DATA BRIO ACADEMY DATA BRIO ACADEMY
TIME SERIES
TIME SERIES
ANALYSIS
ANALYSIS
What is a Time Series? What is a Time Series?Databrio
Databrio
2/18/2016
Trusted by over 1 million members
Try Scribd FREE for 30 days to access over 125 million titles without ads or interruptions!
Start Free Trial Cancel Anytime.
Trusted by over 1 million members
Try Scribd FREE for 30 days to access over 125 million titles without ads or interruptions!
Start Free Trial Cancel Anytime.
When we have a
When we have a chronologicchronologically ordered collection (set) of ally ordered collection (set) of data points, wedata points, we
refer to the data set as time
refer to the data set as time series. So, a time series is a series. So, a time series is a sequence ofsequence of
observations taken sequentially in time series
observations taken sequentially in time series data can have both univariatedata can have both univariate
and multivariate quantitative data collected over time.
and multivariate quantitative data collected over time.
For example, let us say that we have the
For example, let us say that we have the attrition rate data of a company forattrition rate data of a company for
the past 12
the past 12 months. The senior manager wants to know the months. The senior manager wants to know the probable attritionprobable attrition
rate for the 13
rate for the 13thth and 14 and 14thth month, so that he can prepare his current month, so that he can prepare his current
workforce and initiate any recruitment process if necessary. As we
workforce and initiate any recruitment process if necessary. As we have thehave the
data points arranged chronologically, we say that the data
data points arranged chronologically, we say that the data is ais a time seriestime series data
data.. For predicting the For predicting the probable attrition rate for any future period, weprobable attrition rate for any future period, we
have to use time series analysis which has been discussed belo
Trusted by over 1 million members
Try Scribd FREE for 30 days to access over 125 million titles without ads or interruptions!
Start Free Trial Cancel Anytime.
There are two classes of
There are two classes of time series process: Stationary and Non-Stationarytime series process: Stationary and Non-Stationary
So, what is stationarity? Covariance stationarity follows three
So, what is stationarity? Covariance stationarity follows three
conditions-1) Unconditional mean and variance should be constant
1) Unconditional mean and variance should be constant
E(Y
E(Ytt) = E(Y) = E(Yt+jt+j) = µ) = µ
Var (Y
Var (Ytt) = Var(Y) = Var(Yt+jt+j)=σ )=σ 22
2) Covariance depends on time
2) Covariance depends on time j that has j that has elapsed between observations, not onelapsed between observations, not on
reference period.
reference period.
Cov(Y
Cov(Ytt,Y,Yt+jt+j) = Cov(Y) = Cov(Yss,Y,Ys+js+j) = γ) = γ
Any
Any time series datatime series data which follows the which follows the above mentioned conditions are knownabove mentioned conditions are known
as stationary time series. Similarly, if a time series data do
as stationary time series. Similarly, if a time series data do not conform tonot conform to
the above conditions, they are termed as
the above conditions, they are termed as non-stationary time series data. Fornon-stationary time series data. For
a non-stationary time series, the mean, variance and
a non-stationary time series, the mean, variance and the covariance changes.the covariance changes.
There is no long-run mean to which the series returns. Also, the variance is
There is no long-run mean to which the series returns. Also, the variance is
tie-dependent
tie-dependent, for eg., it , for eg., it could go to infinity as the number could go to infinity as the number of observationof observation
goes to infinity.
goes to infinity.
Unit root tests are used to find
Unit root tests are used to find out non-stationary time series. One of theout non-stationary time series. One of the
commonly used tests for
Trusted by over 1 million members
Try Scribd FREE for 30 days to access over 125 million titles without ads or interruptions!
Start Free Trial Cancel Anytime.
The process flow for time-series analysis is as follows:
The process flow for time-series analysis is as follows:
At first, using unit root
At first, using unit root tests find out whether the time series is stationarytests find out whether the time series is stationary
or not. If it is
or not. If it is stationary, procestationary, proceed to find out ed to find out the best ARMA model usingthe best ARMA model using
different diagnostic tests. After selecting the best
different diagnostic tests. After selecting the best suited model, forecast forsuited model, forecast for
future periods and again use different diagnostic tests to find out
future periods and again use different diagnostic tests to find out how goodhow good
the forecast is.
the forecast is.
If in case the unit
If in case the unit root test like “Dickeyroot test like “Dickey--Fuller” test shows the time series toFuller” test shows the time series to
be non-stationary, then you have to transform the
be non-stationary, then you have to transform the data into stationary series.data into stationary series.
Differencing is widely used to transform the data into stationary series.
Differencing is widely used to transform the data into stationary series.
Once, the data is
Once, the data is transformed into stationary time series, follow the previoustransformed into stationary time series, follow the previous
steps to forecast the model.
Trusted by over 1 million members
Try Scribd FREE for 30 days to access over 125 million titles without ads or interruptions!
Start Free Trial Cancel Anytime.
Stationary Process:
Stationary Process:
After identification of a stationary time series
After identification of a stationary time series process, estimation and modelprocess, estimation and model
selection is done. Stationary Process can be of three basic types:
selection is done. Stationary Process can be of three basic types:
1.
1. Autoregressive(AR)-Autoregressive(AR)-It means that the It means that the variable is a function of variable is a function of its ownits own
lagged values upto a maximum lag of
lagged values upto a maximum lag of p.p.
2.
2. Moving Average(MA)-It means the variable is a function of Moving Average(MA)-It means the variable is a function of thethe
disturbances upto a maximum lag of q.
disturbances upto a maximum lag of q.
3.
3. Combined(ARMACombined(ARMA)-It includes both the elements, i.e. )-It includes both the elements, i.e. have lagged values have lagged values ofof
the variable and lagged values of the disturbance.
the variable and lagged values of the disturbance.
So, for estimation of time series and model selection, decide whether the time
So, for estimation of time series and model selection, decide whether the time
series is a pure AR/ MA
series is a pure AR/ MA or ARMA process. Then estimate the specificationsor ARMA process. Then estimate the specifications
like auto-covariance, auto-correlation and partial
like auto-covariance, auto-correlation and partial auto- correlation.auto- correlation.
Finally,
Finally, choose the best model choose the best model based on based on the significance of coefficients, whitethe significance of coefficients, white
noise residuals, fit vs parsimony and ability to forecast.
Trusted by over 1 million members
Try Scribd FREE for 30 days to access over 125 million titles without ads or interruptions!
Start Free Trial Cancel Anytime.
CONTACT US :
CONTACT US :
ADDRESS :
ADDRESS :
PIN CODE PIN CODE :: PHONE PHONE : : E-MAIL E-MAIL :: WEBSITEWEBSITE : : http://www.databrio.com/http://www.databrio.com/
Data Brio Academy 1st Floor, 135/L SP Mukherjee Road, Data Brio Academy 1st Floor, 135/L SP Mukherjee Road, Near RashBehari Crossing, Lake Range, Kalighat,
Near RashBehari Crossing, Lake Range, Kalighat, Kolkata, West Bengal
Kolkata, West Bengal 700026 700026 033 24660329 033 24660329 [email protected] [email protected]