Agriculture trade
Due to climate change and the rapid population growth, food security becomes a pressing issue, especially for tropical developing countries. Tropical countries suffer from shortage of food and poverty, making them the most vulnerable. Changes in climate system impact agriculture product yield, as well as their demand. Studies suggest that tropical regions are predicted to have large decrease in agriculture production in 2050. The total values of importing agriculture products over tropical countries will largely increase as well. Who are exporting agriculture and who are importing? How the trade will change? I investigate the agriculture trade from 1986-2019 and predict the change out to 2035.
I used the 1986-2019 food and agricultural trade dataset from here. There are a total of 462 different agriculture products and 222 reported countries.
During 1986-2019, the total exported values of food preparation necessities have the highest world total export values of $10239.17 billion. Wheat's total export values rank the second highest at $4278.24 billion.
In 2015-2019, US, Netherlands, Brazil, Germany, France and China take up about 37.1% of the world's total export value. U.S., Netherlands, Germany, France and China took up about ~33.4% of the world's total import value. Countries that exported the most agricultural products tend to be countries that imported the most.
I used ARIMA (Auto-regressive integrated moving average) model to predict the agriculture trade. ARMIA is a popular and widely used statistical method for time series forecasting. It explains time series for a given frequency or lag and also predicts future values. When running a linear regression, the assumption is that all of the observations are independent of each other. The Augmented Dickey-Fuller test is used to evaluate the stationarity of data. Only data with more than 17-year records (>=60% are available). Three-year rolling averages of the data are used. The prediction is made out to 2035.
I created a pipeline of ARMIA model predictions. I use US soybean export value and quantity as an example.
From 2015 - 2019, the US was the country with the largest agriculture product export value, which took up ~10.4% of the world's total agricultural product exported value.
US soybean exported value and quantity increased from 1985 to 2012 and decreased slightly onwards. The US exported rice values and quantity increased from 1985 to 2011 and decreased onwards. The model is able to reproduce the US soybean and rice exported value and quantity evolutions, as shown in the figure below.
Both US soybean and rice exported value is predicted to decrease, consistent with previous studies. However, this is based on one model and annual data. The predicted result has a wide confidence interval, adding to large uncertainty. Data with higher frequency would be more helpful. Models, e.g., LSTM and NN, require much more dense data.
