





































Microsoft Word - ELP-V2N2-p229


Economics, Law and Policy 
ISSN 2576-2060 (Print) ISSN 2576-2052 (Online) 

Vol. 2, No. 2, 2019 
www.scholink.org/ojs/index.php/elp 

229 
 

Original Paper 

An Analysis of the Trend of China’s Export Trade to the USA 

Based on R Language 

LUO Shuang-lin1, ZHOU Min1* & TANG Chong-tai2 
1 School of Economics and Trade, Hunan University of Technology and Business, HuNan, China 
2 CARFAX, New York 20120, USA 

* ZHOU Min, School of Economics and Trade, Hunan University of Technology and Business, HuNan 

41205, China 

 

Received: November 17, 2019 Accepted: November 25, 2019  Online Published: November 29, 2019 

doi:10.22158/elp.v2n2p229              URL: http://dx.doi.org/10.22158/elp.v2n2p229 

 

Abstract 

After President Trump came to power, in order to change the “imbalance” between China and US 

trade, he launched a trade war with China, which led to increase uncertainty in China-US trade and 

increased export volatility. Based on R language environment, this paper compares the advantages and 

disadvantages of seasonal ARIMA (p, d, q) model and double-index ETS (A, N, A) model in short-term 

forecast of China’s total export value to the United States. Then, the double-index ETS (A, N, A) model 

is selected to predict the trend of China’s export trade to the United States in the months of 2009-2020. 

The forecast results show that China’s export to the United States has seasonal characteristics. The 

export fluctuation is smaller than that in 2018, but the total value of exports has decreased significantly. 

Finally, some suggestions are put forward. 

Keywords 

seasonal ARIMA (p, d, q) model, double-index ETS (A, N, A) model, total export value, forecast 

 

1. Introduction 

Against the background of the trade war between China and the United States, some scholars 

conducted studies on the prediction of the development trend of sino-American trade through CGE 

model, game model and tv-stvar model, and the results showed that bilateral trade, GDP growth and 

social welfare would be affected. In recent years, many scholars have used relevant models to predict 

and analyze social phenomena, industrial structure, economic growth and scientific and technological 

innovation in R language environment. R language is an open source data analysis solution with 

powerful statistical calculation and graphic display design capabilities. Maria Brigida Ferraro (2015) 



www.scholink.org/ojs/index.php/elp                   Economics, Law and Policy                        Vol. 2, No. 2, 2019 

230 
Published by SCHOLINK INC. 

used R language to conduct cluster analysis of big data on the main factors affecting obesity. Xue xin 

(2019) predicted and analyzed the exchange rate based on R language neural network. Wu mingxin 

(2017) applied R language into the field of auditing and conducted big data processing on auditing 

finance. Zhang zhe (2013) established a generalized time series model and a regression-time series 

model under the R language environment to forecast and analyze China’s export trade volume and tax 

revenue. Li ping (2010) used R language statistical analysis software to conduct quantitative analysis of 

high-tech industry. 

The innovation of this paper is that based on R language environment, seasonal ARIMA (p, d, q) model 

and double index ETS (A, N, A) model are applied in the field of international trade to predict the 

short-term trend of sino-American trade value, so as to study the impact of sino-American trade war 

and provide guidance for dealing with trade war. 

 

2. Analysis of the Current Situation of China-US Trade 

According to Figure 1, it can be seen that during the period from 2001 to 2018, China’s total imports 

and exports to the United States showed a steady growth every year except for a decrease in 2009 and 

2016. China has always maintained a large surplus. Except for a slight decrease in the surplus in 2009, 

the surplus in the other years has basically become stable. 

 

 
Figure 1. Total Imports and Exports from China to the United States (Unit: thousands of dollars) 

Source: “China Statistical Yearbook” & “China Customs Statistics in 2018”. 

 

For a long time, the US has been China’s largest export market, accounting for about 20% of China’s 

exports. In 2018, the US tariff protection measures against China have affected the proportion of us 

exports in China to some extent. However, in 2018, the trade value between China and the US reached 

us $63351,941, an increase of 8.54% year on year. The total value of exports was 47,8423,17 million 

us dollars, up 11.33% year on year. The total value of imports was us $155,509,623. Year-on-year 

growth of 0.75%; The surplus reached $323.32694 million, an increase of 17.24%. Judging from the 



www.scholink.org/ojs/index.php/elp                   Economics, Law and Policy                        Vol. 2, No. 2, 2019 

231 
Published by SCHOLINK INC. 

data, the impact of the trade war between China and the United States on China in 2018 is limited. 

