This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: Res Ec 797A: Forecasting Order Of Topics And Readings For Lecture 1 (September 6, 2011)
Abbreviations BM: Bernie Morzuch notes
GK: Gayner and Kirkpatrick Source m Epic Syllabus 2 Material coverage; Text; Exams; Assignments
GK 1 Time line in forecasting Anonymous 1 Picture: Retail Sales  Intended to show trend and seasonality.  Seasonality especially is a very important
component for one or more of your variables in
Assignment 1. Assignment 1 Data Collection: Due September 13, 2011
(Two pages) BM 1 Purpose of Res Ec 797A
 Determining a Sen'es’ Data Generating Process
 Forecasting
Components of a Time Series
 Level
— Trend
 Seasonality
 White Noise Department of Resource Economics
University of Massachusetts Fall 2011 Res Ec 797A: Special Topics in Forecasting (The name of the course will be changed to
Applied Univariate and Econometric TimeSeries Techniques) Instructor: Bernie Morzuch
221 Stockbridge Hall Phone: 5455718
email: [email protected] Ofﬁce Hours: As you need me or by appointment Readings:
Applied Econometric Time Series by Walter Enders, Second Edition (Wiley). It is
available at the Textbook Annex. (Please note: There is a third edition. I want to stick
with the second edition). Prerequisites:
A graduatelevel course in regression, knowledge of matrix algebra, and an intennediate
level ability in statistics are necessary. Assignments:
There will be weekly assignments. These must be done in a timely fashion. The purpose
of each assignment is to ensure that you are mastering and digesting the topics as they are
being covered. Assignments will be worth 50% of your ﬁnal grade. Exams:
There will be two exams: a midterm (that is truly at the midtenn!) and a ﬁnal exam.
The ﬁnal exam will concentrate on material covered since the midteim exam. Each exam is worth 25% of your ﬁnal grade. The midterm exam will be administered
during a twohour period outside of the normallyscheduled class pen'od. The ﬁnal exam
will be scheduled during ﬁnals week. Objectives:
1 will attempt to do the following:
(1) Summarize important timeseries methodologies and quantitative forecasting
techniques in use and the issues surrounding their use;
(2) Discuss in detail some of the most important (though not necessarin the most
widely used) techniques;
(3) Develop the theory necessary to understand the techniques used. This is not an econometric theory, statistical theory, or forecasting theory course.
Forecasting is a pragmatic art. My goal is to have the course reﬂect this pragmatism. Course Organization:
The material will be presented in a lecture format. Implementation of the appropriate
computer software will be explained as we proceed through the lectures. You will not be
required to make inclass presentations nor will you have to do a paper for this course. Course Outline:
I will begin with background material on forecasting. This includes terminology,
followed by simple and common univariate estimation procedures not found in Enders. I
will provide you with readings that explain these univariate techniques in detail. I then
move to several notsosimple univariate methods. All of this paves the way for
exploring the behavior of univariate time series and, in particular, testing for their
stationarity. This semester we have 26 class periods. Approximately half of the course
(12 lectures) will be devoted to this background material and to univariate techniques. For the remaining 14 lectures, we will be covering the topics in Enders presented below.
When you pick up the text and page through these readings, you may ﬁnd them a bit
frightening. I will attempt to lessen the difﬁculty by deciphering these pages for you.
Enders is close to stateoftheart regarding timeseries econometrics. During the last lecture, we address what happens to forecasts when we combine forecasts
from different methods. We complete the course by comparing the forecasting
performance of all of the techniques that we've studied through the semester. Enders
 Difference equations: Chapterl
 Stationary time—series models: Chapter 2  Autoregressive conditional heteroscedasticity (ARCH): Chapter 3, pp. 112145
 Testing for trends and unit roots: Chapter 4 . Unit root processes, pp. 156181 ' Unit root tests, pp. 181229 . Multiequation timeseries models: Chapter 5
. Intervention analysis, pp. 240247
 Transfer function models, pp. 247264
' Vector autoregression, pp. 264283
 Granger causality, pp. 283290  Cointegration and errorcorrection models: Chapter 6
~ Cointegration, pp. 319328
 Errorcorrection models, pp. 328—339 Time Line In Forecasting Historical
y data Y
BEGINNING Y END
T
Sample ______________ _ _
Y1 Yt Yn Observations Withinsample
, forecasts . EX post Ex ante A A Backcasts forecasts A forecasts A
Yn+1 YN YN+1 YN+k Q‘Bvackcasting period é————— Forecasting period ——>
%— Backcasts m v0 Source: Gayner and Kirkpatrick, 1994, p.11. 60.00 5000 1 unaﬁ 4000 HO—ﬂm 3000 2,000 ‘ 13 25 37 Retail Sales 49 NSAISA 61
Month I NSA 73 85 ““ " SAL/CHAS  _.._:___ 97 109 University of Massachusetts
Department of Resource Economics RESEC 797A Fall 2011
Assignment 1 Data Collection Because this is a course with a practical orientation, you will use data to make quantitative
forecasts. Collect time series of three variables. I encourage you to collect data related to your
research interest. If you already have data that meet the criteria set down below, so much the
better. Variable 1 should be a series that is reported monthly or possibly quaiterly. Its most important
feature is that you expect it to display a seasonal pattern. Possible contenders are prices of
