Elements of Forecasting

in Business, Finance, Economics and Government Francis X. Diebold

Department of Economics University of Pennsylvania

Slides for Projection

Copyright F.X. Diebold. All rights reserved.

Introduction to Forecasting:

Applications, Methods, Books, Journals and Software

1. Forecasting in Action a. Operations planning and control b. Marketing c. Economics d. Financial speculation e. Financial risk management f. Capacity planning g. Business and government budgeting i. Demography j. Crisis management

2. Forecasting Methods: An Overview

[Review of probability, statistics and regression]

Six Considerations Basic to Successful Forecasting

Forecasts and decisions The object to be forecast Forecast types The forecast horizon The information set Methods and complexity, the parsimony principle, and the shrinkage principle

Statistical Graphics for Forecasting

Why graphical analysis is important Simple graphical techniques Elements of graphical style Application: graphing four components of real GNP

Modeling and Forecasting Trend

Modeling trend Estimating trend models Forecasting trend Selecting forecasting models using the Akaike and Schwarz criteria Application: forecasting retail sales

Modeling and Forecasting Seasonality

The nature and sources of seasonality Modeling seasonality Forecasting seasonal series Application: forecasting housing starts

Characterizing Cycles

Covariance stationary time series White noise The lag operator Wolds theorem, the general linear process, and rational distributed lags Estimation and inference for the mean, autocorrelation and partial autocorrelation functions Application: characterizing Canadian employment dynamics

Modeling Cycles: MA, AR and ARMA Models

Moving-average (MA) models Autoregressive (AR) models Autoregressive moving average (ARMA) models Application: specifying and estimating models for forecasting employment

Forecasting Cycles

Optimal forecasts Forecasting moving average processes Forecasting infinite-ordered moving averages Making the forecasts operational The chain rule of forecasting Application: forecasting employment

Putting it all Together: A Forecasting Model with Trend, Seasonal and Cyclical Components

Assembling what we've learned Application: forecasting liquor sales Recursive estimation procedures for diagnosing and selecting forecasting models

Forecasting with Regression Models Conditional forecasting models and scenario analysis Accounting for parameter uncertainty in confidence intervals for conditional forecasts Unconditional forecasting models Distributed lags, polynomial distributed lags, and rational distributed lags Regressions with lagged dependent variables, regressions with ARMA disturbances, and transfer function models Vector autoregressions Predictive causality Impulse-response functions and variance decompositions Application: housing starts and completions

Evaluating and Combining Forecasts

Evaluating a single forecast Evaluating two or more forecasts: comparing forecast accuracy Forecast encompassing and forecast combination Application: OverSea shipping volume on the Atlantic East trade lane

Unit Roots, Stochastic Trends, ARIMA Forecasting Models, and Smoothing

Stochastic trends and forecasting Unit roots: estimation and testing Application: modeling and forecasting the yen/dollar exchange rate Smoothing Exchange rates, continued

Volatility Measurement, Modeling and Forecasting The basic ARCH process The GARCH process Extensions of ARCH and GARCH models Estimating, forecasting and diagnosing GARCH models Application: stock market volatility

3. Useful Books, Journals and Software Books Statistics review, etc.:

Wonnacott, T.H. and Wonnacott, R.J. (1990), Introductory Statistics, Fifth Edition. New York: John Wiley and Sons. Pindyck, R.S. and Rubinfeld, D.L. (1997), Econometric Models and Economic Forecasts, Fourth Edition. New York: McGraw-Hill. Maddala, G.S. (2001), Introduction to Econometrics, Third Edition. New York: Macmillan. Kennedy, P. (1998), A Guide to Econometrics, Fourth Edition. Cambridge, Mass.: MIT Press.

Time series analysis:

Chatfield, C. (1996), The Analysis of Time Series: An Introduction, Fifth Edition. London: Chapman and Hall. Granger, C.W.J. and Newbold, P. (1986), Forecasting Economic Time Series, Second Edition. Orlando, Florida: Academic Press. Harvey, A.C. (1993), Time Series Models, Second Edition. Cambridge, Mass.: MIT Press. Hamilton, J.D. (1994), Time Series Analysis. Princeton: Princeton University Press.

Special insights: Armstrong, J.S. (Ed.) (1999), The Principles of Forecasting. Norwell, Mass.: Kluwer Academic Forecasting. Makridakis, S. and Wheelwright S.C. (1997), Forecasting: Methods and Applications, Third Edition. New York: John Wiley.
Bails, D.G. and Peppers, L.C. (1997), Business Fluctuations. Englewood Cliffs: Prentice Hall.

Taylor, S. (1996), Modeling Financial Time Series, Second Edition. New York: Wiley.

Journals Journal of Forecasting International Journal of Forecasting Journal of Business Forecasting Methods and Systems

Journal of Business and Economic Statistics Review of Economics and Statistics Journal of Applied Econometrics

Software

General: Eviews, S+, Minitab, SAS, etc.

Cross-section: Stata Open-ended: Matlab

Online Information

Resources for Economists

A Brief Review of Probability, Statistics, and Regression for Forecasting

Discrete Random Variable Discrete Probability Distribution Continuous Random Variable Probability Density Function Moment Mean, or Expected Value Location, or Central Tendency Variance Dispersion, or Scale Standard Deviation Skewness Asymmetry Kurtosis Leptokurtosis Normal, or Gaussian, Distribution

Marginal Distribution Joint Distribution Covariance Correlation Conditional Distribution Conditional Moment Conditional Mean Conditional Variance Population Distribution Sample Estimator Statistic, or Sample Statistic Sample Mean Sample Variance Sample Standard Deviation Sample Skewness Sample Kurtosis

Distribution t Distribution F Distribution Jarque-Bera Test

Regression as curve fitting Least-squares estimation:

Fitted values:

Residuals:

Regression as a probabilistic model Simple regression:

Multiple regression:

Mean dependent var 10.23

S.D. dependent var 1.49

Sum squared resid 43.70

F-statistic 30.89

S.E. of regression 0.99

R-squared 0.58

or

Adjusted R-squared 0.56

Akaike info criterion 0.03

Schwarz criterion 0.15

Durbin-Watson stat 1.97

Regression of y on x and z LS // Dependent Variable is Y Sample: 1960 2007 Included observations: 48 Variable C X Z Coefficient 9.884732 1.073140 -0.638011 Std. Error 0.190297 0.150341 0.172499 t-Statistic 51.94359 7.138031 -3.698642 Prob. 0.0000 0.0000 0.0006 10.08241 1.908842 3.429780 3.546730 27.82752 0.000000

R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat

0.552928 0.533059 1.304371 76.56223 -79.31472 1.506278

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic)

Scatterplot of y versus x

Scatterplot of y versus x Regression Line Superimposed

Scatterplot of y versus z Regression Line Superimposed

Residual Plot Regression of y on x and z

Six Considerations Basic to Successful Forecasting 1. The Decision Environment and Loss Function

2. The Forecast Object Event outcome, event timing, time series. 3. The Forecast Statement Point forecast, interval forecast, density forecast, probability forecast

4. The Forecast Horizon h-step ahead forecast h-step-ahead extrapolation forecast 5. The Information Set

6. Methods and Complexity, the Parsimony Principle, and the Shrinkage Principle Signal vs. noise Smaller is often better Even incorrect restrictions can help

Decision Making with Symmetric Loss Demand High Build Inventory Reduce Inventory 0 $10,000 Demand Low $10,000 0

Decision Making with Asymmetric Loss Demand High Build Inventory Reduce Inventory 0 $20,000 Demand Low $10,000 0

Forecasting with Symmetric Loss High Actual Sales High Forecasted Sales Low Forecasted Sales 0 $10,000 Low Actual Sales $10,000 0

Forecasting with Asymmetric Loss High Actual Sales High Forecasted Sales Low Forecasted Sales 0 $20,000 Low Actual Sales $10,000 0

Quadratic Loss

Absolute Loss

Asymmetric Loss

Statistical Graphics For Forecasting

1. Why Graphical Analysis is Important Graphics helps us summarize and reveal patterns in data Graphics helps us identify anomalies in data Graphics facilitates and encourages comparison of different pieces of data Graphics enables us to present a huge amount of data in a small space, and it enables us to make huge datasets coherent

2. Simple Graphical Techniques Univariate, multivariate Time series vs. distributional shape Relational graphics

3. Elements of Graphical Style Know your audience, and know your goals. Show the data, and appeal to the viewer. Revise and edit, again and again.

