You are on page 1of 55

CHAPTER 3:

SIMPLE FORECASTING MODELS


Trang, Ha Thi Thu (Phd)
SEM, HUST
Chapter Contents

3.1. Introduction
3.2. Naïve Model
3.3. Forecasting with Averaging Models:
3.3.1. Simple Average Model
3.3.2. Moving Average Model
3.3.3. Double Moving Average Model

3.4. Exponential Smoothing:


3.4.1. Simple Exponential Smoothing
3.4.2. Holt’s Method
3.4.3. Winters’ Seasonal Exponential Smoothing
3.1. Introduction

• Elaborate statistical models are not always required


to develop accurate forecasts.
• The principle of parsimony suggests that the
simpler the model the better.
• The main advantage of simple models is that they
serve as a benchmark with which to gauge
applicability, reliability, and necessity of the more
sophisticated models.
3.1. Introduction

• These time series models are good tools in forecasting short-term


events.
• The cardinal premise underlying all time series models is that the
historical pattern of the dependent variable can be used as the basis
for developing forecasts.
• In these models, historical data for the forecast variable are analyzed in
an attempt to discern any underlying pattern(s).
3.1. Introduction

• Time series or autoregressive forecasting models will be most useful


when economic conditions can be expected to remain relatively stable.
• Reliance of time series models on analysis and extrapolation of
historical patterns carries several important implications with respect
to technique selection:
1. Time series are best when applied to short-term forecasts.
2. Time series models prove most satisfactory when historical data
contain either no systematic data pattern or when the changes are
occurring very slowly or consistently.
3. Data requirements and ease of implementation are a function of
the specific time series technique selected.
3.1. Introduction
3.2. Naïve Model

• Uses recent past as the best indicator of the future.


3.2. Naïve Model

• The error associated with this model is computed as:

ˆ
et  Yt  Yt
3.2. Naïve Model

Example for the Naïve Model


Week Sales (in $1,000) Forecast
1 9 -
2 8 9
3 9 8
5 12 9
6 9 12
7 12 9
8 11 12
9 ? 11
3.2. Naïve Model

Example for the Naïve Model


Week Sales (in Forecast Error Absolute Squared
$1,000) Error Error
1 9
2 8 9 −1 1 1
3 9 8 1 1 1
5 12 9 3 3 9
6 9 12 −3 3 9
7 12 9 3 3 9
8 11 12 −1 1 1
Sum 2 12 30
Mean 0.33 2 5
3.2. Naïve Model

Example for the Naïve Model


• What you should keep in mind is that, although MAD is often
used as the measurement of error in evaluating a forecast, an
alternative criterion is the MSE.

• Note that the difference between MAD and MSE is that the
latter penalizes a forecast much more for extreme deviations
than it does for small ones.
3.2. Naïve Model

Example for the Naïve Model

 Whenever a manager evaluates alternative forecasting


techniques in terms of their accuracy, it is necessary to go
beyond the computation of error.
 Managers are generally concerned with two forms of
accuracy:
 Accuracy of the technique in predicting the underlying
patterns or relationship of past data.
 Accuracy of the changes in the pattern: how fast
forecasting procedure can respond to that basic
change.
3.3. Forecasting with Averaging Models:

• 3.3.1. Simple Average Model


• 3.3.2. Moving Average Model
• 3.3.3. Double Moving Average Model
3.3. Forecasting with Averaging Models:

• 3.3.1. Simple Average Model


• 3.3.2. Moving Average Model
• 3.3.3. Double Moving Average Model
3.3. Forecasting with Averaging Models:

 Frequently, management faces the situation in which


forecasts need to be updated daily, weekly, or monthly for
inventories containing hundreds or thousands of items.
 it is not possible to develop sophisticated forecasting
techniques for each item.
 some quick, inexpensive, very simple short-term
forecasting tools are needed
 The basic premise of these models is that a weighted-
average of past observations can be used to smooth the
fluctuations in the data in the short term.
 The assumption underlying these techniques is that the
fluctuations in past values represent random departures from
some underlying structure.
3.3. Forecasting with Averaging Models:

3.3.1. Simple Average Model


3.3.2. Moving Average Model
3.3.3. Double Moving Average Model
3.3.1. Simple Average Model

• Similar to the naïve model, this model uses part of the


historical data to make a forecast.

n
 Yt
ˆ
Y t 1  t 1
n
3.3.1. Simple Average Model

• When forecasting a large number of series simultaneously


(e.g., for inventory management),
 Problem with data storage.
Updated simple average, new period:

Only the most recent forecast and the most recent


observation need be stored as time moves forward.
3.3.1. Simple Average Model

• Recent observations play an important role in the forecast.


