Prediction of Rural Residents Consumption Expenditure Based on Lasso and Adaptive Lasso Methods

Transcription

1 Oen Journal of Statistics, 2016, 6, htt:// ISSN Online: ISSN Print: X Prediction of Rural Residents Consumtion Exenditure Based on Lasso and Adative Lasso Methods Xiaoting Tao, Haomin Zhang College of Science, Guilin University of Technology, Guilin, China How to cite this aer: Tao, X.T. and Zhang, H.M. (2016) Prediction of Rural Residents Consumtion Exenditure Based on Lasso and Adative Lasso Methods. Oen Journal of Statistics, 6, htt://dx.doi.org/ /ojs Received: October 20, 2016 Acceted: December 23, 2016 Published: December 27, 2016 Coyright 2016 by authors and Scientific Research Publishing Inc. This work is licensed under the Creative Commons Attribution International License (CC BY 4.0). htt://creativecommons.org/licenses/by/4.0/ Oen Access Abstract When the variable of model is large, the Lasso method and the Adative Lasso method can effectively select variables. This aer rediction the rural residents consumtion exenditure in China, based on resectively using the Lasso method and the Adative Lasso method. The results showed that both can effectively and accurately choose the aroriate variable, but the Adative Lasso method is better than the Lasso method in rediction accuracy and rediction error. It shows that in variable selection and arameter estimation, Adative Lasso method is better than the Lasso method. Keywords Lasso Method, Adative Lasso Method, Consumtion, Prediction 1. Introduction Consumtion, investment and exort have always been referred to as the troika of economic growth. In China, investment and exort are the main ower of economic growth for a long time. But comared with the investment and the exort, the level of residents consumtion in our country is very low, esecially the rural residents consumtion which deressed for a long time. So, the government also stressed several times that we need to focus on exanding domestic demand, esecially consumer demand. Therefore, discovering the main factors influencing the rural residents consumtion in China has a very imortant ractical significance. There are many emirical studies of literature about consumer sending in China. For examle, by using anel data of 31 Chinese rovinces from 2000 to 2011 through regression analysis, Zhao [1] found that the main factors influenced the rural residents DOI: /ojs December 27, 2016

2 consumtion level is the net income of rural residents. From the ersective of total and category exenditure, Liu and Wu [2] analyzed the local government exenditure on livelihood and rural consumtion from the macro ersective, and made an emirical test by using the anel data of 31 Chinese rovinces from 1998 to Chu and Yan [3] established a variable intercet model of the fixed effect of local government exenditure on rural residents consumtion demand. They found that the effect of local government exenditure of suorting agriculture on rural residents consumtion demand is ositive, but transfer ayment is not significant. Chen and Liu [4] set u different tyes of rural credit model which affects the rural residents consumtion exerience, found that whether short-term or long-term, consumer credit can boost rural consumtion better than roduction. Li [5] emirically analyzed the relations between the social security exenditure and the rural consumtion in China. Wen and Meng [6] used ELES model from the urban and rural residents consumtion structure comarison, rural residents marginal roensity to consume, marginal budget share, basic consumer sending, the actual consumer sending, consumtion income elasticity of demand and consumtion structure changes, analyzed China s rural residents consumtion structure and its evolution in across time resectively. Hu, Tian and Xia [7] emirically analyzed of China s fiscal exenditure for suorting agriculture s imact on rural residents consumtion by using the annual time series data from 1978 to Yu and Zhang [8] created a multivariate rediction model based on the combination of Lasso method and BP neural network, and rediction of China s urban and rural resident s consumtion exenditure. In this aer, we are using R statistical software s Lars ackages and Msgs ackages, modeling and rojections for rural residents consumtion exenditure about China by using the Lasso method and the Adative Lasso resectively. 2. Lasso Method and Adative Lasso Method Suosing this aer has model data ( Xi 1, Xi2,, Xi; Yi ), considering usually linear model: Yi = β0 + β, 1 j Xij + εi (1) Which i = 1, 2,, n, ε1, ε2,, εn is indeendent identically distributed, and E [ ε i ] = 0. Usually write model (1) as the following form: Y = Xβ + ε, (2) Which Y n 1 as the resonse vector, X n as the indeendent variable matrix, β 0 as constant, β 1 as the coefficient vector, assume that the data has been standardized, T remember β = ( β1, β2,, β ), so Lasso method to estimate is defined as: 2 ( ) ( ) ( ˆ lasso ˆ lasso n β0, β ) = argmin( ) ( Y ) 0, i 1 i 0 X β β β j 1 ijβ = = j β j 1 j s = j= (3) 1167

