QMETH 530, Spring, 2005: Forecasting Models in Business (4)

MW 8:30 – 10:20 AM in BLM307

OBJECTIVES:

The objective of this course is to introduce statistical forecasting methods for management. Statistical forecasting methods view the world as a collection of processes that generate data. Forecasting data, which will be generated from a process, is based on a statistical model of the way the process generates data. Such a model is called a forecasting model. A rich collection of standard forecasting models exists. Therefore, a manager need not invent a new model every time s/he forecasts. Instead, her/his task is to identify an appropriate forecasting model from the collection.

Below is the list of standard forecasting models that we learn in this course. They are core models of forecasting. For each model, management functions where the model is most used are listed in parentheses.

  • FM1smoothing (marketing, operations management)
  • FM2fixed trend and seasonality (marketing, economics)
  • FM3stationary ARMA for cycles (finance and operations management)
  • FM4integrated ARMA for variable trends (accounting, finance, economics)
  • FM5regression on time series data (macroeconomics and finance)
  • FM6 intervention analysis (operations management)
  • FM7GARCH for volatility (finance)

Excelworks well for data preparation and also for FM1: smoothing. But we need a dedicated statistical package for implementing other models. For this, the school has acquired site license for Eviews. You can purchase your own copy of the software at a substantial discount. See the order form appended at the end of this syllabus.

COURSE INFORMATION AND REQUIREMENTS:

Course Instructor:Hiro Tamura, Professor, MS

Office location:Mackenzie 362

Office hours:MW 10:30 – 11:30 AM, 4:00 – 5:00 PM.

E-mail address:htamura@u

Tel & Dept. Fax:Tel 206-543-4399; Dept. Fax 206-543-3968

Grading:

Homework 30%

Project35

Final Exam (Take-home)35

Total100%

COURSE INFORMATION AND REQUIREMENTS (cont.):

Texts and References:

Required:

Diebold, F. X. (2004) Elements of Forecasting Third Edition South-Western.

(A link to the text web site is available on the course web.)

References (available at the Foster Business Library reserve desk):

D: Dielman, Terry E. (2005) Applied Regression Analysis. Fourth Edition. Thomson.

NB: Newbold, Paul and Bos, Theodore. (1994) Introductory Business and Economic Forecasting. Second Edition. South-Western.

COURSE SCHEDULE:

Key terms: Terms you are expected to understand for the session.

Read: Try to read before each class meeting.

Cases: Each mini-case illustrates application of a specific forecasting model.

Eviews:Eviews commands for learning

1.3/28/MCourse Orientation

Overview of Forecasting and Applications

Key terms: process, time series data, information set, forecast horizon, forecast statement, forecast loss function; sources.

components of time series data (trend, season, cycle, and irregular),

forecasting model = DGP (data generating process)

Read: Diebold, Ch. 1 and 2.

2. 3/30/WIntroduction to Eviews

UWComputing Resources: BLM 401, CSSCR (145 Savery)

Basic Eview commands:

creating a workfile (workfile, frequency, range)

timeplot (varname.line), creating a time (series time=@trend()+1)

trend line (ls varname c time)

summary statistics (scalar m = @mean(varname) , @var @stdev)

transformations(log(varname), varname(-1), d(varname))

Cases: Data Analysis with Eviews

Statistical Graphics for Forecasting

Key terms: aesthetics, aspect ratio, golden ratio

timeplot (varname.line), Actual, Fitted, Residual Graph

histogram (varname.hist)

Read: Diebold, Ch. 3 (in particular Section 3)

Eview: Freeze/ Line / Shade

3. 4/4/M FM1 Exponential Smoothing

Key terms: simple exponential smoothing, h-step ahead forecast

recursive algorithm, error correction form

Read: Dielman Ch. 11, NB Ch. 6: 6.1-6.3

Cases: Case EX: Exponential Smoothing for Forecasting

Eviews: smooth(s, sc)

Eviews Practice 1 Due

4. 4/6/W FM1 (cont.) Exponential Smoothing for Trend and Seasonality

Key terms: Holt-Winters method

Read:Dielman Ch. 11, NB Ch. 6: 6.4, 6.7, 6.10.

Cases: Case HL: Holt’s Linear Trend Algorithm

Case HW: Holt-Winters Algorithm for Seasonal Data

Eviews: smooth(“.”, s.c.) “.” = n, a, m

5. 4/11/MFM1 (cont.) Construction of Seasonal Index

Key terms: moving average

Read:NB Ch. 5.3, 5.4

Cases: Case: Seasonal Index

Eviews: @movav, @quarter, @month, @seas

series_name(int.)- lead if pos. int., lag if neg. int., seas

HW#1 Due

6. 4/13/WFM2 Fixed Trend Models

Key terms: linear, log-linear, quadratic, logistic, Gompertz

method of least squares, non-linear regression, initial estimates

regression coefficients, SE. of regression, recursive estimation

white noise

Read: Diebold, Ch. 4

Cases: Case FT1: Fund Performance Comparison

Case FT1A: Growth Rate of GDP

Case FT2: MLB-Average Salary

Case FT3: Cardiac Operations

Eviews: ls, param,

7. 4/18/MFM2 (cont.)

