2017 4th International Conference on Economics and Management (ICEM 2017) ISBN: 978-1-60595-467-7 Analysis on the Development Trend of Per Capita GDP in Yunnan Province Based on Quantile Regression Yong-sheng YANG 1,a,*, Li LI 1,2,3,b and Lei XIAO 4,5,c 1 Pan-Asia Business School, Yunnan Normal University, Kunming, Yunnan, China 2 Yunnan Association for Promotion of Trans-Asian Financial Cooperation and Development, Kunming, Yunnan, China 3 Champion Property & Casualty Insurance Co., Ltd., Kunming, China 4 School of Marxism Studies, Kunming University, Kunming, Yunnan, China 5 Research Center of Socialism with Chinese Characteristics, Kunming University, Kunming, Yunnan, China a 43063010@qq.com, b leisurelily@qq.com, c 391644664@qq.com *Corresponding author Keywords: Quantile Regression, Per Capita GDP, AR Model, Least Squares Regression. Abstract. Per capita GDP is an important index to measure the level of economic development and comprehensive economic strength of a country or region. In this paper, the quantile regression method is used to analyze the temporal data of per capita GDP in Yunnan province, and the quantile regression model of per capita GDP development in Yunnan province is established to comprehensively analyze the economic development of Yunnan province. The results show that with the increase of the quantile level and the number of prediction periods, the relative error of the quantile regression model is gradually reduced, the model fitting effect is better than the linear regression model and the time series model, and the quantile regression has good ability of prediction. 1. Introduction Gross domestic product (GDP) is an important indicator of the national economic strength and economic scale. Per capita GDP of Yunnan is from 226 yuan in 1978 to 25083 yuan in 2013. Selecting a suitable model to analyze its trend, the growth law, the fluctuation and the relationship between the development processes, has important significance for government to develop appropriate macro-control policies. Liu Ying adopted the Box-Jenkins model to predict per capita GDP in China [1]. Dong Zhenguo used the autoregressive moving average model (ARIMA) of the time-dependent model [2]. Jin Shan used ARIMA (1, 1, 1) model to analyze the data of Guizhou from 1950 to 2006, and revealed the growth law of Guizhou [3]. Quantile regression theory was first proposed by Koenker and Bassett in 1978. He used the conditional quantile of the dependent variable to regress the independent variable, obtained the regression model under different quantiles. Koerker and Zhao studied the quantile regression method of conditional heteroskedastic autoregressive model [4]. The time series model generally requires that the sequence is a stationary non-white noise sequence, and the model s order is very subjective, which affects the prediction effect of the model. The quantile regression is a modeling method for estimating the linear relationship between a set of regression variables and the quantile of the explained variables. It emphasizes the change of the conditional quantile, which can compensate for the shortcomings of the least squares pair of spikes, thick tails and heteroskedasticity. Therefore, the quantile regression can be applied to the development trend analysis of per capita GDP in Yunnan province. 165
2. Modeling Analysis 2.1 Data source The per capita GDP data of Yunnan province from 1978 to 2010 are obtained from the Statistical Yearbook of Yunnan Province. And the data are divided into two parts. The per capita GDP data from 1978 to 2010 are used as training set. The data from 2011 to 2013 are used as test set. 2.2 Pre-processing of data Firstly, we use Eviews to get unit root test results of per capita GDP. The unit root test value is -0.65, which is indicating that the sequence is not stable and cannot be directly modeled. Then, the original data should be linearized by logarithmic transformation, and the transformed sequence is LOGY. Unit root test results of LOGY show that the logarithm of data eliminates the exponential trend, and show linear growth trend, but still not stable. Therefore, it is necessary to perform the firstorder difference transformation of the new LOGY sequence, and the new sequence is DLY. The unit root test result is shown in figure 1. Figure 1. Unit root tests of difference series. From the unit root test results, it can be seen that the t value is -3.01, which is lower than the level of 5% and 10%, and the p value is 0.0453 less than 0.05, which further shows that the DLY sequence has been stable at 95% confidence level. 2.3 The time series modeling of DLY sequence Results of ACF and PACF analysis are shown in figure 2. Figure 2. ACF and PACF analysis results of DLY. It can be seen that the autocorrelation diagram is obviously trailing and the partial autocorrelation graph can be regarded as a first-order truncation and can be also regarded as a trailing. So the model may be AR (1) or ARMA (1, 1). The AR (1) model is superior to ARMA (1,1) from the calculated AIC and BIC values. 166
