Document Type : Research Paper

Authors

1 Ph.D. Candidate in Statistics, Allameh Tabataba’i University, Tehran, Iran

2 Associate Professor, Department of Statistics, Allameh Tabataba’i University, Tehran, Iran

Abstract

Modeling plays a crucial role in economic and financial research, forming the foundation for analysis, decision-making, policy development, and planning. Assumptions made during the modeling process are particularly important for estimation and forecasting, as they can significantly influence the results. One of the most widely used classical time series models is the autoregressive model, in which current values are expressed as a finite linear combination of past values. However, in real-world scenarios, many variables interact with each other. To capture these interdependencies, vector time series models-an important class of multivariate time series models-are employed. The vector autoregressive (VAR) models are commonly used in economic and financial modeling. VAR models are typically formulated assuming that the shocks (or noise terms) follow a normal distribution. However, in economic and financial contexts-particularly in macroeconomics-shocks do not often follow a symmetric distribution. The present article focused on a VAR model in which the shocks follow a multivariate skew normal (MSN) distribution. The expectation conditional maximization (ECM) algorithm were used to estimate the model parameters. Finally, using real-world datasets from Canada and Iran-where the shocks exhibit skewness-the study found that the VAR model with MSN-distributed shocks is more efficient than the VAR model with multivariate normal distribution for shocks.

Introduction

The multivariate normal distribution is commonly used to model shocks in VAR models. However, in fields such as economics, finance, the stock market, and medicine, various factors can introduce skewness (asymmetry) into the shocks, resulting in non-symmetric distributions. In such cases, the normal distribution becomes an inappropriate choice. To address this, the multivariate skew distribution-which accounts for asymmetry-should be used for modeling shocks. Despite its relevance, this approach has received limited attention in previous research. The family of multivariate skew distributions is broad and complex, posing practical challenges. The present study aimed to test the VAR model in which the shocks follow a multivariate skew normal (MSN) distribution, using the real-world datasets from Canada and Iran.

Materials and Methods

Consider the VAR model of order p:
 
 
 
where   ,  o is location parameter,  is the scale parameter, and S is the skew parameter. Its density function is given by:
 
where  , , , and   denotes density function of MN, while  denotes the standard cumulative distribution function.
To find the maximum likelihood estimates of the parameters requires derivatives of the log-likelihood function; however, these derivatives do not have closed-form expressions. Therefore, they must be approximated using numerical methods. The maximum likelihood estimators were obtained via the expectation conditional maximization (ECM) algorithm. Based on the hierarchical representation of the multivariate skew normal distribution, we have:
,
(o, I)
,
The logarithm of the conditional likelihood function in VAR(p) can be formulated as:
 
 
where . The expectation of the logarithm of the conditional likelihood is denoted by , and the steps for the maximizing  are as below:

Step 1: Assuming no skewness, estimate the initial values for the coefficients and scale parameters.
Step 2

 
 
                    

Step 3

 
 

Step 4

 
 
 

Step 5: Repeat steps 2 to 5 until the convergence condition of the algorithm is established:

 

Results and Discussion

The performance of the proposed method was evaluated using two real-world datasets from Canada and Iran. The Dickey-Fuller test was employed to determine the stationarity of the data, while the Akaike Information Criterion (AIC), Hannan-Quinn Criterion (HQC), and Schwartz Bayesian Criterion (BIC) were used to select the order of the VAR model. The Canadian dataset consists of seasonally adjusted employment and unemployment data from 1980 to 2000. According to the Mardia test, the shocks follow a multivariate skew normal (MSN) distribution. Therefore, we estimated the parameters of the VAR(1) model. The AIC and BIC results are presented in Tables 1 and 2, respectively.
 Table 1. The Estimated Parameters of Model for Stationary Differenced Canadian Data




Estimation


 


 




 


0.9103


0.7094




 


0.2139


-.0149




 


-0.2018


-0.4663




 


0.3303


0.0255




 


-


0.0703




 


-


0.1066




 


0.1646


0.1595




 


-0.09106


-0.1003




 


0.09106


-0.1003




 


0.11400


0.0977




Table 2. The AIC and BIC for Data




Distribution


AIC


BIC


-Log-Like




 


282.0266


286.836


139.011




 


260.6781


265.4915


128.339




The collected data on agriculture, forestry, and fishing (AFF) and employment of women (EW) in Iran span the years 1991 to 2021. According to the Mardia test, the shocks follow a multivariate skew normal (MSN) distribution. The parameter estimates for the VAR(1) model, along with the AIC and BIC values, are presented in Table 3.
   Table 3. The Estimated Parameters of Model for Stationary Differenced Iranian Data




Estimation


 


 




 


-


0.1104




 


-


-0.0798




 


1.2397


1.2267




 


0.2748


0.2836




 


0.2748


0.2836




 


0.3855


0.3796




AIC


146.1145


145.9185




BIC


148.8491


148.6531




The fitted model can be formulated as follows:
 
 
where    and     represent  AFF  and EW, respectively.
According to the AIC and BIC criteria presented in Tables 2 and 3 for the Canadian and Iranian data, the VAR model with MSN-distributed shocks is more appropriate than the VAR model with MN-distributed shocks.