 

3. Forecast and Analysis of China’s Export to the United States Based on R Language 

According to China’s general administration of customs issued in June 2014-June 2019, the data of 

Chinese exports to the United States will be 2018 before and after data is divided into training and 

testing, and use the double index of ETS (A, N, A) model and seasonal ARIMA (p, d, q) model to 

predict respectively, and comparing the prediction results of two models, select the best prediction 

results. 

3.1 Model Framework and Principles 

3.1.1 ETS (A, N, A) Model 

Exponential model is the most common model used to predict the future value of time series. The idea 

of exponential smoothing method is derived from the improvement of moving average prediction 

method, which comprehensively uses adjacent values, overall trend and seasonality to conduct 

prediction analysis, but gives more weight to adjacent values. This kind of model is proved to be good 

for short-term prediction in practice. ETS function in forecast package in R language can fit the index 

model. Among them, ETS function can be divided into three index models: ses, holt, and hw, 

respectively. Ses, holt, and hw functions are convenient packages of ETS function, and the functions 

have preset parameter values. After data input, the best model is selected as double exponential model. 

General ETS function is as follows: 

                           (1) 

Where ts is the timing sequence to be analyzed, and there are three letters defining the model. The first 

letter represents the error term, the second letter represents the trend term, and the third letter represents 

the seasonal term. Optional letters include: additive model (A), multiply model (M), none (N), 

automatic selection (Z). 

3.1.2 ARIMA Model 

ARIMA model (autoregressive integrated moving average moving average mode) is a commonly used 

stochastic time series model with high accuracy for short-term prediction. It was founded by American 

statisticians box and Jenkins with the following basic ideas: Some time series are a set of random 

variables dependent on time t, and the change of the whole series has certain regularity. By establishing 

a mathematical model and analyzing and studying, the structure and characteristics of time series can 

be essentially understood, and the most effective prediction results can be obtained. 

ARIMA model is made up of autoregressive model AR (P), MA (q) and autocorrelation model poor 

score (d) of three parts, so that half of the ARIMA model has the characteristics of the autoregressive 

and moving average characteristics of the process, and through poor score (d) let originally 

non-stationary time series become stable, improve the accuracy of the subsequent forecast. 

The general expression of ARIMA (p, d, q) model is: 



www.scholink.org/ojs/index.php/elp                   Economics, Law and Policy                        Vol. 2, No. 2, 2019 

232 
Published by SCHOLINK INC. 

      (2) 

In general, if the data is a time series with seasonal effects, a product seasonal model is required 

 for simulation, P and Q are the order of seasonal autoregression and 

moving average, D is the order of seasonal difference, and S is the seasonal cycle. 

3.2 Data Basis (Monthly Data) Analysis 

 

 

Figure 2. Total Value of China’s Exports to the United States in R Language Environment (Unit: 

ten thousand yuan) 

 

 

Figure 3. Autocorrelation of Current Values from June 2014 to June 2019.06 

 

The data are mainly from the General Administration of Customs of China, with a total of 61 sample 

data. Using software is R, make in time for the horizontal axis, exports for the longitudinal axis of the 

sequence diagram, as shown in Figure 3, March 2015, in February 2016 and February 2017, in March 

2018 and February 2019 was the lowest each quarter, in October 2014, in September 2015 and 

December 2016 and November 2017 and November 2018 were appeared in the annual peak, the whole, 

China’s exports to the United States trade has obvious seasonal characteristic, present the total cost of 

the export trade fluctuation trend of rising and it peaked in 2018-2019 at a six-year high. At the same 



www.scholink.org/ojs/index.php/elp                   Economics, Law and Policy                        Vol. 2, No. 2, 2019 

233 
Published by SCHOLINK INC. 

time, as shown in Figure 4, the time series of total export trade value shows a certain tardiness in 

autocorrelation and obvious seasonality, i.e., non-stationary. Therefore, the data belongs to 

non-stationary time series. In order to eliminate the growth trend, difference is needed. In order to 

eliminate the seasonal trend, further seasonal difference is needed. 

