Applied Time Series Analysis with R

Applied Time Series Analysis with R

1 Stationary Time Series

    1.1 Time Series
    1.2 Stationary Time Series
    1.3 Autocovariance and Autocorrelation Functions for Stationary Time Series
    1.4 Estimation of the Mean, Autocovariance, and Autocorrelation for Stationary Time Series
    1.5 Power Spectrum
    1.6 Estimating the Power Spectrum and Spectral Density for Discrete Time Series
    1.7 Time Series Examples

2 Linear Filters
    2.1 Introduction to Linear Filters
    2.1.1 Relationship between the Spectra of the Input and Output of a Linear Filter
    2.2 Stationary General Linear Processes
    2.2.1 Spectrum and Spectral Density for a General Linear Process
    2.3 Wold Decomposition Theorem
    2.4 Filtering Applications
    2.4.1 Butterworth Filters

3 ARMA Time Series Models
    3.1 MA Processes
    3.2 AR Processes
        3.2.1 Inverting the Operator
        3.2.2 AR(1) Model
        3.2.3 AR( p) Model for  p ≥ 1
        3.2.4 Autocorrelations of an AR( p) Model
        3.2.5 Linear Difference Equations
        3.2.6 Spectral Density of an AR( p) Model
        3.2.7 AR(2) Model
            3.2.7.1 Autocorrelations of an AR(2) Model
            3.2.7.2 Spectral Density of an AR(2)
            3.2.7.3 Stationary/Causal Region of an AR(2)
            3.2.7.4 ψ-Weights of an AR(2) Model
        3.2.8 Summary of AR(1) and AR(2) Behavior
        3.2.9 AR( p) Model
        3.2.10  AR(1) and AR(2) Building Blocks of an AR( p) Model
        3.2.11  Factor Tables
        3.2.12  Invertibility/Infinite-Order AR Processes
        3.2.13  Two Reasons for Imposing Invertibility

    3.3 ARMA Processes
        3.3.1 Stationarity and Invertibility Conditions for an ARMA( p,q) Model
        3.3.2 Spectral Density of an ARMA( p,q) Model
        3.3.3 Factor Tables and ARMA( p,q) Models
        3.3.4 Autocorrelations of an ARMA( p,q) Model
        3.3.5 ψ-Weights of an ARMA( p,q)
        3.3.6 Approximating ARMA( p,q) Processes Using High-Order AR( p) Models
    3.4 Visualizing AR Components
    3.5 Seasonal ARMA(p,q) × (PS,QS)S Models

    3.6 Generating Realizations from ARMA( p,q) Processes
        3.6.1 MA( q) Model
        3.6.2 AR(2) Model
        3.6.3 General Procedure
    3.7 Transformations
        3.7.1 Memoryless Transformations
        3.7.2 AR Transformations

4 Other Stationary Time Series Models
    4.1 Stationary Harmonic Models
        4.1.1 Pure Harmonic Models
        4.1.2 Harmonic Signal-Plus-Noise Models
        4.1.3 ARMA Approximation to the Harmonic Signal-Plus-Noise Model
    4.2 ARCH and GARCH Processes
        4.2.1 ARCH Processes
            4.2.1.1 The ARCH(1) Model
            4.2.1.2 The ARCH( q 0) Model
        4.2.2 The GARCH(p0, q0) Process
        4.2.3 AR Processes with ARCH or GARCH Noise

5 Nonstationary Time Series Models
    5.1 Deterministic Signal-Plus-Noise Models
        5.1.1 Trend-Component Models
        5.1.2 Harmonic Component Models

    5.2 ARIMA(p,d,q) and ARUMA(p,d,q) Processes
        5.2.1 Extended Autocorrelations of an ARUMA( p,d,q) Process
        5.2.2 Cyclical Models
    5.3 Multiplicative Seasonal ARUMA (p,d,q) × (Ps, Ds, Qs)s Process
    5.3.1 Factor Tables for Seasonal Models of the Form of Equation
    5.4 Random Walk Models
        5.4.1 Random Walk
        5.4.2 Random Walk with Drift
    5.5 G-Stationary Models for Data with Time-Varying Frequencies

