Forecasting Methods And Stock Market Analysis (Page 4 of 8) in pdf

106

Virginica Rusu and Cristian Rusu

that there is no information on the causality relationship, which aﬀects the

variable that has to be forecast.

Generally speaking, a time series model will be preferred when: (1) we

have little information on the factors that aﬀect the behavior of the variable,

(2) we have a great amount of data, or (3) the main aim is short-term

prediction.

The time series analysis starts by building a data model whose properties

are similar to the generation of the analyzed process. If we suppose that the

properties of the analyzed process, included in the model, will continue into

the future, then the model can be used for prediction.

The extrapolation with time series diﬀers from simple extrapolation. The

diﬀerence occurs because the time series analysis presupposes that the series

whose behavior has to be forecast was generated by a stochastic (random)

process, whose structure can be characterized and described. In other words,

a time series model gives a description of the nature of the (random) process

that generated the time series. The description is not carried out in terms

of cause-eﬀect, as in the case of the regression model, but in terms of the

form in which the event is incorporated into the process.

The time series models can be either deterministic or stochastic. The

models that make no reference to the source or random variation of the

series are deterministic. A stochastic model of a time series will give more

information as a deterministic model, allowing an improved forecast.

The traditional methods of time series analysis suppose that the series

are composed of four elements: the tendency, the cyclic component,

the seasonal component, and random changes. The ﬁrst three of the

above-mentioned components are deterministic, systematic, whereas the last

one is a residual component that provides the analyzed phenomena with the

feature of a stochastic process.

The roles of the components in the forecast process are diﬀerent, depend-

ing on the length of the time interval for which the forecast is carried out.

In short-term forecasts, the residual component has a major importance. In

long-term forecasts, the most important one is the trend component.

A widely used forecasting method is the exponential smoothing. The

simple version of the exponential smoothing is suitable to the series that

do not have an evident trend. The Holt-Winters method is suitable for the

series with a strong tendency component.

The moving average methods take into consideration the most recent

data of a dynamic series. The inﬂuence of the recent data decreases with

the increasing number of the values (periods) that are used. If the dynamic

series has random changes on large time intervals, a larger amount of data

will be used. If the dynamic series has a certain conﬁguration and the

random changes are abrupt and occur at small time intervals, then a smaller

Forecasting Methods And Stock Market Analysis Page 4