3.3 Model Prediction and Analysis 

In order to more directly reflect the impact of us sanctions on China’s export trade with the us in the 

two years from 2018 to 2019, this paper selected China’s monthly export trade data from June 2014 to 

June 2019 for time series analysis,  model and  

model mainly uses the short-term prediction of China’s export trade value to determine whether the 

total value of China’s export to the United States will decrease significantly or increase after the United 

States imposes sanctions on China. 

3.3.1 Grouping 

The data of China’s total export to the United States from June 2014 to June 2019 are divided into two 

groups: the first group is the total export value data before December 2017, named Training; the second 

group is the total export value data after January 2018, named Testing. In 2018 and January-June 2019, 

when the trade war between China and the United States was at its peak, a division can improve the 

accuracy of the model ( model and model) and the 

prediction of the total value of China’s exports to the United States. 

3.3.2 Make Comparative Analysis 

The degree of fitting of the model can be determined according to the information criterion. Several 

information criteria can be used, such as red information criteria, AIC information criteria, AIC revised 

AICc information criteria, and bayesian information criteria, BIC information criteria. According to 

Table 4, according to the ARIMA model presented by R and the AIC information criterion, AICc 

information criterion and BIC information criterion of ETS (A, N, A) model, the fitting degree of 

ARIMA model is better than ETS (A, N, A) model. 

 

Table 1. Comparison Table of Fitting Degree 

 AIC AICc BIC 

ARIMA（0,1,1）（0,1,0）[12]model 954.15 954.59 954.59 

ETS（A,N,A）model 1405.355 1423.132 1431.773 

 

 

 

 

 



www.scholink.org/ojs/index.php/elp                   Economics, Law and Policy                        Vol. 2, No. 2, 2019 

234 
Published by SCHOLINK INC. 

Table 2. ARIMA (0, 1, 1) (0, 1, 0)[12] Model Error Correlation Values 

 RMSE MAE MAPE MASE 

Training set 1508882 1074948 4.997333 0.5105552 

Test set 3829596 3321557 14.615813 1.5775997 

 

Table 3. ETS (A, N, A) Model Error Correlation Values 

 RMSE MAE MAPE MASE 

Training set 1344986 1021103 4.814283 0.4849811 

Test set 2304295 1842249 7.454881 0.8749907 

 

RMSE represents root mean square error, MAE represents mean absolute error, MAPE represents 

mean absolute percentage error and MASE represents mean absolute scale error. These errors are used 

to measure the error degree of prediction results of ARIMA (0, 1, 1) (0, 1, 0) [12] model and ETS (A, N, 

A) model. Through intuitive comparison Tables 5 and 6 error numerical, ETS (A, N, A) model is far 

less than the error of the model ARIMA (0, 1, 1) (0, 1, 0)[12]  model, and the sample data of the 

average value of 231.0439665 billion yuan, ETS (A, N, A) model RMSE, MAE, and two groups of 

data under MAPE compared with the average of the sample data is small, at the same time, the MASE 

is less than 1, so from the Angle of error, ETS (A, N, A) model has better prediction results than 

ARIMA (0, 1, 1) (0, 1, 0)[12] model. 

To sum up, the ARIMA (0, 1, 1) (0, 0)[12]  model compared with ETS (A, N, A) model is more 

complex, the fitting degree of ARIMA (0, 1, 1) (0, 0)[12] model occupy A certain advantage, but in the 

short-term forecast, in the case of data presents obvious seasonal, ETS (A, N, A) model prediction error 

rate is lower, so choose the ETS (A, N, A) model in the prediction of the trend for China’s exports to 

the United States. 

3.3.3 ETS (A, N, A) Model Testing 

Ljung-box test has better sample nature (that is, more effective in statistical sense) than box-pierce test. 

According to Ljung-box test, under ETS (A, N, A) model, p-value=0.07373>0.05. Therefore, there is 

no sufficient reason to reject the null hypothesis and the residuals should be considered independent of 

each other. 

