determination of an appropriate ARIMA(p,q) model to represent
an observed stationary time series involves a number of
interrelated problems. These includes the choice of p and q,
the coefficients and some other statistics.
In this Add-In an Excel function is included to estimate the coefficients of an ARMA(p,q) model (p is the order of autoregressive terms AR, q is the order of moving average term MA).
and useful statistics will be displayed.
(std. errors, t-statistics, p-values, (adjusted) R-squared,
SSR, Akaike information criterion, Schwartz criterion, Durbin
Watson,...). The impulse response function, predictions and
inverted MA/AR roots will also be computed.
To estimate the coefficient this tool uses a non linear estimation technique (Levenberg-Marquardt algorithm). Note that on the one hand estimation of non linear models is much more expensive then solving a model by OLS, on the other hand estimation is an approximation by numerical techniques. If moving average terms are included this [web:reg] ARMA back forecast the moving average terms. (Box and Jenkins: "Time series analysis: forecasting and control", 1976)
The output of the function is difficult
for many persons to recognize. For this reason I integrated
a VBA form. The input is simplified and the outputs is formatted
and diagrams will be created.
There are also some time series functions
included to transform a time series.
An example worksheet to estimate an
ARMA(p,q) model and a documentation is also included.
ARMA Excel Add-In