MBA 532Business Statistics
by
Rushan AbeygunawardanaDepartment of Statistics, University of Colombo
MBA 532:Business Statistics MBA-2011
Time Series Analysis
IntroductionA time series is a collection of observations made sequentially in time.
In economics and business
Share price on successive days
Export total in successive months, Yearly sales figures
Weekly bank interest rates
In meteorology
Daily rainfall, daily temperature
Hourly wind speed
In demography
Population of a country in successive years
In marketing
Sales in successive months
The purposes of time series analysis are;
To understand or model the behavior of the observed series.
To predict or forecast future of a series based on the past values of the series.
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Continuous and Discrete Time Series
A time series is said to be continuous when observations are made continuously in time. The term “continuous’ is used for series of this type even when the measured variable can only take a discrete set of values.
A time series is said to discrete when observations are taken only at specific times, usually equally spaced. The term “discrete” is used for series of this type even when the measured variable is a continuous variable.
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Continuous and Discrete Time Series
In the discrete time series the observations are usually taken at equally intervals.
Discrete time series can arise in several ways.
Sampled: Read off (or digitize) a continuous time series model with the values at equal interval of time.
Aggregate (accumulate) : E.g.: monthly exports, weekly rainfall
The special feature of time series analysis is the fact that successive observations are usually not independent and that the analysis must taken into account the time order of the observations.
When successive observations are dependent the future values may be forecast form the past data.
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Deterministic Time Series and Stochastic Time Series
Deterministic Time Series
If a time series can be forecast exactly, it is said to be
deterministic time series.
Stochastic Time Series
For stochastic time series the exact prediction is
impossible and must be replaced by the idea that
future values have a probability distribution which is
conditioned by knowledge of the past values.
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Objectives of Time Series Analysis
There are several possible objectives in time series
analysis and these objectives may be classified as;
Description
Explanation
Forecasting
Control
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Objectives of Time Series AnalysisDescription
In the time series analysis the first step is plot the data and obtained simple descriptive measures of the main properties of the series. (i.e. tread, seasonal variations, outliers, turning points etc.)
In the plot of the raw sales data;
There are upward trends
There are downward trends
There are turning points
For some series, the variation is dominated by such “obvious” features (trend and seasonal variation) and a fairly simple model may be perfectly adequate.
For some other series, more advance techniques will be required. And more complex models (such as various types of stochastic process) are required to describe the series.
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Objectives of Time Series AnalysisExplanation
When observations are taken on two or more variables, it may be possible to use the variation in one series to explain the variation in the other series.
To see how see level is affected by temperature and pressure
To see how sales are affected by price and economic conditions
To see how sales of soft drink varies with the daily temperature
A stochastic model is fitted and then forecast the future values and then input process is adjusted so as to keep the process on target.
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Objectives of Time Series Analysis
Forecasting (Predicting)
Given an observed time series, one may want to
predict the future values of the series.
Forecasting is vary important task in the analysis of
economic and industrial time series.
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Objectives of Time Series Analysis
Control
When a time series is generated, which measure the
“quality” of a manufacturing process, the aim of
analysis may be to control the process. Control
processes are of several different kinds.
In quality control, observations are plotted on a
control chart and controller takes actions using
pattern of the graph.
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Types of variationsTraditional models of time series analysis are mainly concerned with decomposing the variation in a series into;
Trend
Seasonal variation
Cyclic variation
Irregular variation
Trend (T)
This is defined as “long term change in mean”. Trend measured the average change in the variable per unit time. It shows the graduate and general pattern of development, which is often described by a straight line or some type of smooth curve.
When speaking of a trend, we must take into account the number of observations available and make subjective assessment of what is long term trend.
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Types of variations
Seasonal variation (S)
Many time series exhibits variation which is annual in period. This variation arises due to the seasonal factors. This yearly variation is easy to understand and we shall see that is can be measured explicitly and /or removed from the data to give de-seasonalized data.
Sales of school stationeries in December and January
Sales of charismas trees
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Types of variationsCyclic variation
Apart from the seasonal effects, sometimes series shows variation at a fixed period of time due to some another physical causes.
