Forecasting is a current topic of growing important in business and

economic analysis. It is an attempt to foresee the future by examining the past.

It consists of generating unbiased estimates of the fbture magnitude of some

variable, on the basis of past and present knowledge and experience. Generally

a forecast may be considered as a statement about future. There are many ways

of making forecasts. In economics forecasting methods include: (i) guessing,

rule of thumb or informal models (ii) expert judgment (iii) extrapolation (iv)

leading indicators (v) surveys (vi) time series models and (vii) economic

systems. It is difficult to judge the frequency with which each of these methods

is used in practice, but most occur regularly in our everyday lives.

Basically, forecasting process involves the following components: (i)

Determining the objective (ii) Developing a model (iii) Testing the model (iv)

Applying the model (v) Evaluating and revising the model,

Since three decades considekble amount of research has betn conducted

in the field of forecasting. Attempts have been made to make forecasting as

9

scientific as possible. Statistics is the base of scientific forecasting. The

problem of forecasting is essentially statistical in nature. There are mainly two

aspects of scientific forecasting. The 'first is the analysis of the past conditions

and the second is the analysis of current conditions in relation to a prospective

future trend. Scientific forecasting requires detailed information about the past

movements and special Wors affecting the movement.

 In stathtks,g-f refers to the extending or projecting time series

into the futu;e lwscd on past behaviour of the quantitative data. Forecasts are

made by estimating fbttme values of the external factors by means of

prediction, projection and forecast and fiom these values, calculating the

estimate of the dependent variable.

The three terms prediction, projection and forecast can be distinguishes

separately.

Prediction is an estimate based solely on past data of the time series under

investigation. It is purely mechanical extrapolation.

A Projection is a prediction where the extrapolated values are subject to

certain numerical assumptions.

A Forecast is an estimate which relates the series in which we are interested to

external factors.

With the development of more sophisticated forecasting techniques,

along with the advent of computers, forecasting has received more and more
I

attention. Forecasts are needed in finance, marketing and production areas; in

Government and profit seeking organizations; in National Political Parties etc,

The main point to remember is that' a forecast must result in present action to

improve the future.

 Tbc main ~~ in the 6ald of fbmcasbg are Yuk (1927).
M.EuJcy (1930). w&A (1931). Slunky (1937). Wold (1930, Cox (1961.

1968) Mc Laughlin (1962). Box and Jenkins (1962, 1976), Brown (1962).

Harrison (1965). Naylor and Seaks (1972). Nelson (1973). mil (1975), GFoss

and Peterson ( 1976). Montgomery and Johnson ( 1 976). Thornopoulos (1 980),

Klien and Young (1980). Young (1982). Holden, Peel and Thompson (1990),

Christoffersen and Diebold (19%). Makridakis. Wheel Wright and Mc Gee

(1998). Pindyck and Rubinfeid (1998). G u e m and Pcna (2000).

Timmermann and Moody (200 1). Newbold (2002). Froietti (2005) and others..

In the present research study, an attempt has been made to propose some

new forecasting methods based on regression and time series analysis.

1.2 STATEMENT OF THE PRESENT RESEARCH PROBLEM

The past three decades have witnessed a number of developments in

estimation and prediction that have direct relevance and applicability in

organizational forecasting. The forecasting literatwe is only now begi~ingto

focus on translating what is theoretically possible and computationally feasible

I

into a f- that can be easily understood and applied.

Several forecasting techniques may be divided into two major categories

namely, Quantitative and Qualitative or tcchnoiogical mahods. Qualiarti%

techniques can be divided into t h e series and causal mtthods and qualitative
ot aectrndogicat mothods caa be divided into expbmxy and normative
mtthoda

Quantitative fwecasting can be applied when three conditions exist:

i. information about the p i is available.

ii. This information can be quantified in the form of numerical data.

iii. It can be assumed that some aspects of the past pattern will continue

into the future.

In qualitative forecasting. all information and judgement relating to an

item are used to forecast the items demand. This technique is often used when

little or no demand history is available. These forecasts may be based on

marketing research studies. the Delphi method or similar methods.

In causal forecasting, a cause and effect type o f rcscarch is sought. Hm,

the forecaster seeks a relation between an item's demands and other factors

such as business induitrial and National indices. ?his relationship is used to

forecast the fuhuc demands of the item.

Ln time series analysis forecasting, a statistical analysis on p t &men&

is used to generate the forecasts. A basic assumption is that the underlyb

Time series models can often be

trends of the past will continue into the h.

usad more easily to forecast, whereas causal models can be used with greater
success for policy and dccision making.
 Fcaeawhg @chdps tfiat am based on regredon analysis are

substantially 'diffaart in their underlying c9nccpts and thaory h m the

techniques of time series M y s i s , d n g and decomposition. Rtgressim

techniques are g t d l y r e f d to as causal or explanatory approaches to

forecasting. T)#y attempt to predict the firturc by discovering and measuring

the effect of important independent variables on the dependent variable to be

forecast. Because of their costs, these methods are gemrally used in long-run

planning and in situations where the value of increased accuracy warmnts the

additional expense.