Conclusion

Considering the VAR(p) model with shocks following a multivariate skew normal (MSN) distribution, the present study employed the maximum likelihood method and the ECM algorithm to estimate the model parameters. Based on two real-world datasets from Canada and Iran, the findings showed that the VAR model with MSN-distributed shocks provides a better fit than the model with MN shocks when the shocks exhibit skewness.

Keywords

Main Subjects

Akaik, H. (1974). A new look at the statistical model identification. IEEE Trans Automat Contr. 19, 716–723.
Arellano-Valle, R B., Ozan, S., Bolfarine, H., Lachos, VH. (2005). Skew-normal measurement error models. Journal of Multivariate Analysis, 96, 265–281. doi.org/10.1016/j.jmva.2004.11.002
Arellano-Valle, R.B. & Azzalini, A. (2006). On the unification of families of skew-normal distributions. Scandinavian Journal of Statistics, 33, 561-574. doi.org/10.1111/j.1467-9469.2006.00503.x
Arellano-Valle, R.B. & Genton, M.G. (2005). Fundamental skew distributions. Journal of Multivariate Analysis, 96, 93– 116.
Arellano-Valle, R.B., Gomez, H.W. & Quintana, F.A. (2004). A new class of skew-normal distributions. Communication in Statistics- Theory and Methods, 33,1465-1480. doi.org/10.1081/STA-120037254
Azzalini, A. & Capitanio, A. (1999). Statistical application of the multivariate skew- normal distribution. Journal of Royal Statistical Society, series B, 61, 579-602. doi.org/10.1111/1467-9868.00194
Azzalini, A. & Capitanio, A. (2014). The Skew-Normal and Related Families. Cambridge CB2 8BS, United Kingdom.
Azzalini, A. (1985). A class of distributions which includes the normal ones. Scandinavian Journal of Statistics, 12, 171-178.
doi.org/10.6092/ASSN.1973-2201/711
Azzalini, A. (2005). The skew-normal distribution and related multivariate families. Scandinavi Journal of Statistics, 32, 159-188.
doi.org/10.1111/1467-9469. 2005.00426.x
Balakrishnan, N. & Scarpa, B. (2012). Multivariate measures of skewness for the skew-normal distribution. Journal of Multivariate Analysis, 104, 73–87. doi.org 10.1016/j.jmva.2011.06.017
Bondon, P. (2009). Estimation of autoregressive models with epsilon-skew-normal innovations. Journal of Multivariate Analysis, 100(8), 1761-1776. doi.org/10.1016/j.jmva.variate2009.02.006
Box, G. & Jenkins, G. (1970). Time Series Analysis: Forecasting and Control. Holden-Day, San Francis.
Esfanjir, A.A. & Rezaei R.H. (1399). Investigating the effect of women’s employment on economic growth in selected countries of Middle East. Quarterly Journal of Woman and Society. 11,207-226. [ In Persian]
Faryaar, H., Macdonald, R. & Watt, J. (2022). Improving the measurement of the contribution of women to the economy: estimates of GDP. Economic analysis division at statistics Canada.36, 28-001. doi.org/10.25318/36280001202201000003-eng
Ghasami, S., Khodadadi, Z. & Maleki, M. (2019). Autoregressive processes with generalized hyperbolic innovations. Communication in Statistics- Simulation and Computation.49(12), 3080-3092. doi.org  /10.1080/ 03610918.2018. 1535066
Gupta, A. & Chang, C. (2002). Multivariate skew-symmetric distributions. Applied Mathematics Letters, 16(5), 634-646. doi.org/10.1016/S0893-9659(03)00060-0
Heidari, H. (2011). An alternative VAR model for forecasting Iranian inflation: An application of bewley transformation. Iranian Journal of Economic Research, 46, 77-96. [ In Persian]
Khorsandi, M., Mohammadi, T., Arab, H. & Sakhaei, E. (2022). The effect of external economic shocks on Iran’s macroeconomic variable: Global VAR approach. Iranian Journal of Economic Research, 27, 9-50.  doi.org/10.22054/ijer.2020.52537.868 [ In Persian]
Kilian, L. & Lütkepohl, H. (2017). Structural Vector Autoregressive Analysis. (Themes in Modern Econometrics), Cambridge, United Kingdom.
Koop, G. & Korobilis, D. (2009). Bayesian multivariate time series methods for empirical macroeconomics. Foundation and Trends in Econometrics, 3, 267-351. doi.org/10.1561/0800000013
Liu, J., Kumar, S. & Palomar, D. (2019). Parameter estimation of heavy-tailed AR model with missing data via stochastic EM. IEEE Transaction on Signal Processing, 67(8), 2159- 2172. doi.org/10.1109/ TSP.2019. 2899816