Daily variation in temperature
Cyclic variation is characterized by recurring up and down movements which are different form seasonal fluctuations in that they extend over longer / shorter period of time, usually two or more years or less than one year.
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Types of variationsIrregular variation
After trend and cyclic variation has been removed from a set of data, we are left with a series of residuals, which may or may not be “random”. Then we may see some anther type of variation which do not show a regular pattern and it is called the irregular variation.
We shall examine various techniques for analyzing series of this type to see if some of the appropriate irregular variation may be explained in terms of probability models such as Moving Average (MA) or Autoregressive (AR) models which will be discussed later.
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Stationary Time Series
This is a time series with;
No systematic change in mean (No trend)
No systematic change in variation (No seasonal variation)
No strictly periodic variation (No cyclic variation).
Most of the probability theory (stochastic process) of time
series is concerned with stationary time series and for this
reason time series analysis often requires one to turn a non-
stationary into a stationary one so as to use these theories.
That is, it may be interested of remove the trend and seasonal
variation from a set of data and then try to model the variation
in residuals by means of stationary stochastic process.
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Time PlotThis is the first step in analyzing a time series data. That is plot the observations against time. The time plot shows whether and how the values in a dataset change over time. You can make a time plot of any numeric data.
Time Plot graphs are similar to X-Y graphs, and are used to display time-value data pairs. A Time Plot data item consists of two data values—the time and the value—which translate into the x and y coordinate, respectively. Each data item is displayed as a symbol, but you can add a line, bubbles, or fill areas to better delineate the data. Because of the nature of the coordinate system, Time Plot graphs do not have categories.
Time graphs are good for graphing the values at irregular intervals, such as sampling data at random times.
Plotting a time series is not a easy task. The choice of scales, the size of the intercept, and the way that the points are plotted (continuous lines or separate points) may substantially affected the way the plot “look” and so the analyst must examine very carefully and make the judgments.
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Common Approaches to Forecasting
Used when historical data
are unavailable
Considered highly
subjective and judgmental
Common Approaches to Forecasting
Causal
Quantitative forecasting methods
Qualitative forecasting methods
Time Series
Use past data to predict
future values
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Common Approaches to Forecasting…
Quantitative forecasting methods can be used when:
Past information about the variable being forecast is available,
The information can be quantified, and
It can be assumed that the pattern of the past will continue into the future
In such cases, a forecast can be developed using a time series method or a causal method:
Time series methods: The historical data are restricted to past values of the variable. The objective is to discover a pattern in the historical data and then extrapolate the pattern into the future. Ex.: trend analysis, classical decomposition, moving averages, exponential smoothing, ARIMA.
Causal forecasting methods: Based on the assumption that the variable we are forecasting has a cause-effect relationship with one or more other variables (e.g.: the sales volume can be influenced by advertising expenditures). Ex.: regression analysis.
Qualitative methods generally involve the use of expert judgment to develop forecasts.
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Traditional Models in Time Series Analysis
A procedure for dealing with a time series is to fit a suitable
model to the data. There are three commonly use time series.