4. Application: Graphing Four Components of Real GNP

Anscombes Quartet (1) x1 y1 10.0 8.04 8.0 6.95 13.0 7.58 9.0 8.81 11.0 8.33 14.0 9.96 6.0 7.24 4.0 4.26 12.0 10.84 7.0 4.82 5.0 5.68 (2) x2 10.0 8.0 13.0 9.0 11.0 14.0 6.0 4.0 12.0 7.0 5.0 y2 9.14 8.14 8.74 8.77 9.26 8.10 6.13 3.10 9.13 7.26 4.74 (3) x3 10.0 8.0 13.0 9.0 11.0 14.0 6.0 4.0 12.0 7.0 5.0 y3 7.46 6.77 12.74 7.11 7.81 8.84 6.08 5.39 8.15 6.42 5.73 (4) x4 8.0 8.0 8.0 8.0 8.0 8.0 8.0 19.0 8.0 8.0 8.0 y4 6.58 5.76 7.71 8.84 8.47 7.04 5.25 12.50 5.56 7.91 6.89

Anscombes Quartet Regressions of yi on xi, i = 1, ..., 4. LS // Dependent Variable is Y1 Variable C X1 R-squared Coefficient 3.00 0.50 0.67 Std. Error T-Statistic 1.12 0.12 2.67 4.24 1.24

S.E. of regression

LS // Dependent Variable is Y2 Variable C X2 R-squared Coefficient 3.00 0.50 0.67 Std. Error T-Statistic 1.12 0.12 2.67 4.24 S.E. of regression 1.24

LS // Dependent Variable is Y3 Variable C X3 R-squared Coefficient 3.00 0.50 0.67 Std. Error T-Statistic 1.12 0.12 2.67 4.24 1.24

S.E. of regression

LS // Dependent Variable is Y4 Variable C X4 R-squared Coefficient 3.00 0.50 0.67 Std. Error T-Statistic 1.12 0.12 2.67 4.24 1.24

S.E. of regression

Anscombes Quartet Bivariate Scatterplots

1-Year Treasury Bond Rate

Change in 1-Year Treasury Bond Rate

Liquor Sales

Histogram and Descriptive Statistics Change in 1-Year Treasury Bond Rate

Scatterplot 1-Year versus 10-Year Treasury Bond Rate

Scatterplot Matrix 1-, 10-, 20-, and 30-Year Treasury Bond Rates

Time Series Plot Aspect Ratio 1:1.6

Time Series Plot Banked to 45 Degrees

Time Series Plot Aspect Ratio 1:1.6

Components of Real GDP (Millions of Current Dollars, Annual)

Modeling and Forecasting Trend 1. Modeling Trend

2. Estimating Trend Models

3. Forecasting Trend

4. Selecting Forecasting Models

Consistency Efficiency

Labor Force Participation Rate Females

Increasing and Decreasing Linear Trends

Labor Force Participation Rate Males

Linear Trend Female Labor Force Participation Rate

Linear Trend Male Labor Force Participation Rate

Volume on the New York Stock Exchange

Various Shapes of Quadratic Trends

Quadratic Trend Volume on the New York Stock Exchange

Log Volume on the New York Stock Exchange

Various Shapes of Exponential Trends

Linear Trend Log Volume on the New York Stock Exchange

Exponential Trend Volume on the New York Stock Exchange

Degrees-of-Freedom Penalties Various Model Selection Criteria

Retail Sales

Retail Sales Linear Trend Regression Dependent Variable is RTRR Sample: 1955:01 1993:12 Included observations: 468 Variable C TIME Coefficient -16391.25 349.7731 Std. Error 1469.177 5.428670 0.899076 0.898859 15866.12 1.17E+11 -5189.529 0.004682 T-Statistic -11.15676 64.43073 Prob. 0.0000 0.0000 65630.56 49889.26 19.34815 19.36587 4151.319 0.000000

R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic)

Retail Sales Linear Trend Residual Plot

Retail Sales Quadratic Trend Regression Dependent Variable is RTRR Sample: 1955:01 1993:12 Included observations: 468 Variable C TIME TIME2 Coefficient 18708.70 -98.31130 0.955404 Std. Error 379.9566 3.741388 0.007725 T-Statistic 49.23905 -26.27669 123.6754 Prob. 0.0000 0.0000 0.0000 65630.56 49889.26 15.82919 15.85578 77848.80 0.000000

R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat

0.997022 0.997010 2728.205 3.46E+09 -4365.093 0.151089

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic)

Retail Sales Quadratic Trend Residual Plot

Retail Sales Log Linear Trend Regression Dependent Variable is LRTRR Sample: 1955:01 1993:12 Included observations: 468 Variable C TIME Coefficient 9.389975 0.005931 Std. Error 0.008508 3.14E-05 T-Statistic 1103.684 188.6541 Prob. 0.0000 0.0000 10.78072 0.807325 -4.770302 -4.752573 35590.36 0.000000

R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat

0.987076 0.987048 0.091879 3.933853 454.1874 0.019949

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic)

Retail Sales Log Linear Trend Residual Plot

Retail Sales Exponential Trend Regression Dependent Variable is RTRR Sample: 1955:01 1993:12 Included observations: 468 Convergence achieved after 1 iterations RTRR=C(1)*EXP(C(2)*TIME) Coefficient C(1) C(2) 11967.80 0.005944 Std. Error 177.9598 3.77E-05 0.988796 0.988772 5286.406 1.30E+10 -4675.175 0.040527 T-Statistic 67.25003 157.7469 Prob. 0.0000 0.0000 65630.56 49889.26 17.15005 17.16778 41126.02 0.000000

R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic)

Retail Sales Exponential Trend Residual Plot

Model Selection Criteria Linear, Quadratic and Exponential Trend Models Linear Trend AIC SIC 19.35 19.37 Quadratic Trend 15.83 15.86 Exponential Trend 17.15 17.17

Retail Sales History, 1990.01 - 1993.12 Quadratic Trend Forecast, 1994.01-1994.12

Retail Sales History, 1990.01 - 1993.12 Quadratic Trend Forecast and Realization, 1994.01-1994.12

Retail Sales History, 1990.01 - 1993.12 Linear Trend Forecast, 1994.01-1994.12

Retail Sales History, 1990.01 - 1993.12 Linear Trend Forecast and Realization, 1994.01-1994.12

Modeling and Forecasting Seasonality 1. The Nature and Sources of Seasonality 2. Modeling Seasonality D1 = (1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, ...) D2 = (0, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, ...) D3 = (0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, ...) D4 = (0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 1, ...)

3. Forecasting Seasonal Series

Gasoline Sales

Liquor Sales

Durable Goods Sales

Housing Starts, 1946.01 - 1994.11

Housing Starts, 1990.01 - 1994.11

Regression Results Seasonal Dummy Variable Model Housing Starts

LS // Dependent Variable is STARTS Sample: 1946:01 1993:12 Included observations: 576 Variable D1 D2 D3 D4 D5 D6 D7 D8 D9 D10 D11 D12 Coefficient 86.50417 89.50417 122.8833 142.1687 147.5000 145.9979 139.1125 138.4167 130.5625 134.0917 111.8333 92.15833 Std. Error 4.029055 4.029055 4.029055 4.029055 4.029055 4.029055 4.029055 4.029055 4.029055 4.029055 4.029055 4.029055 t-Statistic 21.47009 22.21468 30.49929 35.28588 36.60908 36.23627 34.52733 34.35462 32.40524 33.28117 27.75671 22.87344 Prob. 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 123.3944 35.21775 6.678878 6.769630 31.93250 0.000000

R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat

0.383780 0.371762 27.91411 439467.5 -2728.825 0.154140

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic)

Residual Plot

Estimated Seasonal Factors Housing Starts

Housing Starts History, 1990.01-1993.12 Forecast, 1994.01-1994.11

Housing Starts History, 1990.01-1993.12 Forecast and Realization, 1994.01-1994.11