Liu, Y., Sang, R. & Liu, Sh. (2016). Diagnostic analysis for a vector autoregressive model under Student’s t- distributions statistic. Neerlandica, 71(2), 86- 114. doi.org/ 10.1111/stan.12102
Lütkepohl, H. (1991). Introduction to Multiple Time Series Analysis. Springer-Verlag. Berlin and New York.
Lütkepohl, H. (2005). New Introduction to Multiple Time Series Analysis. Springer Science & Business Media.
Lütkepohl, H. (2020).  Structural vector autoregressive models with more shocks than Variables identified via heteroscedasticity. Economics Letters, 195, 19510458. doi.org /10.1016/j.econlet.2020.109458
Maleki, M., Wraith, D., Mahmoudi, R. & Javier, E. (2019). Asymmetric heavy-tailed vector auto-regressive processes with application to financial data. Journal of Statistical Computation and Simulation, 90(2), 324-340. doi.org/10.1080/ 00949655. 2019.168067
Mardia, K. (1970).  Measure of multivariate skewness and kurtosis with application. Biometrika, 36, 519-530. doi.org/10.1093/biomet/57.3.519
Mardia, K. (1974). Applications of some measures of multivariate skewness and kurtosis in testing normality and robustness studies. Sankhy- a, ser. B. 36, 115–128. doi.org/10.1080/00369791.1974.11659939
Mirzaei, H., Razban, N., Mohammadi, T. & Morovat, H. (2023). Analyzing the housing market network among Iran’s. Provinces: New evidence through variance decomposition of dorecast errors, 88, 120-157. doi.org/10.22054/joer.2024.75890.1159
Mohammadi, T., Azizkhani, F., Taei, H. & Javid, B. (1398). Macroeconomic dynamics of deregulation in product markets and work in Mena countries: Panel VAR. Iranian Journal of Economic Reasearch, 80, 37-67. doi.org/10.22054/ijer.2019.11112
Neethling, A., Ferreira, J. & Naderi, M. (2020). Skew generalized normal innovations for AR(p) process endorsing asymmetry. Symmetry, 12(8),1253. doi.org /10.3390/sym12081253
Ni, Sh. & Sun, D. (2005).  Bayesian estimates for Vector autoregressive. Journal of Business and Economic Statistics, 23, 105-117. doi.org /10.1198/073500104000000622.
Pourahmadi, M. (2001). Foundation of Time Series Analysis and Prediction Theory; John Wiley & Sons, Inc: Hoboken, NJ, USA.
Schwarz, G. (1978). Estimating the dimension of a model. Ann Stat, 6(2), 461–464. doi.org /10.1214/aos/ 117634 4136
Sharafi, M. & Nematollahi, A. (2016). AR(1) model with skew-normal innovations. Metrika, 79, 1011–1029. doi.org /10.1007/s00184-016-0587-7.
Sims, C. (1980). Macroeconomic and reality. Econometric, 48, 1-48. doi.org/10.2307/1912017
Stock, J.H. & Watson, M.W. (2001). Vector autoregressive. Journal of Economic Perspectives, 15(4), 101-115. doi.org /10.1257/jep.15.4.10
Sujit, S., Dipak, D. & Marco, B. (2003). A new class of multivariate distributions with applications to Bayesian regression models. Canadian Journal of Statistics, 129(31), 129-150.
 doi.org/10.2307/3316064
Tarami, B. & Pourahmadi, M. (2003). Multi-variate t autoregressive: innovations, prediction variances and exact Likelihood equations. Journal of Time Series Analysis, 24, 739-754. doi.org / 10.1111/j.1467-  9892. 2003.00332.x
Tsay, R.S. (2002). Analysis of Financial Time Series. John Wiley & Sons, Inc. ISBN: 0-471-41544-8
Tsung, L. (2009). Maximum likelihood estimation for multivariate skew normal mixture models. Journal of Multivariate Analysis, 100(2), 257-265. doi.org/10.1016/j.jmva.2008.04.010
Tsung, L., Hsiu, H. & Chiang, Chen. (2009). Analysis of multivariate skew normal models with incomplete data. Journal of Multivariate Analysis, 100, 2337-2351. doi.org/10.1016/j.mva.2009.07.005
Wai, C., Mumtaz, H. & Pinter, G. (2017). Forecasting with VAR models: fat tails and stochastic volatility. International Journal of Forecasting, 33(4), 1124-1143. doi.org/ 10.1016/j.ijforecast.2017.03.001
Wei, W. (2006). Time Series Analysis Univariate and Multivariate Methods. Boston, Pearson Addison Wesley.
Zamani Mehryan, S. & Sayyareh, A. (2015). Statistical inference in autoregressive models with non-negative residuals. Statistical Research and Training Center, Iran JSRI 2015. 12(1), 83-104.