Additive models
Multiplicative models
A combination of 1 and 2
iiiii ICSTY
iiiii ICSTY
iiiii ICSTY
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Smoothing the Time SeriesCalculate moving averages to get an overall impression of the pattern of movement over timeMoving Average:
averages of consecutive time series values for a chosen period of length L
A series of arithmetic means over time
Result dependent upon choice of L (length of period for computing
means)
Example: Five-year moving average
First average:
Second average:
5
YYYYYMA(5) 54321
5
YYYYYMA(5) 65432
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Example: Moving Average Method
Year Sales
1
2
3
4
5
6
7
8
9
10
11
23
40
25
27
32
48
33
37
37
50
40
Annual Sales
0
10
20
30
40
50
60
1 2 3 4 5 6 7 8 9 10 11
Year
Sa
les
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Calculating Moving Averages
Year Sales
1 23
2 40
3 25
4 27
5 32
6 48
7 33
8 37
9 37
10 50
11 40
Average
Year
5-Year
Moving
Average
3 29.4
4 34.4
5 33.0
6 35.4
7 37.4
8 41.0
9 39.4
5
543213
5
322725402329.4
Annual vs. 5-Year Moving Average
0
10
20
30
40
50
60
1 2 3 4 5 6 7 8 9 10 11
YearS
ales
Annual 5-Year Moving Average
Each moving average is for a consecutive block of 5 years
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Exponential SmoothingA weighted moving average
Weights decline exponentially
Most recent observation weighted most
Used for smoothing and short term forecasting (often one period into the future)
The weight (smoothing coefficient) is W
Subjectively chosen
Range from 0 to 1
Smaller W gives more smoothing, larger W gives less smoothing
The weight is:
Close to 0 for smoothing out unwanted cyclical and irregular components
Close to 1 for forecasting
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Exponential Smoothing Model
Exponential smoothing model
11 YE
1iii E)W1(WYE
where:Ei = exponentially smoothed value for period i
Ei-1 = exponentially smoothed value already computed for period i - 1
Yi = observed value in period i W = weight (smoothing coefficient), 0 < W < 1
For i = 2, 3, 4, …
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Exponential Smoothing Example
Suppose we use weight W = 0.2
Time
Period
(i)
Sales
(Yi)
Forecast
from prior
period (Ei-1)
Exponentially Smoothed
Value for this period (Ei)
1
2
3
4
5
6
7
8
9
10
etc.
23
40
25
27
32
48
33
37
37
50
etc.
--
23
26.4
26.12
26.296
27.437
31.549
31.840
32.872
33.697
etc.
23
(.2)(40)+(.8)(23)=26.4
(.2)(25)+(.8)(26.4)=26.12
(.2)(27)+(.8)(26.12)=26.296
(.2)(32)+(.8)(26.296)=27.437
(.2)(48)+(.8)(27.437)=31.549
(.2)(48)+(.8)(31.549)=31.840
(.2)(33)+(.8)(31.840)=32.872
(.2)(37)+(.8)(32.872)=33.697
(.2)(50)+(.8)(33.697)=36.958
etc.
1ii
i
E)W1(WY
E
E1 = Y1 since no prior information exists
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Sales vs. Smoothed Sales
0
10
20
30
40
50
60
1 2 3 4 5 6 7 8 9 10Time Period
Sa
les
Sales Smoothed
Fluctuations have
been smoothed
The smoothed
value in this case
is generally a little
low, since the
trend is upward
sloping and the
weighting factor is
only .2
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Trend-Based Forecasting
Forecast for time period 6:
Year
Time
Period
(X)
Sales
(y)
1999
2000
2001
2002
2003
2004
2005
0
1
2
3
4
5
6
20
40
30
50
70
65
??
Sales trend
01020304050607080
0 1 2 3 4 5 6
Year
sale
s
79.33
(6) 9.571421.905Y
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Introduction to ARIMA models
The Autoregressive Integrated Moving Average (ARIMA) models, or Box-Jenkins
methodology, are a class of linear models that is capable of representing
stationary as well as non-stationary time series.
ARIMA models rely heavily on autocorrelation patterns in data.
Both ACF and PACF are used to select an initial model.
The Box-Jenkins methodology uses an iterative approach:
An initial model is selected, from a general class of ARIMA models, based
on an examination of the TS and an examination of its autocorrelations for
several time lags.
The chosen model is then checked against the historical data to see
whether it accurately describes the series: the model fits well if the residuals
are generally small, randomly distributed, and contain no useful information.
If the specified model is not satisfactory, the process is repeated using a
new model designed to improve on the original one.
Once a satisfactory model is found, it can be used for forecasting.
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Autoregressive Models AR(p)
An AR(p) model is a regression model with lagged values of the
dependent variable in the independent variable positions, hence the
name autoregressive model.
