Page 25 - Azerbaijan State University of Economics

THE JOURNAL OF ECONOMIC SCIENCES: THEORY AND PRACTICE, V.79, # 2, 2022, pp. 19-36

The second problem is that these type models may have the overfit problem in case
of higher-order polynomials. Therefore, the data should be taken into account while
selecting such polynomial models. Increasing the order of model can lead to
overfitting problem. Overfit (or over-adaptation) occurs when the training data used
in building the model is very consistent with the data and as a result, it generates a
model defining specific generalization to the data. In this case, although the training
is in perfect harmony with the data, the model built on the next used data results in a
high error. Xing Wan considers that in order to avoid over-fitting problem in
polynomial regression, a regularization method should be used to suppress the
coefficients of higher-order polynomial. As well as while solving polynomial
regression problems, polynomial regression can be transformed into linear regression
to solve. The more complexity of the model results in the higher degree of overfitting.
Collecting the large volume of data can avoid overfitting problem too (Schönbrodt,
2015; Xing Wan, 2019). While estimating of the parameters of model it requires the
assumption that the errors follow the normal distribution (Ostertagova, 2012).

For the purpose of the study, the statistical data involving the accrual interest income
on loans and gross domestic product (GDP) are taken from the statistical bulletins of
the Central Bank of Azerbaijan (Source: Official web-site of Central Bank of the
Republic of Azerbaijan, https://www.cbar.az/page-40/statistical-bulletin). The
amounts of tax revenues refer to the official web page of Ministry of Finance of the
Republic of Azerbaijan (Source: Official web site of Ministry of Finance of the
Republic of Azerbaijan, http://www.maliyye.gov.az/static-page/periodic-reports).

EMPIRICAL ANALYSIS AND ECONOMETRIC ESTIMATIONS
In addition to the complexity of the interpretation of polynomial regression models, it
has a number of desirable features. The inclusion of a quadratic term in the model
allows to determine the influence of the quadratic effect. In the absence of a quadratic
effect between an independent and a dependent variable, in other words, in cases
where the relationship curve is more complex, the polynomial regression equation of
an independent variable can be included in a higher order (Stimson, James et al.,
1978). The main goal of the study is to find the interest burden that gives the maximum
level to the GDP and the tax revenues, so that in the same time it ensures the maximum
level in the financial results of banking sector.

25