Characterizing Cycles 1. Covariance Stationary Time Series Realization Sample path Covariance stationarity

p() regression of yt on yt-1, ..., yt-

2. White Noise

3. The Lag Operator

4. Wolds Theorem, the General Linear Process, and Rational Distributed Lags Wolds Theorem Let {yt} be any zero-mean covariance-stationary process. Then:

where

and

The General Linear Process

where b0=1 and

Rational Distributed Lags

5. Estimation and Inference for the Mean, Autocorrelation and Partial Autocorrelation Functions

A Rigid Cyclical Pattern

Autocorrelation Function, One-Sided Gradual Damping

Autocorrelation Function, Non-Damping

Autocorrelation Function, Gradual Damped Oscillation

Autocorrelation Function, Sharp Cutoff

Realization of White Noise Process

Population Autocorrelation Function White Noise Process

Population Partial Autocorrelation Function White Noise Process

Canadian Employment Index

Canadian Employment Index Correlogram Sample: 1962:1 1993:4 Included observations: 128 Acorr. 1 2 3 4 5 6 7 8 9 10 11 12 0.949 0.877 0.795 0.707 0.617 0.526 0.438 0.351 0.258 0.163 0.073 -0.005 P. Acorr. 0.949 -0.244 -0.101 -0.070 -0.063 -0.048 -0.033 -0.049 -0.149 -0.070 -0.011 0.016 Std. Error .088 .088 .088 .088 .088 .088 .088 .088 .088 .088 .088 .088 Ljung-Box 118.07 219.66 303.72 370.82 422.27 460.00 486.32 503.41 512.70 516.43 517.20 517.21 p-value 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

Canadian Employment Index Sample Autocorrelation and Partial Autocorrelation Functions, With Plus or Minus Two Standard Error Bands

Modeling Cycles: MA, AR, and ARMA Models The MA(1) Process

If invertible:

The MA(q) Process

where

The AR(1) Process

If covariance stationary:

Moment structure:

Autocovariances and autocorrelations:

For

(Yule-Walker equation) But

. Thus

and

Partial autocorrelations:

The AR(p) Process

The ARMA(1,1) Process

MA representation if invertible:

AR representation of covariance stationary:

The ARMA(p,q) Process

Realizations of Two MA(1) Processes

Population Autocorrelation Function MA(1) Process, =.4

Population Autocorrelation Function MA(1) Process, =.95

Population Partial Autocorrelation Function MA(1) Process, =.4

Population Partial Autocorrelation Function MA(1) Process, =.95

Realizations of Two AR(1) Processes

Population Autocorrelation Function AR(1) Process, =.4

Population Autocorrelation Function AR(1) Process, =.95

Population Partial Autocorrelation Function AR(1) Process, =.4

Population Partial Autocorrelation Function AR(1) Process, =.95

Population Autocorrelation Function AR(2) Process with Complex Roots

Employment MA(4) Model

LS // Dependent Variable is CANEMP Sample: 1962:1 1993:4 Included observations: 128 Convergence achieved after 49 iterations Variable C MA(1) MA(2) MA(3) MA(4) Coefficient 100.5438 1.587641 0.994369 -0.020305 -0.298387 Std. Error 0.843322 0.063908 0.089995 0.046550 0.020489 t-Statistic 119.2234 24.84246 11.04917 -0.436189 -14.56311 Prob. 0.0000 0.0000 0.0000 0.6635 0.0000 101.0176 7.499163 2.203073 2.314481 174.1826 0.000000 -.87

R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat Inverted MA Roots

0.849951 0.845071 2.951747 1071.676 -317.6208 1.246600 .41

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic) -.56+.72i -.56 -.72i

Employment MA(4) Model Residual Plot

Employment MA(4) Model Residual Correlogram

Sample: 1962:1 1993:4 Included observations: 128 Q-statistic probabilities adjusted for 4 ARMA term(s)

Acorr. 1 2 3 4 5 6 7 8 9 10 11 12 0.345 0.660 0.534 0.427 0.347 0.484 0.121 0.348 0.148 0.102 0.081 0.029

P. Acorr. 0.345 0.614 0.426 -0.042 -0.398 0.145 -0.118 -0.048 -0.019 -0.066 -0.098 -0.113

Std. Error .088 .088 .088 .088 .088 .088 .088 .088 .088 .088 .088 .088

Ljung-Box 15.614 73.089 111.01 135.49 151.79 183.70 185.71 202.46 205.50 206.96 207.89 208.01

p-value

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

Employment MA(4) Model Residual Sample Autocorrelation and Partial Autocorrelation Functions, With Plus or Minus Two Standard Error Bands

Employment AR(2) Model

LS // Dependent Variable is CANEMP Sample: 1962:1 1993:4 Included observations: 128 Convergence achieved after 3 iterations Variable C AR(1) AR(2) Coefficient 101.2413 1.438810 -0.476451 Std. Error 3.399620 0.078487 0.077902 t-Statistic 29.78017 18.33188 -6.116042 Prob. 0.0000 0.0000 0.0000 101.0176 7.499163 0.761677 0.828522 1643.837 0.000000

R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat Inverted AR Roots

0.963372 0.962786 1.446663 261.6041 -227.3715 2.067024 .92

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic) .52

Employment AR(2) Model Residual Plot

Employment AR(2) Model Residual Correlogram

Sample: 1962:1 1993:4 Included observations: 128 Q-statistic probabilities adjusted for 2 ARMA term(s)

Acorr. 1 2 3 4 5 6 7 8 9 10 11 12 -0.035 0.044 0.011 0.051 0.002 0.019 -0.024 0.078 0.080 0.050 -0.023 -0.129

P. Acorr. -0.035 0.042 0.014 0.050 0.004 0.015 -0.024 0.072 0.087 0.050 -0.027 -0.148

Std. Error .088 .088 .088 .088 .088 .088 .088 .088 .088 .088 .088 .088

Ljung-Box 0.1606 0.4115 0.4291 0.7786 0.7790 0.8272 0.9036 1.7382 2.6236 2.9727 3.0504 5.4385

p-value

0.512 0.678 0.854 0.935 0.970 0.942 0.918 0.936 0.962 0.860

Employment AR(2) Model Residual Sample Autocorrelation and Partial Autocorrelation Functions, With Plus or Minus Two Standard Error Bands

Employment AIC Values Various ARMA Models MA Order 0 0 1 AR Order 2 3 4 1.01 .762 .77 .79 1 2.86 .83 .77 .761 .79 2 2.32 .79 .78 .77 .77 3 2.47 .80 .80 .78 .79 4 2.20 .81 .80 .79 .80

Employment SIC Values Various ARMA Models MA Order 0 0 1 AR Order 2 3 4 1.05 .83 .86 .90 1 2.91 .90 .86 .87 .92 2 2.38 .88 .89 .90 .93 3 2.56 .91 .92 .94 .97 4 2.31 .94 .96 .96 1.00

Employment ARMA(3,1) Model

LS // Dependent Variable is CANEMP Sample: 1962:1 1993:4 Included observations: 128 Convergence achieved after 17 iterations Variable C AR(1) AR(2) AR(3) MA(1) Coefficient 101.1378 0.500493 0.872194 -0.443355 0.970952 Std. Error 3.538602 0.087503 0.067096 0.080970 0.035015 t-Statistic 28.58130 5.719732 12.99917 -5.475560 27.72924 Prob. 0.0000 0.0000 0.0000 0.0000 0.0000 101.0176 7.499163 0.760668 0.872076 836.2912 0.000000

R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat Inverted AR Roots Inverted MA Roots

0.964535 0.963381 1.435043 253.2997 -225.3069 2.057302 .93 -.97

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic) .51 -.94

Employment ARMA(3,1) Model Residual Plot

Employment ARMA(3,1) Model Residual Correlogram

Sample: 1962:1 1993:4 Included observations: 128 Q-statistic probabilities adjusted for 4 ARMA term(s)

Acorr. 1 2 3 4 5 6 7 8 9 10 11 12 -0.032 0.041 0.014 0.048 0.006 0.013 -0.017 0.064 0.092 0.039 -0.016 -0.137

P. Acorr. -0.032 0.040 0.017 0.047 0.007 0.009 -0.019 0.060 0.097 0.040 -0.022 -0.153

Std. Error .09 .09 .09 .09 .09 .09 .09 .09 .09 .09 .09 .09

Ljung-Box 0.1376 0.3643 0.3904 0.6970 0.7013 0.7246 0.7650 1.3384 2.5182 2.7276 2.7659 5.4415

p-value

0.402 0.696 0.858 0.855 0.774 0.842 0.906 0.710

Employment ARMA(3,1) Model Residual Sample Autocorrelation and Partial Autocorrelation Functions, With Plus or Minus Two Standard Error Bands

Forecasting Cycles

Optimal Point Forecasts for Infinite-Order Moving Averages

where

, and

Interval and Density Forecasts

95% h-step-ahead interval forecast: . h-step-ahead density forecast:

Making the Forecasts Operational The Chain Rule of Forecasting

Employment History and Forecast MA(4) Model

Employment History and Long-Horizon Forecast MA(4) Model

Employment History, Forecast and Realization MA(4) Model

Employment History and Forecast AR(2) Model

Employment History and Long-Horizon Forecast AR(2) Model

Employment History and Very Long-Horizon Forecast AR(2) Model

Employment History, Forecast and Realization AR(2) Model

Putting it all Together: A Forecasting Model with Trend, Seasonal and Cyclical Components The full model:

Point Forecasting

Interval Forecasting

e.g.: (95% interval)

Density Forecasting

Assessing the Stability of Forecasting Models: Recursive Parameter Estimation and Recursive Residuals At each t, t = k, ... ,T-1, compute: Recursive parameter est. and forecast: Recursive residual: If all is well: Sequence of 1-step forecast tests: Standardized recursive residuals: If all is well:

Liquor Sales, 1968.01 - 1993.12

Log Liquor Sales, 1968.01 - 1993.12

Log Liquor Sales Quadratic Trend Regression

LS // Dependent Variable is LSALES Sample: 1968:01 1993:12 Included observations: 312 Variable C TIME TIME2 Coefficient 6.237356 0.007690 -1.14E-05 Std. Error 0.024496 0.000336 9.74E-07 t-Statistic 254.6267 22.91552 -11.72695 Prob. 0.0000 0.0000 0.0000 7.112383 0.379308 -4.152073 -4.116083 1281.296 0.000000

R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat

0.892394 0.891698 0.124828 4.814823 208.0146 1.752858

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic)

Log Liquor Sales Quadratic Trend Regression Residual Plot

Log Liquor Sales Quadratic Trend Regression Residual Correlogram Acorr. 0.117 -0.149 -0.106 -0.014 0.142 0.041 0.134 -0.029 -0.136 -0.205 0.056 0.888 0.055 -0.187 -0.159 -0.059 0.091 -0.010 0.086 -0.066 -0.170 -0.231 0.028 0.811 0.013 -0.221 -0.196 -0.092 0.045 -0.043 0.057 -0.095 -0.195 -0.240 0.006 0.765 P. Acorr. 0.117 -0.165 -0.069 -0.017 0.125 -0.004 0.175 -0.046 -0.080 -0.206 0.080 0.879 -0.507 -0.159 -0.144 -0.002 -0.118 -0.055 -0.032 0.028 0.044 0.180 0.016 -0.014 -0.128 -0.136 -0.017 -0.079 -0.094 0.045 0.041 -0.002 0.026 0.088 -0.089 0.076 Std. Error .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 Ljung-Box 4.3158 11.365 14.943 15.007 21.449 21.979 27.708 27.975 33.944 47.611 48.632 306.26 307.25 318.79 327.17 328.32 331.05 331.08 333.57 335.03 344.71 362.74 363.00 586.50 586.56 603.26 616.51 619.42 620.13 620.77 621.89 625.07 638.38 658.74 658.75 866.34 p-value 0.038 0.003 0.002 0.005 0.001 0.001 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36

Log Liquor Sales Quadratic Trend Regression Residual Sample Autocorrelation and Partial Autocorrelation Functions

Log Liquor Sales Quadratic Trend Regression with Seasonal Dummies

LS // Dependent Variable is LSALES Sample: 1968:01 1993:12 Included observations: 312 Variable TIME TIME2 D1 D2 D3 D4 D5 D6 D7 D8 D9 D10 D11 D12 Coefficient 0.007656 -1.14E-05 6.147456 6.088653 6.174127 6.175220 6.246086 6.250387 6.295979 6.268043 6.203832 6.229197 6.259770 6.580068 Std. Error 0.000123 3.56E-07 0.012340 0.012353 0.012366 0.012378 0.012390 0.012401 0.012412 0.012423 0.012433 0.012444 0.012453 0.012463 t-Statistic 62.35882 -32.06823 498.1699 492.8890 499.3008 498.8970 504.1398 504.0194 507.2402 504.5509 498.9630 500.5968 502.6602 527.9819 Prob. 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 7.112383 0.379308 -6.128963 -5.961008 1627.567 0.000000

R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat

0.986111 0.985505 0.045666 0.621448 527.4094 0.586187

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic)

Log Liquor Sales Quadratic Trend Regression with Seasonal Dummies Residual Plot

Log Liquor Sales Quadratic Trend Regression with Seasonal Dummies Residual Correlogram Acorr. 0.700 0.686 0.725 0.569 0.569 0.577 0.460 0.480 0.466 0.327 0.364 0.355 0.225 0.291 0.211 0.138 0.195 0.114 0.055 0.134 0.062 -0.006 0.084 -0.039 -0.063 -0.016 -0.143 -0.135 -0.124 -0.189 -0.178 -0.139 -0.226 -0.155 -0.142 -0.242 P. Acorr. 0.700 0.383 0.369 -0.141 0.017 0.093 -0.078 0.043 0.030 -0.188 0.019 0.089 -0.119 0.065 -0.119 -0.031 0.053 -0.027 -0.063 0.089 0.018 -0.115 0.086 -0.124 -0.055 -0.022 -0.075 -0.047 -0.048 0.086 -0.017 0.073 -0.049 0.097 0.008 -0.074 Std. Error .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 Ljung-Box 154.34 302.86 469.36 572.36 675.58 782.19 850.06 924.38 994.46 1029.1 1072.1 1113.3 1129.9 1157.8 1172.4 1178.7 1191.4 1195.7 1196.7 1202.7 1204.0 1204.0 1206.4 1206.9 1208.3 1208.4 1215.4 1221.7 1227.0 1239.5 1250.5 1257.3 1275.2 1283.7 1290.8 1311.6 p-value 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36

Log Liquor Sales Quadratic Trend Regression with Seasonal Dummies Residual Sample Autocorrelation and Partial Autocorrelation Functions

Log Liquor Sales Quadratic Trend Regression with Seasonal Dummies and AR(3) Disturbances

LS // Dependent Variable is LSALES Sample: 1968:01 1993:12 Included observations: 312 Convergence achieved after 4 iterations Variable TIME TIME2 D1 D2 D3 D4 D5 D6 D7 D8 D9 D10 D11 D12 AR(1) AR(2) AR(3) Coefficient 0.008606 -1.41E-05 6.073054 6.013822 6.099208 6.101522 6.172528 6.177129 6.223323 6.195681 6.131818 6.157592 6.188480 6.509106 0.268805 0.239688 0.395880 Std. Error 0.000981 2.53E-06 0.083922 0.083942 0.083947 0.083934 0.083946 0.083947 0.083939 0.083943 0.083940 0.083934 0.083932 0.083928 0.052909 0.053697 0.053109 t-Statistic 8.768212 -5.556103 72.36584 71.64254 72.65524 72.69393 73.52962 73.58364 74.14071 73.80857 73.04993 73.36197 73.73176 77.55624 5.080488 4.463723 7.454150 Prob. 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 7.112383 0.379308 -7.145319 -6.941373 3720.875 0.000000

R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat Inverted AR Roots

0.995069 0.994802 0.027347 0.220625 688.9610 1.886119 .95

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic) -.34+.55i -.34 -.55i

Log Liquor Sales Quadratic Trend Regression with Seasonal Dummies and AR(3) Disturbances Residual Plot

Log Liquor Sales Quadratic Trend Regression with Seasonal Dummies and AR(3) Disturbances Residual Correlogram Acorr. 0.056 0.037 0.024 -0.084 -0.007 0.065 -0.041 0.069 0.080 -0.163 -0.009 0.145 -0.074 0.149 -0.039 -0.089 0.058 -0.062 -0.110 0.100 0.039 -0.122 0.146 -0.072 0.006 0.148 -0.109 -0.029 -0.046 -0.084 -0.095 0.051 -0.114 0.024 0.043 -0.229 P. Acorr. 0.056 0.034 0.020 -0.088 0.001 0.072 -0.044 0.063 0.074 -0.169 -0.005 0.175 -0.078 0.113 -0.060 -0.058 0.048 -0.050 -0.074 0.056 0.042 -0.114 0.130 -0.040 0.017 0.082 -0.067 -0.045 -0.100 0.020 -0.101 0.012 -0.061 0.002 -0.010 -0.140 Std. Error .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 .056 Ljung-Box 0.9779 1.4194 1.6032 3.8256 3.8415 5.1985 5.7288 7.2828 9.3527 18.019 18.045 24.938 26.750 34.034 34.532 37.126 38.262 39.556 43.604 46.935 47.440 52.501 59.729 61.487 61.500 69.024 73.145 73.436 74.153 76.620 79.793 80.710 85.266 85.468 86.116 104.75 p-value 0.323 0.492 0.659 0.430 0.572 0.519 0.572 0.506 0.405 0.055 0.081 0.015 0.013 0.002 0.003 0.002 0.002 0.002 0.001 0.001 0.001 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36