A pth-order autoregressive model, or AR(p), takes the form:
Autoregressive models are appropriate for stationary time series, and the
coefficient Ф0 is related to the constant level of the series.
response variable at time
observation (predictor variable) at time
regression coefficients to be estimated
error term at time
t
t k
i
t
Y t
Y t k
t
response variable at time
observation (predictor variable) at time
regression coefficients to be estimated
error term at time
t
t k
i
t
Y t
Y t k
t
0 1 1 2 2 ...t t t p t p tY Y Y Y 0 1 1 2 2 ...t t t p t p tY Y Y Y
MBA 532:Business Statistics MBA-2011
Time Series Analysis
AR(p)Theoretical behavior of the ACF and PACF for AR(1) and AR(2) models:
ACF 0
PACF = 0 for lag > 2
ACF 0
PACF = 0 for lag > 2
AR(2)AR(2)
ACF 0
PACF = 0 for lag > 1
ACF 0
PACF = 0 for lag > 1
AR(1)AR(1)
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Moving Average Models MA(q)
MA(q) model is a regression model with the dependent variable, Yt,
depending on previous values of the errors rather than on the
variable itself.
A qth-order moving average model, or MA(q), takes the form:
MA models are appropriate for stationary time series. The weights ωi do
not necessarily sum to 1 and may be positive or negative.
response variable at time
constant mean of the process
regression coefficients to be estimated
error in time period -
t
i
t k
Y t
t k
response variable at time
constant mean of the process
regression coefficients to be estimated
error in time period -
t
i
t k
Y t
t k
1 1 2 2 ...t t t t q t qY 1 1 2 2 ...t t t t q t qY
MBA 532:Business Statistics MBA-2011
Time Series Analysis
MA(q)Theoretical behavior of the ACF and PACF for MA(1) and MA(2) models:
MA(2)MA(2)
ACF = 0 for lag > 2;
PACF 0
ACF = 0 for lag > 2;
PACF 0
MA(1)MA(1)
ACF = 0 for lag > 1;
PACF 0
ACF = 0 for lag > 1;
PACF 0
MBA 532:Business Statistics MBA-2011
Time Series Analysis
ARMA(p,q) Models
A model with autoregressive terms can be combined with a model having moving average terms to get an ARMA(p,q) model:
ARMA(p,q) models can describe a wide variety of behaviors for stationary time series.
Theoretical behavior of the ACF and PACF for autoregressive-moving average processes:
Note that:
• ARMA(p,0) = AR(p)
• ARMA(0,q) = MA(q)
Note that:
• ARMA(p,0) = AR(p)
• ARMA(0,q) = MA(q)
0 1 1 2 2 1 1 2 2... ...t t t p t p t t t q t qY Y Y Y 0 1 1 2 2 1 1 2 2... ...t t t p t p t t t q t qY Y Y Y
In practice, the values of p and q each rarely exceed 2.
In practice, the values of p and q each rarely exceed 2.
ACF PACF
AR(p) Die out Cut off after the order p
of the process
MA(q) Cut off after the order q of the
process
Die out
ARMA(p,q) Die out Die out
In this context…
• “Die out” means “tend to zero gradually”
• “Cut off” means “disappear” or “is zero”
In this context…
• “Die out” means “tend to zero gradually”
• “Cut off” means “disappear” or “is zero”
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Example: Fitting an ARIMA Model
Index
Index
60544842363024181261
290
280
270
260
250
240
230
220
210
Time Series Plot of Index
The series show an upward trend.The series show an upward trend.
Lag
Auto
corr
ela
tion
16151413121110987654321
1,0
0,8
0,6
0,4
0,2
0,0
-0,2
-0,4
-0,6
-0,8
-1,0
Autocorrelation Function for Index(with 5% significance limits for the autocorrelations)
The first several autocorrelations are persistently large and trailed off to zero rather slowly a trend exists and this time series is nonstationary (it does not vary about a fixed level)
The first several autocorrelations are persistently large and trailed off to zero rather slowly a trend exists and this time series is nonstationary (it does not vary about a fixed level)
Idea: to difference the data to see if we could eliminate the trend and create a stationary series.
Idea: to difference the data to see if we could eliminate the trend and create a stationary series.
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Example: Fitting an ARIMA Model…
A plot of the differenced data appears to vary about a fixed level.
A plot of the differenced data appears to vary about a fixed level.
