Regression-model

Introduction

As regressors are chosen for a linear regression model, AFR package recommends to check for:

1. Optimal size of the time-series data

Function opt_size assess whether time-series data has enough observations for the chosen model.

model<-lm(real_gdp~imp+exp+usdkzt+eurkzt, macroKZ)
opt_size(model)
#> There is acceptable number of observations.
#> It is necessary to have 24 observations.
#> Your regression has 39 observations.
#> Warning in opt_size(model): If there is equal or close number of observations,
#> please check further.

Based on the output of the function, modify the model, i.e. remove or add regressor(s).

2. Choose the best regression model

From the initially built linear regression model regsel_f function allows to choose the best regressors by Akaike Information criterion (AIC) and Adjusted R-squared (Adj R2) parameters. These parameters are set by default, but other parameters can be added too.

To dive into details, check_betas function demonstrates all models with regressors’ betas based on which regsel_f function gives the result. A user can export the output of all models into Excel document for more representative format by using function write_xlsx of writexl package.

check_betas(model)

3. Analysis of the model

As regsel_f gave the best regression model, it can be analysed by diagnostic tests for the compliance with Gauss-Markov theorem for a multiple regression model.

Graphically, the regression model can be visualized for decomposition and forecasting. Function dec_plot demonstrates a contribution of each regressor in a form of stacked bar plot.

dec_plot(model, macroKZ)

Function reg_plot shows actual and forecast data. Forecasting can be performed by Arima or trending.

reg_plot(model, macroKZ)