3 here s 0 is the enalty arameter. The otimal solution of (3) is called the Lasso solution; the entire Lasso solution can be obtained by changing the s values, at this time, this aer uses k-fold CV and Mallows C criteria to choose the best model. k-fold CV is a common method of evaluation model, it roughly ut all of the observation data divided into k equal arts, and then take turns to use one of the k 1 arts for the training set, used to fitting data, the remaining art is a test set, totally calculating k times, get the k index of the mean square error of fitting test set, do an average, then reeat all of the stes of the above, then select the model of the minimum average mean square error. C criteria is also a standard which used to assess a regression model, if select a indeendent variables from the k indeendent variables to involved in regression, then C criteria is defined as: SSE C = n+ 2 2 S SSE Y Y ( ) n = i= 1 i i Therefore, this aer can choose a model with minimum C. Lasso method selects variables, at the same time, it is good in estimates the unknown arameters, can solve the multicollinearity roblems that exist in the model better, esecially the high-dimensional data rocessing. Adative Lasso method to estimate is defined as: In Tye (5), 2 ( ) ( ) ( ˆ alasso ˆ alasso n β0, β ) = argmin( ) ( Y ) 0, i 1 i 0 X β β β j 1 ijβ = = j ( ˆ βi ) j= 1 ω β s i j 1 ω = as the weight coefficient, and γ > 0 as a adjustment ara- i γ meters, ˆi β is the initial estimate about the arametersi, ˆi β can use the least squares estimate, ridge estimate and Lasso estimates. The otimal solution of (5) is called Adative Lasso solution, all Adative Lasso solution can be obtained by changing the svalues, at this moment, Mallows C criteria, AIC criteria, GCV criteria and BIC criteria can be used to choose the best model [9]. The Adative Lasso method using different weight coefficient, with a smaller weight unish the variable which regression coefficient is larger, with a larger weight unish the variable which regression coefficient is smaller, makes the selected variables more accurately. Due to Lasso method use the same weight of all coefficient, and the Adative Lasso methods based on different variables given different weights, with a smaller weight unish the variable which regression coefficient is larger, with a larger weight unish the variable which regression coefficient is smaller, imroved the Lasso method in variable selection, which cannot meet the model selection of consistency and arameter estimation lack of convergence seed to n, makes selected variables more accurate. For this urose, this article resectively building the forecasting model based on the method of Lasso and Adative Lasso, to forecast the rural residents consumtion exenditure in China. 2 (4) (5) 1168

4 3. The Emirical Analysis 3.1. Variable Selection and Data Sources In this aer, on the basis of the theory of economics and the research of Yu and Zhang [8], 16 variables which influenced rural residents consumtion exenditure(y) are selected. Name of the 16 variables secific in Table 1. In this article, the deendency ratio data was from Statistical Yearbook of China Poulation. The interest rate data was from the website of the eole s bank of China, interest rates will be subject to the one-year rate stiulated by the central bank, if there are multile interest rate in a year, then use weighted average, the weight of the interest rate used in accounted for the roortion of 12 months. Other variable data are from eriods, China Statistical Yearbook The Prediction of Rural Residents Consumtion Exenditure Model Based on Lasso Method From Figure 1, this article just need 24 stes to get all the Lasso solution, when arameters s = 1, all variables into the model. Because of the value of k-fold CV is bigger Table 1. The set of indeendent variables. variable meaning variable meaning x 1 Residents disosable income x 9 Young deendency ratio x 2 GDP growth rate x 10 Education situation x 3 Inflation x 11 Sending on social security x 4 The first industrial outut value x 12 Emloyment figure x 5 The tertiary industry outut value x 13 Income distribution ga x 6 The annual fixed asset investment x 14 Sending habits x 7 The interest rate x 15 Highway mileage x 8 oulation x 16 Post and telecommunications business Figure 1. The variable selection ste with Lasso method. 1169

5 than the value of C criteria, so this aer selects the most otimal Lasso solution according to Mallows C criteria. This article gets minimum C value when ste is 20, then the model is otimal, in the end, this aer chooses 11 variables such as x1, x2, x3, x5, x6, x9, x12, x13, x14, x15, x 16. From the result, inflation is the most imortant factors influencing the rural residents consumtion exenditure; And highway mileage, Residents disosable income, the annual fixed assets investment and the tertiary industry outut value have a ositive imact on rural residents consumer sending; But the GDP growth rate, young deendency ratio, emloyment figure, income distribution ga, consumtion habits and ost and telecommunications business has a negative effect on the rural residents consumtion exenditure; Other factors are not significant in affect the rural residents consumtion exenditure, are elected to the model. LASSO model: y = x 0.131x x x x x x x x x x On this basis, this aer redicts China rural residents consumer sending from 2008 to 2014, forecasting results are shown in Table 2. From Table 2, the forecasts of rural residents consumtion exenditure is widesread undervalued, but from the ersective of the relative error, the estimate result is more stable, the rediction error decreases year by year, roves the estimate effect of the model of Lasso method is good The Prediction of Rural Residents Consumtion Exenditure Model Based on Adative Lasso Method Through the use of Adative Lasso method (see Figure 2), this article can redict the rural residents consumtion exenditure. Because of the value of Mallows C criteria, AIC criteria, GCV criteria and BIC criteria is 9.767, 10.11, 11.14, 8.263, the BIC value is minimum which can fitting the model best, so this aer chooses BIC criteria to modify the model and finally selects 13 variables such as x1, x2, x3, x6, x7, x9, x10, x11, x12, x13, x14, x15, x 16. From the result, inflation is the most imortant factors influencing the rural residents consumtion exenditure; And highway mileage, the interest rates, residents disosable income, exending on social security and the annual fixed assets investment are have a ositive imact on rural residents Table 2. Rural residents consumer sending redictions based on the lasso method. year Y (billion) Predicted value (billion) relative error (6) 1170