Index
Diff1
60544842363024181261
5
4
3
2
1
0
-1
-2
-3
-4
Time Series Plot of Diff1
Lag
Auto
corr
ela
tion
16151413121110987654321
1,0
0,8
0,6
0,4
0,2
0,0
-0,2
-0,4
-0,6
-0,8
-1,0
Autocorrelation Function for Diff1(with 5% significance limits for the autocorrelations)
Lag
Part
ial A
uto
corr
ela
tion
16151413121110987654321
1,0
0,8
0,6
0,4
0,2
0,0
-0,2
-0,4
-0,6
-0,8
-1,0
Partial Autocorrelation Function for Diff1(with 5% significance limits for the partial autocorrelations)
Comparing the autocorrelations with their error limits, the only significant autocorrelation is at lag 1. Similarly, only the lag 1 partial autocorrelation is significant. The PACF appears to cut off after lag 1, indicating AR(1) behavior. The ACF appears to cut off after lag 1, indicating MA(1) behavior we will try: ARIMA(1,1,0) and ARIMA(0,1,1)
Comparing the autocorrelations with their error limits, the only significant autocorrelation is at lag 1. Similarly, only the lag 1 partial autocorrelation is significant. The PACF appears to cut off after lag 1, indicating AR(1) behavior. The ACF appears to cut off after lag 1, indicating MA(1) behavior we will try: ARIMA(1,1,0) and ARIMA(0,1,1)
A constant term in each model will be included to allow for the fact that the series of differences appears to vary about a level greater than zero.
A constant term in each model will be included to allow for the fact that the series of differences appears to vary about a level greater than zero.
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Example: Fitting an ARIMA Model…
The LBQ statistics are not significant as indicated by the large p-values for either model. The LBQ statistics are not significant as indicated by the large p-values for either model.
ARIMA(1,1,0)ARIMA(1,1,0)
ARIMA(0,1,1)ARIMA(0,1,1)
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Example: Fitting an ARIMA Model…
Finally, there is no significant residual autocorrelation for the ARIMA(1,1,0) model. The results for the ARIMA(0,1,1) are similar.
Finally, there is no significant residual autocorrelation for the ARIMA(1,1,0) model. The results for the ARIMA(0,1,1) are similar.
Lag
Auto
corr
ela
tion
16151413121110987654321
1,0
0,8
0,6
0,4
0,2
0,0
-0,2
-0,4
-0,6
-0,8
-1,0
Autocorrelation Function for RESI1(with 5% significance limits for the autocorrelations)
Lag
Auto
corr
ela
tion
16151413121110987654321
1,0
0,8
0,6
0,4
0,2
0,0
-0,2
-0,4
-0,6
-0,8
-1,0
Autocorrelation Function for RESI2(with 5% significance limits for the autocorrelations)
Therefore, either model is adequate and provide nearly the same one-step-ahead forecasts.
Therefore, either model is adequate and provide nearly the same one-step-ahead forecasts.
MBA 532:Business Statistics MBA-2011
Time Series Analysis
The first sample ACF coefficient is significantly different form zero. The autocorrelation at lag 2 is close to significant and opposite in sign from the lag 1 autocorrelation. The remaining autocorrelations are small. This suggests either an AR(1) model or an MA(2) model.
The first PACF coefficient is significantly different from zero, but none of the other partial autocorrelations approaches significance, This suggests an AR(1) or ARIMA(1,0,0)
The first sample ACF coefficient is significantly different form zero. The autocorrelation at lag 2 is close to significant and opposite in sign from the lag 1 autocorrelation. The remaining autocorrelations are small. This suggests either an AR(1) model or an MA(2) model.
The first PACF coefficient is significantly different from zero, but none of the other partial autocorrelations approaches significance, This suggests an AR(1) or ARIMA(1,0,0)
Lag
Auto
corr
ela
tion
18161412108642
1,0
0,8
0,6
0,4
0,2
0,0
-0,2
-0,4
-0,6
-0,8
-1,0
Autocorrelation Function for Readings(with 5% significance limits for the autocorrelations)
ARIMA
The time series of readings appears to vary about a fixed level of around 80, and the autocorrelations die out rapidly toward zero the time series seems to be stationary.
The time series of readings appears to vary about a fixed level of around 80, and the autocorrelations die out rapidly toward zero the time series seems to be stationary.