 

 

 

 



www.scholink.org/ojs/index.php/elp                   Economics, Law and Policy                        Vol. 2, No. 2, 2019 

235 
Published by SCHOLINK INC. 

3.3.4 Predict Results 

 

 

Figure 4. Predicted Value of ETS (A, N, A) Model—Line Graph 

 

Run the monthly data of China’s export value to the United States from June 2014 to June 2019 in the 

forecast package in R, and get the predicted value line chart of ETS (A, N, A) model in Figure 4. The 

blue line is the point estimate, and the light gray and dark gray areas represent 95% and 80% 

confidence intervals respectively. As can be seen from the figure, the total value of China’s exports to 

the United States will decline significantly from 2019 to 2020, roughly approaching the level of 

China’s total value of exports to the United States in 2017, because the United States imposes a tariff of 

up to 25% on imports of Chinese products. The most important thing is that it may change the 

long-term stable growth trend of China’s exports to the United States. The predicted values and actual 

values are shown in Table 7 and Table 8 (the influence of holiday factors in February 2019 is obvious, 

so it is removed from the error estimation). 

 

Table 4. Error between Fitting Value and Actual Value from July 2018 to June 2019 

Time The actual value The fitting values Error rate（%） 

2018.07 26766530 26025160 -2.77% 

2018.08 29481821 27034012 -8.30% 

2018.09 31931363 28011028 -12.28% 

2018.10 29319143 26483546 -9.67% 

2018.11 31885204 27145170 -14.87% 



www.scholink.org/ojs/index.php/elp                   Economics, Law and Policy                        Vol. 2, No. 2, 2019 

236 
Published by SCHOLINK INC. 

2018.12 27942025 27339536 -2.16% 

2019.01 25211055 25973371 3.02% 

2019.02 15481382 19555027 26.31% 

2019.03 21498523 21749741 1.17% 

2019.04 21074682 23838726 13.12% 

2019.05 25296472 25320438 0.09% 

2019.06 26826670 25468902 -5.06% 

Mean absolute error rate (except 2019.02) 6.59% 

 

Table 5. Predicted Values of ETS (A, N, A) Model 

Time Piont forecast 
Interval estimate of 80% 

confidence level 

Interval estimate of 95% 

confidence level 

2019.07 26012197 (23621427,28402967) (22355830,29668564) 

2019.08 27496730 (24928483,30064976) (23568936,31424524) 

2019.09 28874115 (26139887,31608342) (24692475,33055755) 

2019.10 27073494 (24182801,29964188) (22652560,31494428) 

2019.11 28224219 (25185104,31263333) (23576294,32872143) 

2019.12 27389672 (24209055,30570290) (22525338,32254007) 

2020.01 25847560 (22531472,29163647) (20776042,30919078) 

2020.02 19773579 (16327343,23219816) (14503016,25044143) 

2020.03 21224636 (17652990,24796282) (15762275,26686997) 

2020.04 23540110 (19847311,27232910) (17892462,29187759) 

2020.05 25267428 (21457327,29077530) (19440381,31094476) 

2020.06 25591927 (21667971,29515883) (19590754,31593100) 

 

4. Analysis of Prediction Results 

According to the above empirical evidence and prediction, the summary and analysis are as follows: 

First of all, from June 2014-June 2019 real data, months China’s exports to the United States trade 

gross value of the current period has obvious seasonal fluctuations, in February or march of each year 

will be the lowest of the year, mainly due to the effect of China about 2 month every year the Spring 

Festival holiday, after the Spring Festival exports gradually recover, generally around November 

reached the highest value. Although there were frictions between China and the United States during 

this period, and the United States frequently used the “double-countervailing” investigation, “337” 

investigation and technical barriers to trade to restrict the import of Chinese products, China’s export 

trade with the United States was still on the rise on the whole, and the surplus was also expanding. 

Even in 2018, when the United States started a trade war, China’s exports to the United States 



www.scholink.org/ojs/index.php/elp                   Economics, Law and Policy                        Vol. 2, No. 2, 2019 

237 
Published by SCHOLINK INC. 

continued to grow. 

Secondly, by using the model of A ETS (A, N, A) it is concluded that the July 2018-June 2019 month 

predicted values and the real value of the mean absolute error rate was 6.59% (remove) in February 

2019, error rate is small, relatively than that of ARIMA (0, 1, 1) (0, 0)[12]  model, especially in the case 

of data with the seasonal characteristics, ETS (A, N, A) model to predict trend has more advantages. 

Finally, the peak value of the predicted value in November 2019 is lower than the peak value of the 

actual value in 2018, and the minimum value of the predicted value in February 2020 is higher than 

that in 2019, indicating that the fluctuation range of China’s exports to the United States during the 

forecast period is smaller than that in 2018, and the total value of exports drops to the level of 2017. 