Econometric models for forecasting are generally much simpler and

involve fewer equations than those designed for policy study. 'Mttse models arr

systems of relationships between the variables under investigation. Their

equations are then estimated fiom the available data, mainly aggregate time

series. A forecast made, from an econometric model may not be exact, because

the estimates of the parameters are.subjtct to sampling errors.

Forecasting methods can be cornpap! by Monte Carlo study or

Stochastic Simulation, where an investigator generates artificial data on which

the models art compared in repeated trials, to calculate how well such rntthods

perform in a controlled environment or their own choosing.

13 -OF THE PRESENT RESEARCH STUDY

llre main objective-ofthe present nsearch study is to develop some new

fomcahg methods based on applied regression analysis.

'RE specific a i m of the present study art:

(i) to brief the =view about the various basic forecasting models

existing in the literature;

(ii) to explain the need for adaptive, filtering and combining forecasts,

for forecasting methods;

(iii) to p p s c a mehod of choosing optimum weights for combimd

foremsting technique;

(iv) to develop new forecasting model based on regression stability;

(v) to propose a new method of estimation for ARCH forecasting model;

(vi) to give some tests for disturbances in ARCH and GARCH

forecasting models.

1.4 SOME BASIC FORECASTING TECHNIQUES

Several fortcasting techniques have' becn proposed diflerrntly by

different Statisticians and Econometricians from time to time. Some important

forecasting methods existing in the literature arc:

I. Naive Method

ii- Single and Double Moving Average Methods

iii. Single and Double Expomntial Smoothing Models

 iv. Brwlm's o a S . ~ M c t h o tbr
d Exponential Slaoothing
V. bit's Two-Pammem Method fbr hpmential S m i

vi. Wikr's T h r e o r ~ e rT d and Sessonality Mcthod for

Exponential Smoothing

vii. A-ve Contml Smoothing Models

viii. Simple R e p i o n Forecasting Model

ix. Multiple Rcgn?ssion Forecasting Model

X. Time Series Regmion F o m t i n g model

xi. Cross Sectional Regression Forecasting Model

xii. Trigonometric Forecasting Models

...
XIIl. Multidimensional Forecasting Models

xiv. Autoregressive Moving Average (ARMA) Forecasting Models

xv. Autoregressive Integrated Moving Average (AFUMA) Fomxsting

Models

xvi. Box-Jenkins Forecasting Models

xvii. Transfer Function Forecasting Model

xviii. Statistical Forecasting Model by using Kalmans Filtering

xix. . Econometric F o m t i n g Models '

1.5 ORGANIZATION OF THE PRESENT STUDY

The organization of the present msearch work itself shows how the

objectives of the present study arc achieved within the described h e work.
 Cbpter-I isanintruductwyonc. It c a t a h t h e g e d ht
m duc
w

but tbe problem of rorraaiqg kid0 objectives and

. . of the

-t m.It also brings- the various basic forecasting t-uea

Chapter - I1 describes the, various basic forecasting models such as

Naive, Moving averages, Simple smoothing, Double moving avetages and

Double smoothing, triple smoothing and adaptive smoothing forecasting

models. It a h gives a brief note on Box - Jenkins forecasting model.

Chapter - It1 deals with the Adaptive, Filtering and Combination hr

forecasting techniques. A harmonic madel for adaptive fortcasting has been

explained besides filtering demands from demand patterns. The procedure of

evaluation of forecasting accuracy is also presented in this chapter.

Chapter - I V gives the need for exponential smoothing forecasting

model along with model selection criterion. It also discusses abut

exponentially weighted moving average forecasting model besides adaptive

smoothing for univariate time series.

Cbapter - V presents the presents the various autoregressive forecasting

modcls with thcir link with

models-such as ARMA. AMMA and S T A ~ A
dynamic li& models.

Chapter - VI proposes some new forecasting techniques in

cconornetrics. The combining method of forecasts has been cxplainad by
proposing optimum values for the wtights in the combined forecast. Using a

test for regression stability, a new f d n g technique has been developed by

splitting tht time series data into ~clro subsets of obwvations. An
~utoregressivcHeterosocdasticity for#=astingmodel has been specified and its
parameters have been estimated. Also the tests for ARCH disturbances and

GARCH disturbances have been explained.

-
Chapter VII epitomizes the conclusions based on the present research

work. It also gives directions for future research in the line of the p m t

research study.

Several relevant articles regarding the f o m ting techniques have been

presented under a separate title 'BIBLIOGRAPHY'.

1.6 CHAPTER SCHEME

The concepts of the proposed research study are presented under the

following heads:

CHAPTER - I INTRODUCTION

CHAPTER - Il : SOME BASIC FORECASTING MODELS

CHAPTER - 111 : ADAPTIVE. FILTERING AND COMBINATION

FOR FORECASTINq TECHNIQUES

CHAPER - IV EXPONENTIAL SMOOTHING FORECASTING

MODELS
CHAPTER - V : AUTOREGRESSlVE FORECASTING MODELS

CHAPTER - VI : SOME NEW FORECASTlNG TECHNIQUES

CHAPTER - VII : CONCLUSIONS

BIBLIOGRAPHY