6 Forecasting
    6.1 Mean-Square Prediction Background
    6.2 Box–Jenkins Forecasting for ARMA(p,q) Models
        6.2.1 General Linear Process Form of the Best Forecast Equation
    6.3 Properties of the Best Forecast
    6.4 π-Weight Form of the Forecast Function
    6.5 Forecasting Based on the Difference Equation
        6.5.1 Difference Equation Form of the Best Forecast Equation
        6.5.2 Basic Difference Equation Form for Calculating Forecasts from an ARMA(p,q) Model
    6.6 Eventual Forecast Function
    6.7 Assessing Forecast Performance
        6.7.1 Probability Limits for Forecasts
        6.7.2 Forecasting the Last k Values
    6.8 Forecasts Using ARUMA(p,d,q) Models
    6.9 Forecasts Using Multiplicative Seasonal ARUMA Models
    6.10 Forecasts Based on Signal-Plus-Noise Models

7 Parameter Estimation
    7.1 Introduction
    7.2 Preliminary Estimates
        7.2.1 Preliminary Estimates for AR( p) Models
        7.2.1.1 Yule–Walker Estimates
        7.2.1.2 Least Squares Estimation
        7.2.1.3 Burg Estimates
        7.2.2 Preliminary Estimates for  MA(q) Models
        7.2.2.1 MM Estimation for an  MA(q)
        7.2.2.2 MA(q) Estimation Using the Innovations Algorithm
        7.2.3 Preliminary Estimates for ARMA( p,q) Models
        7.2.3.1 Extended Yule–Walker Estimates of the AR
        Parameters
        7.2.3.2 Tsay–Tiao Estimates of the AR Parameters
        7.2.3.3 Estimating the MA Parameters
    7.3 ML Estimation of ARMA( p,q) Parameters
        7.3.1 Conditional and Unconditional ML Estimation
        7.3.2 ML Estimation Using the Innovations Algorithm
    7.4 Backcasting and Estimating 
    7.5 Asymptotic Properties of Estimators
        7.5.1 AR Case
        7.5.1.1 Confidence Intervals: AR Case
        7.5.2 ARMA( p,q) Case
        7.5.2.1 Confidence Intervals for ARMA( p,q) Parameters
        7.5.3 Asymptotic Comparisons of Estimators for an MA(1)
    7.6 Estimation Examples Using Data
    7.7 ARMA Spectral Estimation
    7.8 ARUMA Spectral Estimation

8 Model Identification
    8.1 Preliminary Check for White Noise
    8.2 Model Identification for Stationary ARMA Models
        8.2.1 Model Identification Based on AIC and Related Measures
    8.3 Model Identification for Nonstationary ARUMA( p,d,q) Models
        8.3.1 Including a Nonstationary Factor in the Model
        8.3.2 Identifying Nonstationary Component(s) in a Model
        8.3.3 Decision Between a Stationary or a Nonstationary Model
        8.3.4 Deriving a Final ARUMA Model
        8.3.5 More on the Identification of Nonstationary Components
        8.3.5.1 Including a Factor (1 −  B)d in the Model
        8.3.5.2 Testing for a Unit Root
        8.3.5.3 Including a Seasonal Factor (1 −  Bs) in the Model

9 Model Building
9.1 Residual Analysis
    9.1.1 Check Sample Autocorrelations of Residuals versus 95% Limit Lines
    9.1.2 Ljung–Box Test
    9.1.3 Other Tests for Randomness
    9.1.4 Testing Residuals for Normality
9.2 Stationarity versus Nonstationarity
9.3 Signal-Plus-Noise versus Purely Autocorrelation-Driven Models
    9.3.1 Cochrane–Orcutt and Other Methods
    9.3.2 A Bootstrapping Approach
    9.3.3 Other Methods for Trend Testing
9.4 Checking Realization Characteristics
9.5 Comprehensive Analysis of Time Series Data: A Summary

