Lecture 1 Introduction
Basic Ideas of Time Series
• Example 1
Basic Ideas of Time Series
• Example 2: Global warming
Forecasting Problems
• Problem Type 1: Causal regression
• Problem: Predict 𝑋𝑛+𝑘 from 𝑌1 , 𝑌2 , … , 𝑌𝑛
• Objective: Making forecasting based on another related time series data.
• Principles: The new value of X can be related to the old values of Y. Time
series methods that you learned in this course can be used to discover such
hidden relationships.
• Computer: Use R to do statistics.
Forecasting Problems
• Problem Type 2: Autoregression
• Problem: Predict 𝑋𝑛+𝑘 from 𝑋1 , 𝑋2 , … , 𝑋𝑛
• Objective: Making forecasting based on the historical data.
• Principles: The new value and the old values can be related. Time series
methods that you learned in this course can be used to discover such
hidden relationships.
• Computer: Use R to do statistics.
Preparing Data in R
• Read external data files, e.g. txt files and csv files
• Example 1: Global temperature data
• Set the working directory by the command:
setwd("C:/AMS4322 Fall 2023")
gtemp = read.csv("gtemp.csv",header=T)
• gtemp
Preparing Data in R
• Read external data files, e.g. txt files and csv files
• Example 2: U.S. savings and income data
setwd("C:/AMS4322 Fall 2023")
savings = read.table("savings.txt", sep = "\t", skip = 4, header=T)
• savings
Sample Covariance / Autocovariance
• Concepts of covariance/correlation
• Is X useful in predicting Y? Regression class
• Is today useful in predicting tomorrow? Time series class
• Is yesterday useful in predicting tomorrow ? Time series class
• These can be answered via covariance
Sample Covariance / Autocovariance
• Sample covariance:
𝑛
1
𝑋, 𝑌 = (𝑋𝑡 − 𝑋)(𝑌
C𝑜𝑣 ത 𝑡 − 𝑌)
ത
𝑛
𝑡=1
• Sample correlation:
(𝑋, 𝑌)
𝐶𝑜𝑣
𝑋, 𝑌 =
C𝑜𝑟𝑟
(𝑋)𝑉𝑎𝑟
𝑉𝑎𝑟 (𝑌)
Sample Covariance / Autocovariance
• Sample autocovariance:
𝑛−𝑘
1
ത 𝑡+𝑘 − 𝑋)
𝛾ො𝑘 = (𝑋𝑡 − 𝑋)(𝑋 ത
𝑛
𝑡=1
• Sample autocorrelation:
𝛾ො𝑘
𝜌ො𝑘 =
𝛾ො0
• Here, k is called the lag
Sample Covariance / Autocovariance
• In R, you can see the sample autocorrelations from the output of an R
function called acf. You will learn in more details the acf methods
latter on in this course.
• acf(gtemp$Temperature,20)
• acf(savings$SAVINGS, 20)
Concepts of
Stationarity
Concepts of Stationarity
• Mathematical details:
• Precise definition of weak stationarity: The mean, variance, and
autocorrelation structure do not change over time.
• 𝐸[𝑋𝑡 ] = 𝜇 for all t
• 𝐶𝑜𝑣 𝑋𝑡 , 𝑋𝑡−𝑘 = 𝐸 𝑋𝑡 − 𝜇 𝑋𝑡−𝑘 − 𝜇 = 𝛾𝑘 for all t
• 𝛾𝑘 is called autocovariance function. It is meaningful only for the weakly
stationary process.
Concepts of Stationarity
• Example 1: White noise process: Why to learn that? It is the building
block of time series model
• It means a sequence of uncorrelated random variables 𝑎1 , 𝑎2 , 𝑎3 , …
with mean zero and constant variance 𝜎𝑎2
𝜎𝑎2 , 𝑖𝑓 𝑘 = 0
• In this case, 𝐸[𝑎𝑡 ] = 0 and 𝐶𝑜𝑣 𝑎𝑡 , 𝑎𝑡−𝑘 =ቊ
0, 𝑖𝑓 𝑘 ≠ 0
Both do not depend on t. So, white noise process is weakly stationary.
Concepts of Stationarity
• Exercise:
• Let 𝑎1 , 𝑎2 , 𝑎3 , … be a white noise process. A new process is defined as
𝑋𝑡 = 3 + 𝑡𝑎𝑡
• Is X a weakly stationary process?
Concepts of Stationarity
• Exercise:
• Let 𝑎1 , 𝑎2 , 𝑎3 , … be a white noise process. A new process is defined as
𝑋𝑡 = 𝑎1 + 𝑎2 + ⋯ + 𝑎𝑡
• Is X a weakly stationary process?
Concepts of Stationarity
• Descriptive statistics like sample mean, sample variance, sample
autocovariance, etc. of non-stationary time series data are considered
NOT meaningful.
• If the time series data is non-stationary, the people usually do some
transformation to make the time series stationary.
Concepts of Stationarity
• In many cases, the non-stationarity can be removed by taking logarithm
and taking differencing
• The new sequence can be “less” non-stationary
𝑍𝑡 = 𝑋𝑡 − 𝑋𝑡−1
• To study stock price data, it is common to consider
𝑍𝑡 = log 𝑋𝑡 − log 𝑋𝑡−1
• In finance, people call it rate of return
Concepts of Stationarity
• Example of differencing in R:
• X = diff(log(savings$SAVINGS))
• acf(X, 20)
