eviews forecasting

How to use EViews software to forecast time series

/article/1709ad8089a69e4634c4f0ff.html

The purpose of doing econometric analysis is to explore the correlation inherent in the economic phenomenon, and the effectiveness of the forecasting effect is a test of the existence of such a relationship and the magnitude of the explanatory power The standard of this is to test the existence of this relationship and the strength of the explanation. Models are generally divided into two categories, one is based on a single series of models, and the other is our common commonly used multiple series model.

One of the single-series model prediction

The most commonly used method is the exponential smoothing method, which has been described in detail before.

ARMA model

The time series can be transformed into a multiseries model by introducing the time series

Second, the prediction of multiseries models

1, general understanding

When the regression equation between the sequences is determined can be based on the regression equation to be fitted to the prediction. Eviews path: the window of the regression equation ----- forecast.

The measure of forecasting effectiveness is the forecast error, i.e., the deviation between the predicted value and the actual value. The parameters of the forecast output interface are understood as follows:

First, Root Mean Squared Error (RMSE). Find the square root of the equal possible weighted sum of squares of the prediction error

Second, Mean Absolute Error (Mean Absolute Error). Averaging the absolute value of the prediction error.

Third, the average relative error (Mean Abs. Percent Error). Its formula and the average absolute variance are the same as the absolute value of the error is used, but here to divide by the actual value, so the final measure is the relative error. Because the formula is multiplied by 100, so the output results of its value of the unit of measure is a percentage.

Fourth, Theil Inequality Coefficient (Theil Inequality Coefficient). The meaning of its formula is also very clear, the numerator is the root mean square of the error, the denominator is the predicted value is likely to be weighted square and open square root + the actual value is likely to be weighted square and open square root. Therefore, the value of this coefficient interval is 0-1, the closer to 0, that is, the smaller the root mean square of the unit error, that is, the closer the predicted value and the actual value, the better the model fit. The same is true for the values of the first three parameters, the smaller the better the model fit, but what is called small is impossible to determine.

The predicted mean squared error can be decomposed into the sum of three indicators, the bias ratio, the variance ratio, and the covariance ratio, and the sum is 1:

First, the bias ratio (Bias Proportion)

Second, the variance ratio (Variance Proportion)

Third, the covariance ratio (Covariance Proportion)

Third, the covariance ratio (Covariance Proportion)

This is a good example of how the model fits the model. Covariance Proportion)

The Variance Ratio indicates the degree of deviation of the predicted mean from the actual mean of the series (the ratio of the square of the difference between the predicted mean and the actual mean to the mean side of the error); the Variance Ratio indicates the degree of deviation of the predicted variance from the actual variance of the series (the ratio of the square of the standard deviation of the deviations of the distributions of the predicted and the actual values to the mean side of the error); the Covariance ratio measures the magnitude of unsystematic error (the ratio of the covariance of the distributional deviations of the predicted and actual values to the mean square of the error).

If the predictions are good, then the bias and variance ratios should be small and the covariance ratio large.

2, dynamic forecasting and static forecasting

Dynamic forecasting is a multi-step forecast, in addition to the first predicted value of the actual value of the explanatory variables is used to predict the subsequent periods of the predicted value of the recursive forecasting method, lagged explanatory variables (that is, the so-called dynamic term) of the previous prediction is substituted into the estimation (prediction) equations to predict the predicted value of the next period.

Static forecasting, on the other hand, involves making a series of one-step predictions, i.e., it must use the true values of the explanatory variables to make predictions, rather than using the predicted values of the explanatory variables as the explanatory variables to make predictions. Static forecasting requires that the observed values of the exogenous variables and any lagged endogenous variables in the forecast sample are available. If data are not available for a particular period, the predicted value for that period is NA. however, it does not have an impact on later predictions. So, if static forecasting is done it is also necessary to give the values of the explanatory variables used for forecasting as well as the values of the lagged explanatory variables. If the prediction is to be based on the estimated values of the explanatory variables, the window of the series of the explanatory variables needs to be opened first and these estimates need to be added into their corresponding intervals before the prediction is made.

3, the explanatory variables for the formula of the prediction

If the formula of the explanatory variables is a simple form, Eviews in the prediction to determine the predictive sequence will be given two choices: First, the entire formula as a whole as a predictive sequence; Second, the first occurrence of this formula as a predictive sequence, such as the formula for x/2y, then x as the predictive sequence.

But when the formula is more complex, Eviews can only take the whole formula as the predicted sequence. Even, when the formula is too complex, Eviews concedes to itself and gives an error message that it can no longer hold the scene and cannot solve for the formula.