Index
Readin
gs
70635649423528211471
110
100
90
80
70
60
50
40
30
20
Time Series Plot of Readings
Lag
Part
ial A
uto
corr
ela
tion
18161412108642
1,0
0,8
0,6
0,4
0,2
0,0
-0,2
-0,4
-0,6
-0,8
-1,0
Partial Autocorrelation Function for Readings(with 5% significance limits for the partial autocorrelations)
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Both models appear to fit the data well. The estimated coefficients are significantly different from zero and the mean square (MS) errors are similar.
Both models appear to fit the data well. The estimated coefficients are significantly different from zero and the mean square (MS) errors are similar.
ARIMA
AR(1) = ARIMA(1,0,0)
AR(1) = ARIMA(1,0,0)
MA(2) = ARIMA(0,0,2)
MA(2) = ARIMA(0,0,2)
A constant term is included in both models to allow for the fact that the readings vary about a level other than zero.
A constant term is included in both models to allow for the fact that the readings vary about a level other than zero.
Let’s take a look at the residuals ACF…Let’s take a look at the residuals ACF…
MBA 532:Business Statistics MBA-2011
Time Series Analysis
ARIMA
Finally, there is no significant residual autocorrelation for the ARIMA(1,0,0) model. The results for the ARIMA(0,0,2) are similar.
Finally, there is no significant residual autocorrelation for the ARIMA(1,0,0) model. The results for the ARIMA(0,0,2) are similar.
Therefore, either model is adequate and provide nearly the same three-step-ahead forecasts. Since the AR(1) model has two parameters (including the constant term) and the MA(2) model has three parameters, applying the principle of parsimony we would use the simpler AR(1) model to forecast future readings.
Therefore, either model is adequate and provide nearly the same three-step-ahead forecasts. Since the AR(1) model has two parameters (including the constant term) and the MA(2) model has three parameters, applying the principle of parsimony we would use the simpler AR(1) model to forecast future readings.
Lag
Auto
corr
ela
tion
18161412108642
1,0
0,8
0,6
0,4
0,2
0,0
-0,2
-0,4
-0,6
-0,8
-1,0
Autocorrelation Function for RESI1(with 5% significance limits for the autocorrelations)
Lag
Auto
corr
ela
tion
18161412108642
1,0
0,8
0,6
0,4
0,2
0,0
-0,2
-0,4
-0,6
-0,8
-1,0
Autocorrelation Function for RESI2(with 5% significance limits for the autocorrelations)
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Building an ARIMA model The first step in model identification is to determine whether the series is
stationary. It is useful to look at a plot of the series along with the sample ACF.
If the series is not stationary, it can often be converted to a stationary series by differencing: the original series is replaced by a series of differences and an ARMA model is then specified for the differenced series (in effect, the analyst is modeling changes rather than levels)
Models for nonstationary series are called Autoregressive Integrated Moving Average models, or ARIMA(p,d,q), where d indicates the amount of differencing.
Once a stationary series has been obtained, the analyst must identify the form of the model to be used by comparing the sample ACF and PACF to the theoretical ACF and PACF for the various ARIMA models.
Principle of parsimony: “all things being equal, simple models are preferred to complex models”
Once a tentative model has been selected, the parameters for that model are estimated using least squares estimates.
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Building an ARIMA model …
Before using the model for forecasting, it must be checked for adequacy. Basically, a model is adequate if the residuals cannot be used to improve the forecasts, i.e.,
The residuals should be random and normally distributed
The individual residual autocorrelations should be small. Significant residual autocorrelations at low lags or seasonal lags suggest the model is inadequate
After an adequate model has been found, forecasts can be made. Prediction intervals based on the forecasts can also be constructed.
As more data become available, it is a good idea to monitor the forecast errors, since the model must need to be reevaluated if:
The magnitudes of the most recent errors tend to be consistently larger than previous errors, or
The recent forecast errors tend to be consistently positive or negative
Seasonal ARIMA (SARIMA) models contain:
Regular AR and MA terms that account for the correlation at low lags
Seasonal AR and MA terms that account for the correlation at the seasonal lags
MBA 532:Business Statistics MBA-2011
Time Series Analysis
Introduction to ARIMA models
MBA 532:Business Statistics MBA-2011
Time Series AnalysisRushan A B Abeygunawardana 44Wednesday, April 12, 2023
The End