6 consumtion exenditure; But the GDP growth rate, young deendency ratio, education, emloyment figure, income distribution ga, consumtion habits and ost and telecommunications business are have a negative effect on the rural residents consumtion exenditure; Other factors are not significant in affect the rural residents consumtion exenditure, are not elected to the model. Adative LASSO model: y = 0.728x x x x x x x10 (7) x x x 0.064x x x On this basis, this aer redicts China rural residents consumer sending from 2008 to 2014, forecasting results are shown in Table 3. From Table 3, the rediction of rural residents consumtion exenditure is also has the henomenon of widesread undervalued, but from the ersective of the relative error, the estimate result is more stable, the rediction error decreases year by year, and the most imortant oint is that the rediction result of the Adative Lasso method is more close to the real value than the Lasso method, roves the estimate effect of the model of Adative Lasso method is better than the Lasso method. Figure 2. The variable selection with Adative Lasso method. Table 3. Rural residents consumer sending redictions based on the Adative Lasso method. year Y (billion) Predicted value (billion) relative error

7 4. Conclusions and Recommendations Based on the Lasso method and the Adative Lasso method to construct redictive model of rural residents consumtion exenditure in China resectively, inflation is the most imortant factors influencing the rural residents consumtion exenditure; and residents disosable income, inflation, the annual fixed asset investment and highway mileage factors are all have ositive effects on the two models; and the GDP growth rate, oulation, emloyment figure, income distribution ga, consumtion habits and ost and telecommunications business all have a reverse effect on the two models. To this end, this article uts forward the following suggestions: 1) Control the level of inflation in China. Inflation has a great influence on rural residents life. On the one hand, inflation makes the rice of agricultural and sideline roducts increase, and increase the income of farmers. On the other hand, inflation makes the rural consumer goods and services rices increase and the farmers real income down. Thus, the inflation can be controlled by using these ways, such as the tightening of monetary olicy, fiscal olicy and income olicy, ositive suly olicy, currency reform and other measures to restrain inflation. 2) Raising the income level of rural residents. Raise the income level of rural residents, so that they will have more consumtion. The government can imlement active emloyment olicy, rovide more emloyment latform, give more job training for rural surlus labor about mount guard, exand the source of raising income, and rovide modest fiscal olicy of subsidies for agricultural roduction to increase the income of rural residents. References [1] Zhao, Y.D. (2013) The Influence of Local Fiscal Exenditure on Rural Residents Consumtion. Public Finance Research, 5, [2] Liu, Z.Z. and Wu, F. (2014) The Livelihood Process of Local Government Fiscal Exenditure and Rural Consumtion Theoretical Analysis and Emirical Test Based on Total and Category Exenditure. The Theory and Practice of Finance and Economics, 1, [3] Chu, D.Y. and Yan, W. (2009) Local Government Exenditure and Rural Residents Consumtion Demand: An Emirical Analysis Using Panel Data of Provinces from 1998 to Statistical Research, 8, [4] Chen, D. and Liu, J.D. (2013) Rural Credit Imact on Rural Residents Consumtion Based on State Sace Model and the Mediation Effect Test of the Dynamic Analysis for a Long Time. Journal of Financial Research, 6, [5] Li, X.J. (2013) Dynamic Effect Analysis on Rural Residential Consumtion and Exenditure for Social Security. Collected Essays on Finance and Economics, 4, [6] Wen, T. and Meng, Z.L. (2012) Rural Residents Consumtion Structure Evolution Research in China. Journal of Agrotechnical Economics, 7, [7] Hu, D.L., Tian, K. and Xia, J.C. (2013) The Influence on Rural Residents Consumtion of China s Fiscal Exenditure for Suorting Agriculture Emirical Analysis and Policy Recommendations. Public Finance Research, 1, [8] Yu, S.H. and Zhang, J. (2016) The Study on Prediction of Residents Consumtion Exenditure based on Lasso and BP Neural Network. The Theory and Practice of Finance and 1172

8 Economics, 37, [9] Wu, X.Z. (2012) Comlex Data Statistical Method Based on the Alication of R. China Renmin University Press, Beijing. Submit or recommend next manuscrit to SCIRP and we will rovide best service for you: Acceting re-submission inquiries through , Facebook, LinkedIn, Twitter, etc. A wide selection of journals (inclusive of 9 subjects, more than 200 journals) Providing 24-hour high-quality service User-friendly online submission system Fair and swift eer-review system Efficient tyesetting and roofreading rocedure Dislay of the result of downloads and visits, as well as the number of cited articles Maximum dissemination of your research work Submit your manuscrit at: htt://aersubmission.scir.org/ Or contact ojs@scir.org 1173