many agricultural commodities, sales of some luxury consumer goods, incidences of infectious
diseases, etc. Please do not collect any meteorological data. Variable 2 can be monthly, quarterly or annual data. A higher frequency (such as daily stock
prices) is not desirable. This is a series that you do not expect to display a seasonal pattern. Variable 3 is related to either variable 1 or variable 2. Its most important feature is that you can
provide some causeandeffect explanation. Variable 3 should be at the same frequency as the
variable it causes. To choose variable 3 rely on theory for guidance. For example, in economics,
quantity produced might be related to price, quantity exported to exchange rate, employment to inﬂation, etc. (We are not talking high theory here!) Each series must meet the following requirement: Preferably at least 50 observations, definitely at least 40. With monthly data try to get twice that
amount. If there are one or two missing observations, you could insert values for the missing
data. Preferably ﬁnd a seiies where this is not necessary. Some forecasting techniques require a
minimum of 40 observations. You will also want to retain the last few observations for post
sample testing. That is, you will make ex post forecasts. In future exercises retain 8 years, 8
quarters or 18 months of data for postsample testing. Consistency is important! We want to
make comparisons at the end of the semester, both across forecasting methods and across
forecasters (i.e. you). Some ideas for sources of data will be found on the next page. First order of business: using a spreadsheet, produce a separate plot of each series against time.
Brieﬂy describe the data, looking especially for trends, seasonal patterns and outlier
observations. Just exactly what these terms mean will be discussed in future classes. For the
moment, use your understanding of the everyday meaning of the terms. You must present a
detailed writeup for each variable. You must explain your plots, trends, plots, and anything else
in the context of your writeup. You must use proper grammar. I will deduct points for poorly
written assignments. Sources of data
You can ﬁnd a variety of data series (economic and noneconomic) by using Google.
Examples of web sites for economic timeseries data
1. Energy Information Administration, Department of Energy: www.cia.doe. gov
2. Census Bureau: www.census. gov
3. Economic Research Services, U. S. Department of Agriculture: www.ers.usda.gov
4. Federal Reserve System: www.federalresewe. gov 5. Bureau of Economic Analysis, US. Department of Commerce: www.bea. gov Typical examples of pairs of economic variables that may be related
1. Monthly consumption and average price of natural gas for residential use
2. Number of workers employed in the farm sector and farm wages
3. Production (output level) of any agricultural commodity and its price
4. A country's gross domestic product (gdp) and its investment
5. Monthly residential building starts and the mortgage interest rate
6. Industrial production and the unemployment rate
7. A country‘s exchange rate and its level of imports
8. A country‘s exchange rate and its level of exports 9. Inﬂation rate and energy prices Purpose of Res Ec 797A
This course is concerned with analyzing the behavior of a time series. Issues related to
behavior are:  Determining a series' DGP:
Can we identify the reasons or the process that generates the series or data?
We will devote considerable attention to trying to ﬁnd the data generating
process (DGP) for each series that we analyze. ‘ Forecasting:
Once we feel that we have captured the behavior of the series, can we use this
knowledge to predict or forecast where the series will go, beyond the limits of the
data set used to capture its DGP?  Components of A Time Series
A time series typically has one or more of the following components at time period t:
level (Lt); trend (Tl); seasonal variation (8,); cycle (C1); and white noise or an irregular
(It) or €1TOI‘(81) ﬂuctuation. General deﬁnitions of these components are as follows:  level (L1): The value where the series starts or where it tends to return or
gravitate as time progresses.  trend (Tl): An upward or downward movement in the series. The duration of the
movement can be short or long. It does not have to be persistent, because
persistence is a matter of judgment. In this context you will be exposed to two
kinds of trend: deterministic and stochastic.  seasonal variation (8,): Commonly referred to as seasonality. It very often is
applied in the context of a calendar year. It is the regular monthly or quarterly
ﬂuctuation within a year. It is a pattern of change that completes itself within a
year. It is repeated on a yearly basis and is inﬂuenced by some seasonal factor; e. g.,
weather or custom, sales at Christmas, turkeys at Thanksgiving. Seasonality is not
restricted to a year. There can be seasonality within a week or even within a day.
Exploring seasonality depends upon the frequency of the data being analyzed, e. g.,
either high frequency or low frequency. ' cycle (Ct): A long term upward or downward change, the duration of which can be
two to ten years or longer. It is measured from peak to peak and from trough to
trough. In economics, we see that the economy goes through business cycles. A
variable like gross domestic product (gdp) is used to analyze the peaks and troughs
in a business cycle. I will not address cycles in this course.  white noise (at): A random disturbance. l ...
View
Full Document
 Fall '11
 BernardMorzuch

Click to edit the document details