Log Liquor Sales Quadratic Trend Regression with Seasonal Dummies and AR(3) Disturbances Residual Sample Autocorrelation and Partial Autocorrelation Functions

Log Liquor Sales Quadratic Trend Regression with Seasonal Dummies and AR(3) Disturbances Residual Histogram and Normality Test

Log Liquor Sales History and 12-Month-Ahead Forecast

Log Liquor Sales History, 12-Month-Ahead Forecast, and Realization

Log Liquor Sales History and 60-Month-Ahead Forecast

Log Liquor Sales Long History and 60-Month-Ahead Forecast

Liquor Sales Long History and 60-Month-Ahead Forecast

Recursive Analysis Constant Parameter Model

Recursive Analysis Breaking Parameter Model

Log Liquor Sales Quadratic Trend Regression with Seasonal Dummies and AR(3) Disturbances Recursive Residuals and Two Standard Error Bands

Log Liquor Sales Quadratic Trend Regression with Seasonal Dummies and AR(3) Disturbances Recursive Parameter Estimates

Log Liquor Sales Quadratic Trend Regression with Seasonal Dummies and AR(3) Disturbances CUSUM Analysis

Forecasting with Regression Models Conditional Forecasting Models and Scenario Analysis

Density forecast: C Scenario analysis, contingency analysis C No forecasting the RHS variables problem

Unconditional Forecasting Models

C Forecasting the RHS variables problem C Could fit a model to x (e.g., an autoregressive model) C Preferably, regress y on C No problem in trend and seasonal models

Distributed Lags Start with unconditional forecasting model:

Generalize to

C distributed lag model C lag weights C lag distribution

Polynomial distributed lags Solve the problem:

subject to

C Lag weights constrained to lie on low-order polynomial C Additional constraints can be imposed, such as C Smooth lag distribution C Parsimonious

Rational distributed lags

Equivalently,

C Lags of x and y included C Important to allow for lags of y, one way or another

Another way: distributed lag regression with lagged dependent variables

Another way: distributed lag regression with ARMA disturbances

Another way: the transfer function model and various special cases
Name Model Restrictions

Transfer Function Standard Distributed Lag Rational Distributed Lag Univariate AR Univariate MA Univariate ARMA

None B(L)=C(L)=D(L)=1 C(L)=D(L)=1 A(L)=0, C(L)=1 A(L)=0, D(L)=1 A(L)=0

Distributed Lag with Lagged Dep. Variables

, or C(L)=1, D(L)=B(L)

Distributed Lag with ARMA Disturbances

B(L)=1

Distributed Lag with AR Disturbances

B(L)=C(L)=1

Vector Autoregressions e.g., bivariate VAR(1)

C Estimation by OLS

C Order selection by information criteria C Impulse-response functions, variance decompositions, predictive causality C Forecasts via Wolds chain rule

Point and Interval Forecasts Top Panel Interval Forecasts Dont Acknowledge Parameter Uncertainty Bottom Panel Interval Forecasts Do Acknowledge Parameter Uncertainty

U.S. Housing Starts and Completions, 1968.01 - 1996.06

Notes to figure: The left scale is starts, and the right scale is completions.

Starts Correlogram Sample: 1968:01 1991:12 Included observations: 288

Acorr. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 0.937 0.907 0.877 0.838 0.795 0.751 0.704 0.650 0.604 0.544 0.496 0.446 0.405 0.346 0.292 0.233 0.175 0.122 0.070 0.019 -0.034 -0.074 -0.123 -0.167

P. Acorr. 0.937 0.244 0.054 -0.077 -0.096 -0.058 -0.067 -0.098 0.004 -0.129 0.029 -0.008 0.076 -0.144 -0.079 -0.111 -0.050 -0.018 0.002 -0.025 -0.032 0.036 -0.028 -0.048

Std. Error 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059

Ljung-Box 255.24 495.53 720.95 927.39 1113.7 1280.9 1428.2 1554.4 1663.8 1752.6 1826.7 1886.8 1936.8 1973.3 1999.4 2016.1 2025.6 2030.2 2031.7 2031.8 2032.2 2033.9 2038.7 2047.4

p-value 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

Starts Sample Autocorrelations and Partial Autocorrelations

Completions Correlogram

Sample: 1968:01 1991:12 Included observations: 288

Acorr. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 0.939 0.920 0.896 0.874 0.834 0.802 0.761 0.721 0.677 0.633 0.583 0.533 0.483 0.434 0.390 0.337 0.290 0.234 0.181 0.128 0.068 0.020 -0.038 -0.087

P. Acorr. 0.939 0.328 0.066 0.023 -0.165 -0.067 -0.100 -0.070 -0.055 -0.047 -0.080 -0.073 -0.038 -0.020 0.041 -0.057 -0.008 -0.109 -0.082 -0.047 -0.133 0.037 -0.092 -0.003

Std. Error 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059

Ljung-Box 256.61 504.05 739.19 963.73 1168.9 1359.2 1531.2 1686.1 1823.2 1943.7 2046.3 2132.2 2203.2 2260.6 2307.0 2341.9 2367.9 2384.8 2395.0 2400.1 2401.6 2401.7 2402.2 2404.6

p-value 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

Completions Sample Autocorrelations and Partial Autocorrelations

Starts and Completions Sample Cross Correlations

Notes to figure: We graph the sample correlation between completions at time t and starts at time t-i, i = 1, 2, ..., 24.

VAR Order Selection with AIC and SIC

VAR Starts Equation LS // Dependent Variable is STARTS Sample(adjusted): 1968:05 1991:12 Included observations: 284 after adjusting endpoints Variable Coefficient Std. Error 0.044235 0.061242 0.072724 0.072655 0.066032 0.102712 0.103847 0.100946 0.094569 t-Statistic 3.320264 10.77587 3.157587 1.966281 0.118217 0.307759 -1.163069 -0.204078 -0.289779 Prob. 0.0010 0.0000 0.0018 0.0503 0.9060 0.7585 0.2458 0.8384 0.7722 1.574771 0.382362 -4.122118 -4.006482 294.7796 0.000000

C 0.146871 STARTS(-1) 0.659939 STARTS(-2) 0.229632 STARTS(-3) 0.142859 STARTS(-4) 0.007806 COMPS(-1) 0.031611 COMPS(-2) -0.120781 COMPS(-3) -0.020601 COMPS(-4) -0.027404 R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat

0.895566 0.892528 0.125350 4.320952 191.3622 1.991908

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic)

VAR Starts Equation Residual Plot

VAR Starts Equation Residual Correlogram

Sample: 1968:01 1991:12 Included observations: 284

Acorr. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 0.001 0.003 0.006 0.023 -0.013 0.022 0.038 -0.048 0.056 -0.114 -0.038 -0.030 0.192 0.014 0.063 -0.006 -0.039 -0.029 -0.010 0.010 -0.057 0.045 -0.038 -0.149

P. Acorr. 0.001 0.003 0.006 0.023 -0.013 0.021 0.038 -0.048 0.056 -0.116 -0.038 -0.028 0.193 0.021 0.067 -0.015 -0.035 -0.043 -0.009 -0.014 -0.047 0.018 0.011 -0.141

Std. Error 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059

Ljung-Box 0.0004 0.0029 0.0119 0.1650 0.2108 0.3463 0.7646 1.4362 2.3528 6.1868 6.6096 6.8763 17.947 18.010 19.199 19.208 19.664 19.927 19.959 19.993 21.003 21.644 22.088 29.064

p-value 0.985 0.999 1.000 0.997 0.999 0.999 0.998 0.994 0.985 0.799 0.830 0.866 0.160 0.206 0.205 0.258 0.292 0.337 0.397 0.458 0.459 0.481 0.515 0.218

VAR Starts Equation Residual Sample Autocorrelations and Partial Autocorrelations

VAR Completions Equation

LS // Dependent Variable is COMPS Sample(adjusted): 1968:05 1991:12 Included observations: 284 after adjusting endpoints Variable C STARTS(-1) STARTS(-2) STARTS(-3) STARTS(-4) COMPS(-1) COMPS(-2) COMPS(-3) COMPS(-4) Coefficient 0.045347 0.074724 0.040047 0.047145 0.082331 0.236774 0.206172 0.120998 0.156729 Std. Error 0.025794 0.035711 0.042406 0.042366 0.038504 0.059893 0.060554 0.058863 0.055144 t-Statistic 1.758045 2.092461 0.944377 1.112805 2.138238 3.953313 3.404742 2.055593 2.842160 Prob. 0.0799 0.0373 0.3458 0.2668 0.0334 0.0001 0.0008 0.0408 0.0048 1.547958 0.286689 -5.200872 -5.085236 509.8375 0.000000