10  Vector-Valued (Multivariate) Time Series
10.1  Multivariate Time Series Basics
10.2  Stationary Multivariate Time Series
    10.2.1  Estimating the Mean and Covariance for Stationary Multivariate Processes
    10.2.1.1  Estimating  μ
    10.2.1.2  Estimating T(k)
10.3  Multivariate (Vector) ARMA Processes
    10.3.1  Forecasting Using VAR( p) Models
    10.3.2  Spectrum of a VAR( p) Model
    10.3.3  Estimating the Coefficients of a VAR( p) Model
    10.3.3.1  Yule–Walker Estimation
    10.3.3.2  Least Squares and Conditional ML Estimation
    10.3.3.3  Burg-Type Estimation
    10.3.4  Calculating the Residuals and Estimating Γa
    10.3.5  VAR( p) Spectral Density Estimation
    10.3.6  Fitting a VAR( p) Model to Data
    10.3.6.1  Model Selection
    10.3.6.2  Estimating the Parameters
    10.3.6.3  Testing the Residuals for White Noise
10.4  Nonstationary VARMA Processes
10.5  Testing for Association between Time Series
    10.5.1  Testing for Independence of Two Stationary Time Series
    10.5.2  Testing for Cointegration between Nonstationary Time Series
10.6  State-Space Models
    10.6.1  State Equation
    10.6.2  Observation Equation
    10.6.3  Goals of State-Space Modeling
    10.6.4  Kalman Filter
    10.6.4.1  Prediction (Forecasting)
    10.6.4.2  Filtering
    10.6.4.3  Smoothing Using the Kalman Filter
    10.6.4.4   h-Step Ahead Predictions
    10.6.5  Kalman Filter and Missing Data
    10.6.6  Parameter Estimation
    10.6.7  Using State-Space Methods to Find Additive Components of a Univariate AR Realization
    10.6.7.1  Revised State-Space Model
    10.6.7.2   Ψj Real
    10.6.7.3   Ψj Complex
    

11  Long-Memory Processes
11.1  Long Memory
11.2  Fractional Difference and FARMA Processes
11.3  Gegenbauer and GARMA Processes
    11.3.1  Gegenbauer Polynomials
    11.3.2  Gegenbauer Process
    11.3.3  GARMA Process
11.4   k-Factor Gegenbauer and GARMA Processes
    11.4.1  Calculating Autocovariances
    11.4.2  Generating Realizations
11.5  Parameter Estimation and Model Identification
11.6  Forecasting Based on the  k-Factor GARMA Model
11.7  Testing for Long Memory
    11.7.1  Testing for Long Memory in the Fractional and FARMA Setting
    11.7.2  Testing for Long Memory in the Gegenbauer Setting
11.8  Modeling Atmospheric CO2 Data Using Long-Memory Models
    

12  Wavelets
12.1  Shortcomings of Traditional Spectral Analysis for TVF Data
12.2  Window-Based Methods that Localize the “Spectrum” in Time
    12.2.1  Gabor Spectrogram
    12.2.2  Wigner–Ville Spectrum
12.3  Wavelet Analysis
    12.3.1  Fourier Series Background
    12.3.2  Wavelet Analysis Introduction
    12.3.3  Fundamental Wavelet Approximation Result
    12.3.4  Discrete Wavelet Transform for Data Sets of Finite Length
    12.3.5  Pyramid Algorithm
    12.3.6  Multiresolution Analysis
    12.3.7  Wavelet Shrinkage
    12.3.8  Scalogram: Time-Scale Plot
    12.3.9  Wavelet Packets
    12.3.10 Two-Dimensional Wavelets
12.4  Concluding Remarks on Wavelets

13  G-Stationary Processes
13.1  Generalized-Stationary Processes
    13.1.1  General Strategy for Analyzing G-Stationary Processes
13.2  M-Stationary Processes
    13.2.1  Continuous M-Stationary Process
    13.2.2  Discrete M-Stationary Process
    13.2.3  Discrete Euler (p) Model
    13.2.4  Time Transformation and Sampling
13.3   G(λ)-Stationary Processes
    13.3.1  Continuous  G(p;  λ) Model
    13.3.2  Sampling the Continuous  G(λ)-Stationary Processes
    13.3.2.1  Equally Spaced Sampling from  G(p; λ) Processes
    13.3.3  Analyzing TVF Data Using the  G(p; λ) Model
    13.3.3.1   G(p; λ) Spectral Density
13.4  Linear Chirp Processes
    13.4.1  Models for Generalized Linear Chirps
13.5  G-Filtering
13.6  Concluding Remarks