Penalties for Misclassification of a Pure Diagonal Bilinear Process of Order Two as a Moving Average...

O. E. Okereke, I. S. Iwueze, C. O. Omekara

  Open Access OPEN ACCESS  Peer Reviewed PEER-REVIEWED

Penalties for Misclassification of a Pure Diagonal Bilinear Process of Order Two as a Moving Average Process of Order Two

O. E. Okereke1,, I. S. Iwueze2, C. O. Omekara1

1Department of Statistics, Michael Okpara University of Agriculture, Umudike, Abia State, Nigeria

2Department of Statistics, Federal University of Technology, Owerri, Imo State, Nigeria

Abstract

The penalty function based on misclassification of a pure diagonal bilinear process of order two as a moving average process of order two was derived in this study. Computation of penalties using the penalty function revealed that such misclassification increases the error variance. Regression analysis of the penalties on the parameters of the pure diagonal bilinear process suggested a second order polynomial regression model. A test of significance of each of the parameters of the fitted model showed that all the parameter estimates were statistically significant at 5% level of significance. The analysis of variance technique was also used to confirm the adequacy of the fitted model.

At a glance: Figures

Cite this article:

  • Okereke, O. E., I. S. Iwueze, and C. O. Omekara. "Penalties for Misclassification of a Pure Diagonal Bilinear Process of Order Two as a Moving Average Process of Order Two." American Journal of Applied Mathematics and Statistics 2.1 (2014): 47-52.
  • Okereke, O. E. , Iwueze, I. S. , & Omekara, C. O. (2014). Penalties for Misclassification of a Pure Diagonal Bilinear Process of Order Two as a Moving Average Process of Order Two. American Journal of Applied Mathematics and Statistics, 2(1), 47-52.
  • Okereke, O. E., I. S. Iwueze, and C. O. Omekara. "Penalties for Misclassification of a Pure Diagonal Bilinear Process of Order Two as a Moving Average Process of Order Two." American Journal of Applied Mathematics and Statistics 2, no. 1 (2014): 47-52.

Import into BibTeX Import into EndNote Import into RefMan Import into RefWorks

1. Introduction

Much attention is usually given to the structures of the autocorrelation function (ACF) and partial autocorrelation function (PACF) during the identification stage of a time series model. The moving average model is a linear time series model with known structures of autocorrelation function and partial autocorrelation function [10]. Let be a sequence of independent and identically distributed random variables with zero mean and variance . Then is a non zero mean moving average process of order two (MA(2) process) if:

(1.1)

The moving average process in (1.1) has the following first and second moments [3]:

(1.2)
(1.3)

and

(1.4)

The autocorrelation function in (1.4) cuts off at lag two ([2, 3]) .Other properties of the autocorrelation function are and [9].

A non linear time series model which competes with the moving average process in (1.1) in terms of autocorrelation function structure is the pure diagonal bilinear time series process of order two (PDB(2) process) defined by [4]:

(1.5)

where is a sequence of independent and identically distributed random variables with zero mean and constant variance, and are real constants. If and, then the first and second moments of the model in (1.5) are as follows [8]:

(1.6)
(1.7)
(1.8)

It is quite obvious that the ACFs in (1.4) and the one in (1.8) all cut off after lag two. This is indicative of the fact that a moving average process of order two and a pure diagonal bilinear time series process of order two have similar autocorrelation structures. As a result, there is a possibility of misclassifying a pure diagonal bilinear process of order two as a moving average process of order two. The ease with which linear models are fitted and the practice of approximating nonlinear models by linear models can also cause misspecification of the nonlinear pure diagonal bilinear process of order two.

From the foregoing, it is imperative to investigate the statistical implication of the aforementioned model misclassification. In this regard, we will focus on the penalty function associated with misclassification of a PDB(2) process as an MA(2) process.

2. Relationship between the Parameters of the Pure Diagonal Bilinear Process of Order Two and Moving Average Process of Order Two

Having observed that the moving average process of order two and pure diagonal bilinear process of order two have similar autocorrelation structures, it is worthwhile to derive the relationship between the parameters of the two models. These relationships will help us to obtain the penalty function for misclassifying the nonlinear model as the competing linear model. The method of moments which involves equating the first and second moments of the pure diagonal bilinear model to the corresponding moments of the non zero moving average process of order two shall be used for this purpose.

Equating means, we have

(2.1)

Equating variances, we obtain

(2.2)

Equating first order autocorrelations leads to:

(2.3)

Equating second order autocorrelations gives

(2.4)

Dividing (2.4) by (2.3), we have

(2.5)

From (2.5), we obtain

(2.6)

Substituting (2.6) into (2.3), we obtain

(2.7)

where

(2.8)

and

(2.9)

One way of solving (2.7) involves modifying it first to obtain

(2.10)

Here, a, b and k are constants such that

(2.11)
(2.12)
(2.13)

The value of k satisfying (2.11), (2.12) and (2.13) is obtained by solving the equation

(2.14)

Substituting into (2.14), we obtain

(2.15)

Following the methods of [5] and [1], the real solution of (2.15) is found to be

(2.16)

where

(2.17)

and

(2.18)

Hence,

(2.19)

Comparing (2.7) and (2.10), we have

(2.20)

It can be deduced from (2.20) that

(2.21)

or

(2.22)

Solving (2.21), we obtain

(2.23)

or

(2.24)

From (2.22), we have

(2.25)

or

(2.26)

When , we have

(2.27)

For , we obtain

(2.28)

Using , we have

(2.29)

With , we obtain

(2.30)

Simulation concerning (2.23), (2.24), (2.25), (2.26) and their corresponding values of namely , , and respectively shows that only and take on values satisfying the invertibility condition of a moving average process of order two.