R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat

0.936835 0.934998 0.073093 1.469205 344.5453 2.013370

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic)

VAR Completions Equation Residual Plot

VAR Completions Equation Residual Correlogram

Sample: 1968:01 1991:12 Included observations: 284

Acorr. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 -0.009 -0.035 -0.037 -0.088 -0.105 0.012 -0.024 0.041 0.048 0.045 -0.009 -0.050 -0.038 -0.055 0.027 -0.005 0.096 0.011 0.041 0.046 -0.096 0.039 -0.113 -0.136

P. Acorr. -0.009 -0.035 -0.037 -0.090 -0.111 0.000 -0.041 0.024 0.029 0.037 -0.005 -0.046 -0.024 -0.049 0.028 -0.020 0.082 -0.002 0.040 0.061 -0.079 0.077 -0.114 -0.125

Std. Error 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059 0.059

Ljung-Box 0.0238 0.3744 0.7640 3.0059 6.1873 6.2291 6.4047 6.9026 7.5927 8.1918 8.2160 8.9767 9.4057 10.318 10.545 10.553 13.369 13.405 13.929 14.569 17.402 17.875 21.824 27.622

p-value 0.877 0.829 0.858 0.557 0.288 0.398 0.493 0.547 0.576 0.610 0.694 0.705 0.742 0.739 0.784 0.836 0.711 0.767 0.788 0.801 0.686 0.713 0.531 0.276

VAR Completions Equation Residual Sample Autocorrelations and Partial Autocorrelations

Housing Starts and Completions Causality Tests

Sample: 1968:01 1991:12 Lags: 4 Obs: 284 Null Hypothesis: STARTS does not Cause COMPS COMPS does not Cause STARTS F-Statistic 26.2658 2.23876 Probability 0.00000 0.06511

Housing Starts and Completions VAR Impulse-Response Functions

Housing Starts and Completions VAR Variance Decompositions

Starts History, 1968.01-1991.12 Forecast, 1992.01-1996.06

Starts History, 1968.01-1991.12 Forecast and Realization, 1992.01-1996.06

Completions History, 1968.01-1991.12 Forecast, 1992.01-1996.06

Completions History, 1968.01-1991.12 Forecast and Realization, 1992.01-1996.06

Evaluating and Combining Forecasts Evaluating a single forecast Process:

h-step-ahead linear least-squares forecast:

Corresponding h-step-ahead forecast error:

with variance

So, four key properties of optimal forecasts: a. Optimal forecasts are unbiased b. Optimal forecasts have 1-step-ahead errors that are white noise c. Optimal forecasts have h-step-ahead errors that are at most MA(h-1) d. Optimal forecasts have h-step-ahead errors with variances that are non-decreasing in h and that converge to the unconditional variance of the process C All are easily checked. How?

Assessing optimality with respect to an information set Unforecastability principle: The errors from good forecasts are not be forecastable! Regression:

C Test whether Important case:

are 0

C Test whether (0, 1) = (0, 0) Equivalently,

C Test whether (0, 1) = (0, 1)

Evaluating multiple forecasts: comparing forecast accuracy Forecast errors, Forecast percent errors,

Forecast encompassing

C If (a, b) = (1,0), model a forecast-encompasses model b C If (a, b)= (0,1), model b forecast-encompasses model a C Otherwise, neither model encompasses the other

Alternative approach:

C Useful in I(1) situations

Variance-covariance forecast combination Composite formed from two unbiased forecasts:

Regression-based forecast combination

C Equivalent to variance-covariance combination if weights sum to unity and intercept is excluded C Easy extension to include more than two forecasts C Time-varying combining weights C Dynamic combining regressions C Shrinkage of combining weights toward equality C Nonlinear combining regressions

Shipping Volume Quantitative Forecast and Realization

Shipping Volume Judgmental Forecast and Realization

Quantitative Forecast Error

Judgmental Forecast Error

Histogram and Related Statistics Quantitative Forecast Error

Histogram and Related Statistics Judgmental Forecast Error

Correlogram, Quantitative Forecast Error

Sample: 1/01/1988 7/18/1997 Included observations: 499 Acorr. 0.518 0.010 -0.044 -0.039 0.025 0.057 P. Acorr. 0.518 -0.353 0.205 -0.172 0.195 -0.117 Std. Error .045 .045 .045 .045 .045 .045 Ljung-Box 134.62 134.67 135.65 136.40 136.73 138.36 p-value 0.000 0.000 0.000 0.000 0.000 0.000

1 2 3 4 5 6

Sample Autocorrelations and Partial Autocorrelations Quantitative Forecast Error

Correlogram, Judgmental Forecast Error

Sample: 1/01/1988 7/18/1997 Included observations: 499 Acorr. 1 2 3 4 5 6 0.495 -0.027 -0.045 -0.056 -0.033 0.087 P. Acorr. 0.495 -0.360 0.229 -0.238 0.191 -0.011 Std. Error .045 .045 .045 .045 .045 .045 Ljung-Box 122.90 123.26 124.30 125.87 126.41 130.22 p-value 0.000 0.000 0.000 0.000 0.000 0.000

Sample Autocorrelations and Partial Autocorrelations Judgmental Forecast Error

Quantitative Forecast Error Regression on Intercept, MA(1) Disturbances

LS // Dependent Variable is EQ Sample: 1/01/1988 7/18/1997 Included observations: 499 Convergence achieved after 6 iterations Variable C MA(1) Coefficient -0.024770 0.935393 Std. Error 0.079851 0.015850 t-Statistic -0.310200 59.01554 Prob. 0.7565 0.0000 -0.026572 1.262817 -0.159064 -0.142180 437.8201 0.000000

R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat Inverted MA Roots

0.468347 0.467277 0.921703 422.2198 -666.3639 1.988237 -.94

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic)

Judgmental Forecast Error Regression on Intercept, MA(1) Disturbances

LS // Dependent Variable is EJ Sample: 1/01/1988 7/18/1997 Included observations: 499 Convergence achieved after 7 iterations Variable C MA(1) Coefficient 1.026372 0.961524 Std. Error 0.067191 0.012470 t-Statistic 15.27535 77.10450 Prob. 0.0000 0.0000 1.023744 1.063681 -0.531226 -0.514342 465.2721 0.000000

R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat Inverted MA Roots

0.483514 0.482475 0.765204 291.0118 -573.5094 1.968750 -.96

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic)

Mincer-Zarnowitz Regression Quantitative Forecast Error

LS // Dependent Variable is VOL Sample: 1/01/1988 7/18/1997 Included observations: 499 Convergence achieved after 10 iterations Variable C VOLQ MA(1) Coefficient 2.958191 0.849559 0.912559 Std. Error 0.341841 0.016839 0.018638 t-Statistic 8.653696 50.45317 48.96181 Prob. 0.0000 0.0000 0.0000 19.80609 3.403283 -0.304685 -0.279358 3686.790 0.000000

R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat Inverted MA Roots

0.936972 0.936718 0.856125 363.5429 -629.0315 1.815577 -.91

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic)

Wald Test: Null Hypothesis: C(1)=0 F-statistic 39.96862 Chi-square 79.93723

C(2)=1 Probability Probability

0.000000 0.000000

Mincer-Zarnowitz Regression Judgmental Forecast Error

LS // Dependent Variable is VOL Sample: 1/01/1988 7/18/1997 Included observations: 499 Convergence achieved after 11 iterations Variable C VOLJ MA(1) Coefficient 2.592648 0.916576 0.949690 Std. Error 0.271740 0.014058 0.014621 t-Statistic 9.540928 65.20021 64.95242 Prob. 0.0000 0.0000 0.0000 19.80609 3.403283 -0.595907 -0.570581 5016.993 0.000000

R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat Inverted MA Roots

0.952896 0.952706 0.740114 271.6936 -556.3715 1.917179 -.95

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic)

Wald Test: Null Hypothesis: C(1)=0 F-statistic 143.8323 Chi-square 287.6647

C(2)=1 Probability Probability

0.000000 0.000000

Histogram and Related Statistics Squared Quantitative Forecast Error

Histogram and Related Statistics Squared Judgmental Forecast Error

Loss Differential

Histogram and Related Statistics Loss Differential

Loss Differential Correlogram

Sample: 1/01/1988 7/18/1997 Included observations: 499 Acorr. 1 2 3 4 5 6 0.357 -0.069 -0.050 -0.044 -0.078 0.017 P. Acorr. 0.357 -0.226 0.074 -0.080 -0.043 0.070 Std. Error .045 .045 .045 .045 .045 .045 Ljung-Box 64.113 66.519 67.761 68.746 71.840 71.989 p-value 0.000 0.000 0.000 0.000 0.000 0.000

