999, public anticipation of Y2K power outages
increased demand for generators.
• Walters, Rosenberg and Matthews invested to increase
capacity in anticipation of this demand growth – they
vertically integrated their company to increase capacity
and reduce variable costs.
• Demand grew as expected - Industry shipments
increased by 87%. Prices also increased by an average
of 21%.
Which industry or market?
• Every industry or market has a time, product, and
geographic dimension.
• For example: The yearly market for portable generators
in the U.S.
• Time: annual
• Product: portable generators
• Geography: US
• When analyzing a problem, or investment opportunity, it
helps to first define the time, product and geographic
dimensions of the market in question.
Shifts in the demand curve
• Movement along the demand curve indicates the
“quantity demanded” increased.
• Shifts in demand curve can occur for multiple reasons
• Uncontrollable factor – affects demand and is out of a
company’s control.
• Income, weather, interest rates, and prices of substitute and
complementary products owned by other companies.
• Controllable factor – affects demand but can be controlled
by a company
• Advertising, warranties, product quality, distribution speed,
service quality, and prices of substitute or complementary
products also owned by the company
Anecdote: Microsoft
• In the late 1970s, Microsoft developed DOS, an
operating system to control IBM computers.
• The price for DOS depended on the price and availability of
computers that could run it and the applications that ran
under it as well as the price of DOS itself.
• To increase demand for DOS Microsoft:
• Licensed its operating system to other computer manufacturers
• Developed its own versions of complimentary products
• To affect the quantity demanded, they also kept the
price of DOS low.
Demand increase
• At a given price, more quantity demanded
Supply curves
• Definition: Supply curves are functions that relate
the price of a product to the quantity supplied by
sellers.
• Discussion: Why do supply curves slope upwards?
Market equilibrium
• Definition: Market equilibrium is the price at
which quantity supplied equals quantity demanded.
• At the equilibrium price, there is no pressure for
the price to change given the equality of quantity
demanded and supplied.
Market equilibrium (cont.)
• Proposition: In a competitive
equilibrium there are no
unconsummated wealth-creating
transactions.
Price
$12
$11
$10
$9
$8
$7
$6
$5
$4
Demand
1
2
3
4
5
6
7
8
9
Supply
9
8
7
6
5
4
3
2
1
Using supply and demand
• Supply and demand curves can be used to describe changes
that occur at the industry level
Using supply and demand (cont.)
• Discussion: “over the past decade, the price of
computers has fallen, while quantity has risen.” How?
Why?
Prices convey information
• Prices are a primary way that market participants
communicate with one another
• Buyers signal their willingness to pay, and sellers
signal their willingness to sell with prices
• Price information especially important in financial
markets