The comparison between the predicted value and the real value also shows that the trade war between 

China and the United States has little impact on 2018. However, due to the long-term trend of the trade 

war between China and the United States, China’s exports to the United States will be restrained to a 

certain extent during the period from 2019 to 2020, and the prospect of china-us trade is not optimistic. 

 

5. Suggestions 

5.1 We Will Continue to Open Up and Expand Multilateral Economic and Trade Cooperation 

China should continue to open up to the outside world, strengthen the supply-side structural reform in 

foreign trade, and improve the tax reduction and exemption policies and measures to support small and 

medium-sized foreign trade enterprises. We will promote the development of the One Belt and One 

Road market and the construction of bilateral and multilateral free trade areas, further enhance trade 

facilitation and create a favorable new environment for opening up the economy, so as to offset the 

impact of the reduction in exports to the United States. 

5.2 We Continue to Negotiate with the United States on the Basis of Ensuring Bottom-Line Thinking 

We should stick to the bottom-line principle and the principles of equality, justice and mutual benefit in 

communication and negotiation with the us, strive for an early conclusion of an agreement conducive to 

the long-term economic development of the two countries and avoid further deterioration of China-US 

trade. The recent breakdown of trade talks between China and the us and trump’s threat to impose 

additional tariff rates have further increased the uncertainty of china-us trade, which will also seriously 

affect the confidence of the export market in the us in the future. 

5.3 We Will Strengthen Enterprises’ Awareness of Risk Prevention and Intellectual Property Protection, 

and Foster New International Competitive Advantages 

Faced with the trade war between China and the United States, enterprises should be aware of risks, 

adjust their export markets in a timely manner, and avoid risks of tariffs and exchange rates. Properly 

handle the industrial cooperation with relevant American enterprises. At the same time, we need to 

strengthen ipr protection and create a sound business environment in China. We should encourage and 

intensify enterprise innovation, improve the structure of export products, and cultivate competitive 

advantages in the quality, technology, brand and service of export products. In particular, we should 



www.scholink.org/ojs/index.php/elp                   Economics, Law and Policy                        Vol. 2, No. 2, 2019 

238 
Published by SCHOLINK INC. 

avoid focusing on 16 categories of products for export to the United States and implement 

differentiation strategy. 

 

References 

Anthony, W. C., Jim, C., & V. Reddy, D. (2019). The US-China trade war: Dominance of trade or 

technology? Applied Economics Letters, 2019, 1-6. 

https://doi.org/10.1080/13504851.2019.1646860 

Box, G. E. P. et al. (2015). Time series analysis: Forecasting and control. John Wiley & Sons. 

Guijun, L., Fei, W., & Jiansuo, P. (2018). Global value chain perspective of US–China trade and 

employment. World Economy, 2018, 41. 

Hopewell, K. (n.d.). US-China conflict in global trade governance: The new politics of agricultural 

subsidies at the WTO. Review of International Political Economy, 1-25. 

Hughes, L., & Meckling, J. (n.d.). The politics of renewable energy trade: The US-China solar dispute. 

Energy Policy, 105, 256-262. https://doi.org/10.1016/j.enpol.2017.02.044 

Ka, Z. (2013). High Stakes: US-China Trade Disputes under the World Trade Organization (WTO). 

Asian Journal of Social Science, 41(3-4), 352-380. https://doi.org/10.1163/15685314-12341309 

Lin, J. Y., & Wang, X. (2018). Trump economics and China–US trade imbalances. Journal of Policy 

Modeling, 40(3), 579-600. https://doi.org/10.1016/j.jpolmod.2018.03.009 

Maria, B. F., & Paolo, G. (2015). A toolbox for fuzzy clustering using the R programming language. 

Fuzzy Sets and Systems, 279, 1-16. https://doi.org/10.1016/j.fss.2015.05.001 

Tung, R. L. (1982). U.S.-China Trade Negotiations: Practices, Procedures and Outcomes. Journal of 

International Business Studies, 13(2), 25-37. https://doi.org/10.1057/palgrave.jibs.8490548 

Venables, W. N., & Smith, D. M. (2006). R Development Core Team. 

Zeng, K. (2013). High stakes: United States-China trade disputes under the World Trade Organization. 

International Relations of the Asia-Pacific, 13(1), 33-63. https://doi.org/10.1093/irap/lcs014 

 