Sample Autocorrelations and Partial Autocorrelations Loss Differential

Loss Differential Regression on Intercept with MA(1) Disturbances

LS // Dependent Variable is DD Sample: 1/01/1988 7/18/1997 Included observations: 499 Convergence achieved after 4 iterations Variable C MA(1) Coefficient -0.585333 0.472901 Std. Error 0.204737 0.039526 t-Statistic -2.858945 11.96433 Prob. 0.0044 0.0000 -0.584984 3.416190 2.270994 2.287878 105.2414 0.000000

R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat Inverted MA Roots

0.174750 0.173089 3.106500 4796.222 -1272.663 2.023606 -.47

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic)

Shipping Volume Combining Regression

LS // Dependent Variable is VOL Sample: 1/01/1988 7/18/1997 Included observations: 499 Convergence achieved after 11 iterations Variable C VOLQ VOLJ MA(1) Coefficient 2.181977 0.291577 0.630551 0.951107 Std. Error 0.259774 0.038346 0.039935 0.014174 t-Statistic 8.399524 7.603919 15.78944 67.10327 Prob. 0.0000 0.0000 0.0000 0.0000 19.80609 3.403283 -0.702371 -0.668603 3747.077 0.000000

R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat Inverted MA Roots

0.957823 0.957567 0.701049 243.2776 -528.8088 1.925091 -.95

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic)

Unit Roots, Stochastic Trends, ARIMA Forecasting Models, and Smoothing 1. Stochastic Trends and Forecasting

I(0) vs I(1) processes

Random walk:

Random walk with drift:

Stochastic trend vs deterministic trend

Properties of random walks

With time 0 value

Random walk with drift

Assuming time 0 value

ARIMA(p,1,q) model

or

where

and all the roots of both lag operator polynomials are outside the unit circle.

ARIMA(p,d,q) model

or

where

and all the roots of both lag operator polynomials are outside the unit circle.

Properties of ARIMA(p,1,q) processes C Appropriately made stationary by differencing C Shocks have permanent effects -- Forecasts dont revert to a mean C Variance grows without bound as time progresses -- Interval forecasts widen without bound as horizon grows

Random walk example Point forecast Recall that for the AR(1) process,

the optimal forecast is

Thus in the random walk case,

Interval and density forecasts Recall error associated with optimal AR(1) forecast:

with variance

Thus in the random walk case,

h-step-ahead 95% interval forecast: h-step-ahead density forecast:

Effects of Unit Roots C Sample autocorrelation function fails to damp C Sample partial autocorrelation function near 1 for and then damps quickly C Properties of estimators change e.g., least-squares autoregression with unit roots True process: ,

Estimated model:

Superconsistency: Bias:

stabilizes as sample size grows

-- Ofsetting effects of bias and superconsistency

Unit Root Tests

Dickey-Fuller Trick regression:

distribution

Allowing for nonzero mean under the alternative Basic model:

which we rewrite as

where

C vanishes when

(null)

C is nevertheless present under the alternative, so we include an intercept in the regression Dickey-Fuller distribution

Allowing for deterministic linear trend under the alternative Basic model:

or

where

and

C Under the null hypothesis we have a random walk with drift,

C Under the deterministic-trend alternative hypothesis both the intercept and the trend enter and so are included in the regression.

Allowing for higher-order autoregressive dynamics AR(p) process:

Rewrite:

where

, and

Unit root:

(AR(p-1) in first differences)

distribution holds asymptotically

Allowing for a nonzero mean in the AR(p) case

or

where

, and the other parameters are as above.

In the unit root case, the intercept vanishes, because distribution holds asymptotically.

Allowing for trend under the alternative

or

where

and

In the unit root case,

and

distribution holds asymptotically.

General ARMA representations: augmented Dickey-Fuller tests

C k-1 augmentation lags have been included C , , and hold asymptotically under the null

Simple moving average smoothing Original data: Smoothed data: Two-sided moving average is One-sided moving average is One-sided weighted moving average is C Must choose smoothing parameter, m

Exponential Smoothing Local level model:

C Exponential smoothing can construct the optimal estimate of -- and hence the optimal forecast of any future value of y -on the basis of current and past y C What if the model is misspecified?

Exponential smoothing algorithm Observed series, Smoothed series, Forecasts, (1) Initialize at t=1: (2) Update: (3) Forecast: C Smoothing parameter (estimate of the local level)

Demonstration that the weights are exponential Start:

Substitute backward for :

where

C Exponential weighting, as claimed C Convenient recursive structure

Holt-Winters Smoothing

C Local level and slope model C Holt-Winters smoothing can construct optimal estimates of and -- and hence the optimal forecast of any future value of y by extrapolating the trend -- on the basis of current and past y

Holt-Winters smoothing algorithm (1) Initialize at t=2:

(2) Update:

t = 3, 4, ..., T. (3) Forecast: C C is the estimated level at time t is the estimated slope at time t

Random Walk Level and Change

Random Walk With Drift Level and Change

U.S. Per Capita GNP History and Two Forecasts

U.S. Per Capita GNP History, Two Forecasts, and Realization

Random Walk, Levels Sample Autocorrelation Function (Top Panel) Sample Partial Autocorrelation Function (Bottom Panel)

Random Walk, First Differences Sample Autocorrelation Function (Top Panel) Sample Partial Autocorrelation Function (Bottom Panel)

Log Yen / Dollar Exchange Rate (Top Panel) Change in Log Yen / Dollar Exchange Rate (Bottom Panel)

Log Yen / Dollar Exchange Rate Sample Autocorrelations (Top Panel) Sample Partial Autocorrelations (Bottom Panel)

Log Yen / Dollar Exchange Rate, First Differences Sample Autocorrelations (Top Panel) Sample Partial Autocorrelations (Bottom Panel)

Log Yen / Dollar Rate, Levels AIC Values Various ARMA Models MA Order 0 0 AR Order 1 2 3 -7.171 -7.319 -7.322 1 -5.171 -7.300 -7.314 -7.323 2 -5.953 -7.293 -7.320 -7.316 3 -6.428 -7.287 -7.317 -7.308

Log Yen / Dollar Rate, Levels SIC Values Various ARMA Models MA Order 0 0 AR Order 1 2 3 -7.131 -7.265 -7.253 1 -5.130 -7.211 -7.246 -7.241 2 -5.899 -7.225 -7.238 -7.220 3 -6.360 -7.205 -7.221 -7.199

Log Yen / Dollar Exchange Rate Best-Fitting Deterministic-Trend Model LS // Dependent Variable is LYEN Sample(adjusted): 1973:03 1994:12 Included observations: 262 after adjusting endpoints Convergence achieved after 3 iterations Variable C TIME AR(1) AR(2) Coefficient 5.904705 -0.004732 1.305829 -0.334210 Std. Error 0.136665 0.000781 0.057587 0.057656 t-Statistic 43.20570 -6.057722 22.67561 -5.796676 Prob. 0.0000 0.0000 0.0000 0.0000 5.253984 0.341563 -7.319015 -7.264536 15461.07 0.000000

R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat Inverted AR Roots

0.994468 0.994404 0.025551 0.168435 591.0291 1.964687 .96

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic) .35

Log Yen / Dollar Exchange Rate Best-Fitting Deterministic-Trend Model Residual Plot

Log Yen / Dollar Rate History and Forecast AR(2) in Levels with Linear Trend

Log Yen / Dollar Rate History and Long-Horizon Forecast AR(2) in Levels with Linear Trend

Log Yen / Dollar Rate History, Forecast and Realization AR(2) in Levels with Linear Trend

Log Yen / Dollar Exchange Rate Augmented Dickey-Fuller Unit Root Test

Augmented Dickey-Fuller Test Statistic

-2.498863

1% Critical Value 5% Critical Value 10% Critical Value

-3.9966 -3.4284 -3.1373

Augmented Dickey-Fuller Test Equation LS // Dependent Variable is D(LYEN) Sample(adjusted): 1973:05 1994:12 Included observations: 260 after adjusting endpoints Variable Coefficient Std. Error t-Statistic Prob. 0.0131 0.0000 0.0795 0.0535 0.0132 0.0088 -0.003749 0.027103 -7.327517 -7.245348 8.432417 0.000000

LYEN(-1) -0.029423 D(LYEN(-1)) 0.362319 D(LYEN(-2)) -0.114269 D(LYEN(-3)) 0.118386 C 0.170875 @TREND(1973:01) -0.000139 R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat 0.142362 0.125479 0.025345 0.163166 589.6532 2.010829