• As new observations become available, a new average is
computed.
• The choice of using a smaller or larger number of
observations has implications for the forecast.
• The method of simple averages is an appropriate when:
• the forces generating the series to be forecast have
stabilized and
• the environment in which the series exists is generally
unchanging.
3.3.1. Simple Average Model

Examples:
• Forecast the quantity of sales resulting from a consistent
level of salesperson effort;
• Forecast the quantity of sales of a product in the mature
stage of its life cycle;
• Forecast the number of appointments per week requested
of a dentist, doctor, or lawyer whose patient or client base is
fairly stable.
3.3.1. Simple Average Model
Examples:
3.3.1. Simple Average Model
Examples:
3.3.1. Simple Average Model
Examples:
Forecast Y11 and Y12
3.3.2. Moving Average Model

• A moving average of order k, MA(k):


3.3.2. Moving Average Model

Note that
• the moving average technique deals only with the latest k
periods of known data;
• the number of data points in each average does not change
as time advances.
• The moving average model does not handle trend or
seasonality very well, although it does better than the
simple average method.
3.3.2. Moving Average Model
Annual electricity sales to residential customers in South Australia.
1989–2008.
3.3.2. Moving Average Model
Annual electricity sales to residential customers in South Australia.
1989–2008.
3.3.2. Moving Average Model
Annual electricity sales to residential customers in South Australia.
1989–2008.
3.3.2. Moving Average Model

Some remarks:
 Moving averages are frequently used with quarterly or
monthly data to help smooth the components within a time
series.
 For quarterly data, a four-quarter moving average, MA(4),
yields an average of the four quarters.
 For monthly data, a 12-month moving average, MA(12),
eliminates or averages out the seasonal effects.
 The larger the order of the moving average, the greater the
smoothing effect.
3.3.2. Moving Average Model

Double Moving Average Model

• Used when we have a linear trend in the data.

• Two different moving averages are computed in this model.

• The idea is to remove the trend.


3.3.2. Moving Average Model

Double Moving Average Model

• Used when we have a linear trend in the data.

• Two different moving averages are computed in this model.

• The idea is to remove the trend.


3.3.2. Moving Average Model

Double Moving Average Model


• First, compute the moving average of order k.

• Then compute the second moving average.


3.3.2. Moving Average Model

Double Moving Average Model


Make the forecast p periods into the future

Where,
• k = the number of periods in the moving average
• p = the number of periods ahead to be forecast
3.3.2. Moving Average Model

Example
3.4. Exponential Smoothing:

3.4.1. Simple Exponential Smoothing

3.4.2. Holt’s Method

3.4.3. Winters’ Seasonal Exponential


Smoothing
3.4. Exponential Smoothing Models

• The model relies on the assumption that the data are


stationary.
• Most recent observations play a more important role
than the distant past.
3.4. Exponential Smoothing Models

 The model depends on three pieces of data:


Most recent actual
Most recent forecast
Smoothing constant.
 The value of alpha assigned as a smoothing constant is
critical to the forecast.
 The best alpha should be chosen on the basis of
minimal sum of error squared.
3.4. Exponential Smoothing Models

Where,
t+1 = the new smoothed value or the forecast value for the next
period
= the smoothing constant (0 < < 1)
Yt = the new observation or the actual value of the series in period t
t = the old smoothed value or the forecast for period t
3.4. Exponential Smoothing Models

When
• a ~ 1 : the new forecast will be essentially the current observation.
• a ~ 0: the new forecast will be very similar to the old forecast, and the
current observation will have very little impact.
3.4. Exponential Smoothing:

Several approaches are followed in selecting the


smoothing constant.
• If a great amount of smoothing is desired, a small
alpha should be chosen.
• The choice of alpha is affected by the characteristics
of the time series.
If sharp ups and downs are noticed in the data,
the best smoothing constant is 0.1. That is alpha
chosen should equal 0.1.
If the data show that the past is very different
from the present, then alpha of 0.9 is appropriate.
3.4. Exponential Smoothing:

Several approaches are followed in selecting the


smoothing constant.
• If a great amount of smoothing is desired, a small
alpha should be chosen.
• The choice of alpha is affected by the characteristics
of the time series.
If sharp ups and downs are noticed in the data,
the best smoothing constant is 0.1. That is alpha
chosen should equal 0.1.
If the data show that the past is very different
from the present, then alpha of 0.9 is appropriate.
3.4. Exponential Smoothing:

Exponential smoothing is used for routine sales


forecasting of
inventory,
production,
distribution, and
retail planning.
3.4.2 Holt’s Exponential Smoothing Model

Holt’s Exponential Smoothing Model


3.4. Exponential Smoothing:

Holt’s Exponential Smoothing Model


To handle linear trend.
Holt’ method smooth the trend and the slope in the
time series by using different constants for each.
How do we find the best combination of smoothing
constant?
Low values of alpha and beta should be used when
there are frequent random fluctuations in the data.
High values of alpha and beta should be used when
there is a pattern such as trend in the data.
3.4. Exponential Smoothing:

Holt’s Exponential Smoothing Model


Methods to determine the initial values for L and T:
1. Set the first estimate of the smoothed level equal
to the first observation (L1=Y1). The trend is then
estimated to be zero (T1=0).
2. Use the average of the first five or six
observations as the initial smoothed value L. The
trend is then estimated using the slope of a line
fit to these five or six observations.
3.4. Exponential Smoothing:

Winters’ Seasonal Exponential Smoothing


• Allows for both trend and seasonal patterns to be
taken into account.
• This is an extension of the Holt’s method of
smoothing.
• In computing the forecast, we add an equation for
seasonality as an index.
3.4. Exponential Smoothing:

Winters’ Seasonal Exponential Smoothing


3.4. Exponential Smoothing:

Winters’ Seasonal Exponential Smoothing


3.4. Exponential Smoothing:

Winters’ Seasonal Exponential Smoothing


To determine the weights α, β, and γ :
1. Selected subjectively
2. Generated by minimizing a measure of forecast
error, such as the MSE  use an optimization
algorithm to find the optimal smoothing
constants.
3.4. Exponential Smoothing:

Winters’ Seasonal Exponential Smoothing


To determine the initial values for the smoothed
series, the trend, and the seasonal indices:
1. One approach is to set: L1 = Y1, T1 = 0, S1 = 1
2.Other approaches, for example, develops a
regression equation using Yt = a + bt.
Lt = a; T1 = b,
Initial values for the seasonal components are
obtained from a dummy variable regression using
detrended data.
3.4. Exponential Smoothing:

In summary,
 Simple Exponential Smoothing: Technique for data
with no trend or seasonality.
 Holt’s Method: Technique for data with trend but
no seasonality.
 Holt-Winters Seasonal Method: Technique for
data with trend and seasonality.
A taxonomy of exponential smoothing methods
A taxonomy of exponential smoothing methods
A taxonomy of exponential smoothing methods

To apply the Holt-Winters method in R, use ets()


which stands for error, trend, and seasonality.
In total you have 36 model options to choose
"AAA" = a model with additive error, additive trend,
and additive seasonality.
APPLICATION TO MANAGEMENT

• With a judicious choice of order k, the moving average


method can do a good job of adjusting to shifts in levels.
• It is economical to update, and it does not require much data
storage.
• The moving average method is most frequently used when repeated
forecasts are necessary.
• Exponential smoothing is a popular technique whose strength
lies in good short-term accuracy combined with quick, low-
cost updating.
• The technique is widely used when regular monthly or weekly
forecasts are needed for a large number, perhaps thousands, of
items.
• Inventory control is one example where exponential smoothing
methods are routinely used.

You might also like