2.4 Linear regression modeling of LOGY sequence The growth of per capita GDP is largely related to the per capita GDP of the previous year. Therefore, this paper chooses LOGY as the dependent variable, LOGY (-1) and T as independent variables, and the regression results are as follows: LOGY = 14.6246 + 0.0076T + 0.9418 LOGY ( 1) (1) 3. Quantile Regression Modeling of LOGY Sequence 3.1 Definition of quantile regression If the distribution function of random variable Y is F( y ) = P( Y y ), then the τ quantile of Y is Q( τ ) = inf{ y : F( y) τ}. 3.2 Estimation of quantile regression By analyzing the LOGY sequence data of per capita GDP after taking the logarithm in Yunnan province, it is concluded that the regression results of quantiles under different points are shown in figure 3: Figure 3. Estimation of quantile regression. The coefficients of the three variables are different under different quantiles, and the coefficients of LOGY (-1) can be tested at the significance level of 0.01 under different quantile and are all greater than T coefficient. With the increase of the quantile, the coefficients of the intercept term T decrease, the coefficients of C and LOGY (-1) increase. When the quantile is 0.714, the coefficient of the intercept term C rises obviously, the coefficients of T and LOGY (-1) decrease significantly. 4. Comparison of Each Model Prediction The prediction of per capita GDP of Yunnan province in 2011-2016 was carried out by AR (1), least squares regression and quantile regression, and we compare the predicted results for 2011 and 2013 with the real values in table 1. According to table 1, the relative error of AR (1) is smaller than that of least squares regression, which shows that the fitting effect of short-term prediction is satisfactory. But we know that AR (1) is based on the logarithm of the original data and the difference, the loss of the original data may be large, and can only be used for short-term forecast. With the extension of the forecast period, the prediction error may be gradually risen, the prediction process is calculated indirectly through the 167
model, so the error may be relatively large. The least squares regression model can only get the average level, and the quantile regression model more fully depicts the development of per capita GDP in Yunnan province. At different levels of quantiles, the relative errors predicted are different. At the same quantile level, the relative error decreases with the increase of the forecast period and gradually reduces with the increase of the quantile level. The fitting value is generally high in the high score of 0.857 and generally low in the low score of 0.147, so this will be able to predict the future trend of per capita GDP in Yunnan province under different circumstances. model AR (1) least square regression Quantile regression 0.143 0.286 0.429 0.5 0.571 0.714 0.857 Table 1. Comparison of three model predictions. year 2011 2012 2013 Real number 19265 22195 25083 Predictive number 17841 20529 23574 relative error 0.07 0.08 0.06 Predictive number 17646 20153 23014 relative error 0.08 0.09 0.08 Predictive number 16759 18328 20185 relative error 0.13 0.17 0.20 Predictive number 16874 18493 20330 relative error 0.12 0.17 0.19 Predictive number 17640 20158 23049 relative error 0.08 0.09 0.08 Predictive number 17948 20834 24164 relative error 0.07 0.06 0.04 Predictive number 17941 20811 24113 relative error 0.07 0.06 0.04 Predictive number 18285 21711 25867 relative error 0.05 0.02 0.03 Predictive number 18202 21414 25153 relative error 0.06 0.04 0.00 The following is the 2014-2016 per capita GDP data in Yunnan province predicted by the quantile regression model corresponding to high-score of 0.857. Table 2. Predictions of high score points. Quantile regression 2014 2015 2016 Predictive number 29496 34531 40358 As can be seen from the forecast results, per capita GDP in Yunnan province is increasing year by year, which is inseparable from the government s policy promoting steady economic development. Since 2014, under the influence of tightening macroeconomic environment and weak market demand, the economic pressure of the province has been increasing, but under the relevant macro-control of the government, the province s economy has been generally stable. Therefore, the estimating the development trend of per capita GDP in Yunnan province by using high score of 0.857 is relatively reasonable. 5. Conclusions In this paper, the per capita GDP of Yunnan province is modeled by the quantile regression model, and the results are compared with those of the time series model and the linear regression model. It is found that the quantile regression model can extract the information hidden in the data in more detail, and can get the estimated value under different points, and the prediction results are more comprehensive. Empirical results show that per capita GDP in Yunnan province increases year by 168
year, showing a rapid growth. In order to maintain the steady development of this growth and realize the rapid development of the economy in Yunnan province, we can focus on solving the problems of slowing investment, business difficulties and slowing consumption. Acknowledgements This research was financially supported by 2016 year Yunnan Philosophy and Social Science Planning Project (YB2016016), Scientific Research Foundation of Yunnan Provincial Education Department (2016ZZX080), Yunnan Normal University Dengfeng rooted Outstanding Scientific Innovation Group Project, Doctor Research Foundation of Yunnan Normal University and Scientific Research Foundation of Kunming University (XJZZ1617). References [1] Y. Liu, Z. H. Zhang, Per Capita GDP in China (1952-2002) Time Series Analysis. The Statistics and Decision Making, 2 (2005) 61-62. [2] Z. G. Dong, Per Capita GDP Time Series Model of China. Academic Theory, 15 (2005) 38. [3] S. Jin, Application of ARIMA Model in Guizhou GDP. Financial Perspective, 10 (2007) 154-155. [4] R. Koenker, Q. Zhao, Conditional Quantile Estimation and Inference for ARCH Models. Econometric Theory, 12 (1996) 193-813. 169