0.011775 0.061785 0.064897 0.061020 0.068474 5.27E-05

-2.498863 5.864226 -1.760781 1.940116 2.495486 -2.639758

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic)

Log Yen / Dollar Rate, Changes AIC Values Various ARMA Models MA Order 0 0 AR Order 1 2 3 -7.308 -7.312 -7.316 1 -7.298 -7.307 -7.314 -7.309 2 -7.290 -7.307 -7.307 -7.340 3 -7.283 -7.302 -7.299 -7.336

Log Yen / Dollar Rate, Changes SIC Values Various ARMA Models MA Order 0 0 AR Order 1 2 3 -7.281 -7.271 -7.261 1 -7.270 -7.266 -7.259 -7.241 2 -7.249 -7.252 -7.238 -7.258 3 -7.228 -7.234 -7.217 -7.240

Log Yen / Dollar Exchange Rate Best-Fitting Stochastic-Trend Model

LS // Dependent Variable is DLYEN Sample(adjusted): 1973:03 1994:12 Included observations: 262 after adjusting endpoints Convergence achieved after 3 iterations Variable C AR(1) Coefficient -0.003697 0.321870 Std. Error 0.002350 0.057767 t-Statistic -1.573440 5.571863 Prob. 0.1168 0.0000 -0.003888 0.027227 -7.308418 -7.281179 31.04566 0.000000

R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat Inverted AR Roots

0.106669 0.103233 0.025784 0.172848 587.6409 1.948933 .32

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic)

Log Yen / Dollar Exchange Rate Best-Fitting Stochastic-Trend Model Residual Plot

Log Yen / Dollar Rate History and Forecast AR(1) in Differences with Intercept

Log Yen / Dollar Rate History and Long-Horizon Forecast AR(1) in Differences with Intercept

Log Yen / Dollar Rate History, Forecast and Realization AR(1) in Differences with Intercept

Log Yen / Dollar Exchange Rate Holt-Winters Smoothing

Sample: 1973:01 1994:12 Included observations: 264 Method: Holt-Winters, No Seasonal Original Series: LYEN Forecast Series: LYENSM Parameters: Alpha Beta Sum of Squared Residuals Root Mean Squared Error End of Period Levels: Mean Trend 1.000000 0.090000 0.202421 0.027690 4.606969 -0.005193

Log Yen / Dollar Rate History and Forecast Holt-Winters Smoothing

Log Yen / Dollar Rate History and Long-Horizon Forecast Holt-Winters Smoothing

Log Yen / Dollar Rate History, Forecast and Realization Holt-Winters Smoothing

Volatility Measurement, Modeling and Forecasting The main idea:

Well look at: Basic structure and properties Time variation in volatility and prediction-error variance ARMA representation in squares GARCH(1,1) and exponential smoothing Unconditional symmetry and leptokurtosis Convergence to normality under temporal aggregation Estimation and testing

Basic Structure and Properties Standard models (e.g., ARMA): Unconditional mean: constant Unconditional variance: constant Conditional mean: varies Conditional variance: constant (unfortunately) k-step-ahead forecast error variance: depends only on k, not on t (again unfortunately)

1. The Basic ARCH Process

ARCH(1) process:

Unconditional mean: Unconditional variance: Conditional mean: Conditional variance:

2. The GARCH Process

Time Variation in Volatility and Prediction Error Variance Prediction error variance depends on t-1 e.g., 1-step-ahead prediction error variance is now

Conditional variance is a serially correlated RV Again, follows immediately from

ARMA Representation in Squares has the ARMA(1,1) representation:

where Important result:

The above equation is simply

Thus

is a noisy indicator of

GARCH(1,1) and Exponential Smoothing Exponential smoothing recursion:

Back substitution yields:

where

GARCH(1,1):

Back substitution yields:

Unconditional Symmetry and Leptokurtosis C Volatility clustering produces unconditional leptokurtosis C Conditional symmetry translates into unconditional symmetry Unexpected agreement with the facts!

Convergence to Normality under Temporal Aggregation CTemporal aggregation of covariance stationary GARCH processes produces convergence to normality. Again, unexpected agreement with the facts!

Estimation and Testing Estimation: easy! Maximum Likelihood Estimation

If the conditional densities are Gaussian,

We can ignore the

term, yielding the likelihood:

Testing: likelihood ratio tests Graphical diagnostics: Correlogram of squares, correlogram of squared standardized residuals

Variations on Volatility Models We will look at: Asymmetric response and the leverage effect Exogenous variables GARCH-M and time-varying risk premia

Asymmetric Response and the Leverage Effect: TGARCH and EGARCH Asymmetric response I: TARCH Standard GARCH:

TARCH:

where

positive return (good news): effect on volatility negative return (bad news): + effect on volatility 0: Asymetric news response >0: Leverage effect

Asymmetric Response II: E-GARCH

Log specification ensures that the conditional variance is positive. Volatility driven by both size and sign of shocks Leverage effect when <0

Introducing Exogenous Variables

where: is a parameter vector X is a set of positive exogenous variables.

Component GARCH Standard GARCH:

for constant long-run volatility Component GARCH:

for time-varying long-run volatility

, where

Transitory dynamics governed by Persistent dynamics governed by Equivalent to nonlinearly restricted GARCH(2,2) Exogenous variables and asymmetry can be allowed:

Regression with GARCH Disturbances

GARCH-M and Time-Varying Risk Premia Standard GARCH regression model:

GARCH-M model is a special case:

Time-varying risk premia in excess returns

Time Series Plot NYSE Returns

Histogram and Related Diagnostic Statistics NYSE Returns

Correlogram NYSE Returns

Time Series Plot Squared NYSE Returns

Correlogram Squared NYSE Returns

AR(5) Model Squared NYSE Returns

Dependent Variable: R2 Method: Least Squares Sample(adjusted): 6 3461 Included observations: 3456 after adjusting endpoints Variable C R2(-1) R2(-2) R2(-3) R2(-4) R2(-5) R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat Coefficient 4.40E-05 0.107900 0.091840 0.028981 0.039312 0.116436 0.052268 0.050894 0.000184 0.000116 24827.58 1.975672 Std. Error 3.78E-06 0.016137 0.016186 0.016250 0.016481 0.016338 t-Statistic 11.62473 6.686547 5.674167 1.783389 2.385241 7.126828 Prob. 0.0000 0.0000 0.0000 0.0746 0.0171 0.0000 7.19E-05 0.000189 -14.36434 -14.35366 38.05372 0.000000

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic)

ARCH(5) Model NYSE Returns

Dependent Variable: R Method: ML - ARCH (Marquardt) Sample: 1 3461 Included observations: 3461 Convergence achieved after 13 iterations Variance backcast: ON Coefficient C Variance Equation C ARCH(1) ARCH(2) ARCH(3) ARCH(4) ARCH(5) R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood 3.16E-05 0.128948 0.166852 0.072551 0.143778 0.089254 -0.000381 -0.002118 0.008550 0.252519 11811.54 1.08E-06 0.013847 0.015055 0.014345 0.015363 0.018480 29.28536 9.312344 11.08281 5.057526 9.358870 4.829789 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.000522 0.008541 -6.821461 -6.809024 1.861036 Std. Error 0.000689 z-Statistic 0.000127 Prob. 5.437097 0.0000

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion Durbin-Watson stat

Correlogram Standardized ARCH(5) Residuals NYSE Returns

GARCH(1,1) Model NYSE Returns

Dependent Variable: R Method: ML - ARCH (Marquardt) Sample: 1 3461 Included observations: 3461 Convergence achieved after 19 iterations Variance backcast: ON Coefficient C Variance Equation C ARCH(1) GARCH(1) R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood 1.06E-06 0.067410 0.919714 -0.000191 -0.001059 0.008546 0.252471 11889.09 1.49E-07 0.004955 0.006122 7.136840 13.60315 150.2195 0.0000 0.0000 0.0000 0.000522 0.008541 -6.868008 -6.860901 1.861389 Std. Error 0.000640 z-Statistic 0.000127 Prob. 5.036942 0.0000

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion Durbin-Watson stat

Correlogram Standardized GARCH(1,1) Residuals NYSE Returns

Estimated Conditional Standard Deviation GARCH(1,1) Model NYSE Returns

Estimated Conditional Standard Deviation Exponential Smoothing NYSE Returns

Conditional Standard Deviation History and Forecast GARCH(1,1) Model

Conditional Standard Deviation Extended History and Extended Forecast GARCH(1,1) Model