3. Penalty Function for Misclassification of a PDB(2) Process as an MA(2) Process

Penalty function based on model misclassification in time series analysis is defined by [6] as a function of error standard deviations. Let be the standard deviation of the errors associated with a PDB(2) process. Suppose represents the standard deviation of the errors corresponding to an MA(2) process. Then the penalty function for the misclassification of PDB(2) as an MA(2) is given as

(3.1)

We can write (3.1) as

(3.2)

Using (2.2), we obtain

(3.3)

Substituting (3.3) into (3.2) leads to

(3.4)

and in (3.3) are as defined in (2.27) and (2.23) respectively. Table 1 contains the penalties (P) corresponding to various values of , , and .

Considering the complete table containing 2129 sets of values, we can see that the penalty function for misclassification of a PDB(2) process as an MA(2) process (P) takes on positive values for all values of , , and . The positive value of the penalty for misclassification of a PDB(2) process as an MA(2) process shows that this misclassification leads to increase in variance of the errors. This finding agrees with the results obtained by [6] with regard to misclassification of a PDB(1) process as an MA(1) process.

For predictive purposes, we have to find the relationship between P and and. First, we plot P against each of and. Figure 1 shows the plot of P against.

Table 1. Penalties for various Values of Parameters of MA(2) Process and PDB (2) Process

It can easily be seen from Figure 1 that there is a curvilinear relationship between P and . In Figure 2, we have the plot of P against.

Figure 2 also reveals that there is a curvilinear relationship between P and .

Combining the information in Figure 1 and Figure 2, the regression model in (3.5) is suggested for the relationship between P and and .

(3.5)

where , , , and are the parameters of the regression equation (3.5) and is the associated error term. The least squares estimation of (3.5) based on the admissible values of . and leads to the predictive equation [7].

(3.6)

Table 2 contains the summary of the test for significance of each of the parameters of the model in (3.6).

Table 2. Test for Significance of the Parameters of the Regression Model for Penalty for Misclassication of a PDB(2) Process as an MA(2) Process

Each of the parameters appears to be significant at level of significance since the corresponding p value is less than 0.05. Next, we test for significance of the overall regression using the analysis of variance technique as shown in Table 3.

Table 3. ANOVA Table for Testing for Significance of the Fitted Penalty Model

The p value of 0.00 in the Table 3 implies that the fitted regression model is suitable for describing the relationship between P and and.

4. Conclusion

In this study, we determined the effect of misclassifying a pure diagonal bilinear process of order two as a moving average process of order two. A penalty function was defined and was used to compute penalties for misclassification of the pure diagonal bilinear process of order two as the moving average process of order two based on various sets of values of the parameters of the two processes. The computed penalties assumed positive values. This indicated increase in error variance due to misclassification of pure diagonal bilinear process of order two as a moving average process of order two. A quadratic regression model was found suitable for predicting the penalties based on the parameters of the pure diagonal bilinear process of order two.

References

[1]  Bessels, S. (2006). One step beyond the solvable equation. www.staff.science.uu.nc/…/Afstudeerscriptie_Sander_Bessels.pdf (This site was visited in June, 2013).
In article      
 
[2]  Box, G. E. P., Jenkins, G. M. and Reinsel, G. C.(1994). Time Series Analysis: Forecasting and Control. 3rd ed., Prentice-Hall, Englewood Cliffs, N. J.
In article      
 
[3]  Chatfield, C. (1995). The Analysis of Time Series. An Introduction. 5th ed. Chapmann and Hall, London.
In article      
 
[4]  Granger, C. W. J. and Andersen, A. (1978). An Introduction to Bilinear Time Series Models. Vanderhoeck and Ruprecht, Gottingen.
In article      
 
[5]  Hahn, K. (2005). Solving cubic and quartic polynomials. www.Karlscalculus.org/pdf/cubicquartic.pdf.
In article      
 
[6]  Iwueze, I. S. and Ohakwe, J. (2009). Penalties for misclassification of first order and linear moving average time series processes. Interstat Journal of Statistics, No3, http//interstatjournals.net/Year/2009/articles/0906003.pdf.
In article      
 
[7]  Okereke, O. E. (2013). Characterization of moments of pure diagonal bilinear process of order two and moving average process of order two, Unpublished Ph. D Dissertation.
In article      
 
[8]  Okereke, O. E and Iwueze, I. S. (2013). Region of comparison for the second order moving average and pure diagonal bilinear processes. International Journal of Applied Mathematics and Statistical Sciences, 2(2): 17-25.
In article      
 
[9]  Okereke, O. E, Iwueze, I. S and Johnson, O. (2013). Extrema for autocorrelation coefficients of moving average processes. Far East Journal of Theoritical Statistics, 42(2): 137-150.
In article      
 
[10]  Wei, W. W. S. (2006). Time Series Analysis, Univariate and Multivariate Methods. 2nd ed. Pearson Addision Wesley, New York.
In article      
 
comments powered by Disqus
  • CiteULikeCiteULike
  • MendeleyMendeley
  • StumbleUponStumbleUpon
  • Add to DeliciousDelicious
  • FacebookFacebook
  • TwitterTwitter
  • LinkedInLinkedIn