An Analysis of the Trend of China’s Export Trade to the USA Based on R Language

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.

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.

Analysis of the Current Situation of China-US Trade
According to Figure   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).

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: 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.  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.

Model Prediction and Analysis
In order to more directly reflect

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