Recursive Estimation

Cases: Case FT1B: Funds Performance Comparison

Eviews:ls/view/stability tests/recursive estimates/recursive coefficients

Modeling Seasonality

Key terms: seasonal dummy variables

Read: Diebold Ch. 5

Cases: Case FS Modeling Seasonality

Eviews: @seas

Eviews Practice 2 Due

8. 4/20/W FM2 (cont.) Diagnostics

Key terms: Actual, Fitted, Residual graph, R-squared (adjusted and un-adjusted)

white noise, the base model, illlusory trend, Durbin-Watson test, auto-correlated residual

Read: Ch. 1 Appendix, Ch 4

Cases: Case TS: Test of significance of the model, Of a single coefficient

Case: Illusory Regression - Demonstration

Case DW: Testing Autocorrelation for Residuals

Eviews: @nrnd, @qfdist(p, v1, v2), @fdist(f-stat, v1, v2), @qtdist(p, v), @tdist(x, v)

(@cnorm(x), @qnorm(p) for the standard normal distribution.)

9. 4/25/MFM3 Correlogram

Test for Randomness Using Correlogram

Key terms: autocorrelation, correlogram (ACF), standard error of AC

Ljung-Box Q-test,-distribution

Read: Diebold Ch. 6: 6.2, 6.5, and problem 5 on page 28.

Cases:Case CGM1: Annual Rainfall in the SeaTac Area

Case CGM2: Correlogram of Non Random Series

Eviews: view/ correlogram, @cchisq(x, v), @chisq(x,v), @qchisq(p, v)

Tests for Normality

Key terms: normal plot, skewness, kurtosis, Jarque-Bera statistic
Read: Diebold Ch. 1. Additional Problems and Complements 5

Eviews: view/descriptive statistics, view/ distribution/ quantile-quantile graph

HW2 Due

10. 4/27/W FM3 (cont.) Autoregressive (AR) Models – Identification and Fitting

Key terms: stationary series, AR(p), partial autocorrelation, inverted ar roots

Read: Diebold Ch. 6: 6.1, 6.3, 6.4, 6.5; Ch. 7:7.2

Cases: Case AR1: Forecasting Item Movement Using Autoregressive Process

Eviews: ls series_name c ar(1), ar(2)

11. 5/2/M FM3 (cont.) Moving Average (MA) Models – Identification and Fitting

Key terms: MA(q)

Read: Diebold Ch. 7:7.1, 7.3, 7.4.

Cases: Case MA: Forecasting Item Movement Using Moving Average Process

Eviews: ls series_name c ma(1), ma(2).

FM3 (cont.) Autoregressive Moving Average (ARMA) Models

Identification and Fitting

Key terms: ARMA(p, q)

Cases:Case ARMA: Forecasting Item Movement Using ARMA Process

Eviews: ls series_name c ar(1) ar(2) ma(1) ma(2).

Eviews Practice 3 Due

12. 5/4/W FM3 (cont.) Forecasting for ARMA

Key terms: Out-of-Sample Forecast, RMSE, Mean Absolute Error, Mean Absolute Percentage Error, Theil’s Decomposition of MSE.

Read: Diebold Ch. 8

Cases:Case Pred_ARMA: Point Forecast for ARMA Models

Case OSEval: Out-of-Sample Forecast Evaluation

Eviews: View/Forecast, sample

13. 5/9/MFM3 (cont.) Putting It All Together

Key terms: planning

Read: Diebold Ch. 9

Cases: Case PT: Building a Fixed Trend Forecasting Model

Case Diebold 9.2 Forecasting Liquor Sales

HW3 Due

14. 5/11/W FM4 Variable (Stochastic) Trend Modeling

Key terms: variable (stochastic) trend, random walk, random walk with drift,

I(1) process, first difference

Read: Diebold Ch. 12:1, and 12:3

Cases:Case RW: Behavior of Stock Price – Random Walk

Case ST: Stochastic Trend Modeling – Forecasting GDP

Case Diebold 12.3 Modeling and Forecasting The Yen/Dollar Exchange Rate

Fixed vs. Stochastic Trend – Long Run Path Comparison

Read: Nelson, C.R. and Plosser, C. I. (1982) Trends and Random Walks in Macroeconomic Time Series. Journal of Monetary Economics10

Cases: Case LR: Fixed vs. Stochastic Trend – Long Run Path Comparison

Unit Root Tests

Key terms:unit root () tests

Read Diebold Ch. 12:2

Cases: Case URT1: Unit Root Tests 1 (Dickey-Fuller Tests)

Case URT2: Unit Root Tests 2 - More Examples

Case URT3: Unit Root Tests 3 Augmented Dickey Fuller Tests
Eviews View/Unit Root Test

15. 5/16/MFM4 (cont.) Variable (Stochastic) Trend With Seasonality

Key terms: Seasonal Difference, Seasonal ARMA

Cases:Case S_ST: Stochastic Trend with Seasonality

Eviews: d(series_name, 1, s),sar(s), sma(s)

Eviews Practice 4 Due

16. 5/18/WFM5 Regression on Time Series Data

Key terms: spurious regression, distributed lag models

Read: Diebold Ch. 12

Cases:Case RT: Spurious Regression on Time Series

Case Demand for Gasoline

17. 5/23/MFM6 Intervention Analysis

Key terms: intervention series: pulse, step, transfer function

Read:

Cases:Case INTERV-1: Monitoring Labor Hours

Case INTERV-2: Effect of Tax Rebate on Savings Rate

HW4 Due

18. 5/25/WFM7 GARCH Models for Dynamic Volatility

Key terms: conditional heteroscedasticity, ARCH, GARCH, TARCH

Read: Diebold Ch. 13

Cases: Case GARCH-1:Conditionally Heteroscedastic Models

Case GARCH-2:Extensions of GARCH

Eviews: arch(p, q, options)

19.5/30/MMEMORIAL DAY HOLIDAY

20. 6/1/WReview / Course Evaluation / Final Exam Distribution

21. 6/6/MFinal Exam, Evaluation of Team Members, Course Project Due

1