Overview

This note sets out how we retrieve US export data from the US Census(U.S. Census Bureau, 2025) trade database. As an example we show the picture for US polypropylene exports - a time series that carries the code 3902100000. The data is expressed in US$.

Method

The commodity type series (from the HTS code list) and all dates from the census website were selected. That selection was downloaded as a csv file and read into R. And then a monthly time series was created by parsing the raw csv data.

The raw file contained 628 observations in 3 variables: the series, time in YYYY-MM format, and, a value field. But closer inspection of the raw data showed two series (390210 and 3902100000) had been downloaded and there were records for both annual and monthly data.
So we filtered the data for the relevant series, parsed the month dates, and made the value data numeric.
The saved time series was now in an xts format that could be visualized

Line chart description

trend

From January 2002-June 2008 the series shows an overall upward trend. Exports start at around $60 million and generally increase peaking in June 2008 at $264 million. There is then a sharp correction following the 2008 financial crisis that saw the series bottom at just over $100 million. The index then oscillates between a range of $100-200 million. In the covid crisis of 2020 monthly exports dropped below $100 million but then rebounded to 2022 level above $200 million. Since then the value has fallen and as of febuary 2025 stood at $141 million.

variability

First Differences

The plot shows the absolute changes in export values (in USD millions). Large spikes (e.g., a $40M increase or decrease) indicate periods of high variability. Inspection from the chart suggest variability tends to lie around plus or minus $20 million.

Percentage Changes

This plot highlights relative changes. For example, a -40% drop in June 2008 ($264 million) to Dec 2008 ($100 million). Average variability by inspection appears to generally lie plus or minus 10%.

Rolling Standard Deviation

The 3-month rolling SD plot shows how variability changes over time. Higher values (e.g., $40M or more) indicate periods of greater month-to-month fluctuation, such as during 2008 (possibly due to the financial crisis). Since the covid pandemic the trend in variability had been reducing.

Boxplot by Month

This shows that May and June tend to be months of greatest exports. Higher variability in exports (reflected in wider box and so larger interquartile range) is found in august and november. Months of least variability tend to be february and december.

Random Component

The random component from decomposition isolates variability not explained by trend or seasonality. Large fluctuations (e.g., ±$30M) indicate significant month-to-month variability due to external factors.

ACF

The ACF plot helps identify patterns in the variability. And here shows some significant spikes at lag 12 that could suggest yearly seasonality in the variability.

Range

The monthy export values range from $60 million in January 2002 to $264 million in June 2008.

Recent 12 months

In the past 12 months from February 2024 through February 2025 exports rose from $152 million to peak at $191 million in August 2024 before then dropping to the latest data point of $141 million.

Table of statistics

Statistic	Value
Mean	146989530
Standard Deviation	38142509
Minimum	60647711
Maximum	264133051
1st Quartile	124235254
Median	144062737
3rd Quartile	165271131

Mean ($146,989,530): The average export value is approximately 147 million USD.
Standard Deviation ($38,142,509): Measures the amount of variation or dispersion in the dataset. A higher standard deviation indicates that the values are more spread out from the mean. Here, it suggests moderate variability in export values.
Minimum ($60,647,711): This represents the lowest recorded export value, which is approximately 61 million USD as of January 2002.
Maximum ($264,133,051): Highest month of exports registered at approximately 264 million USD in July 2008.
1st Quartile ($124,235,254) or 25th percentile: Indicates that 25% of the data points are below approximately $124 million.
Median ($144,062,737): The middle value of the dataset when ordered from smallest to largest. It separates the higher 50% of the data from the lower 50%. And the median export value is approximately 144 million USD. Because the median is below the mean it suggests the distribution of monthly values is skewed to the right (positively skewed). That implies the tail on the right side of the distribution is longer and fatter than the left side. and means there are more higher values that pull the mean up while the median remains closer to the center of the data. In the context of trade it suggests that most export transactions are of moderate value but there are some transactions with exceptionally high values that drive up the average.
3rd Quartile ($165,271,131) or 75th percentile: indicates that 75% of the data points are below this value. It provides insight into the upper range of export values.
Overall Interpretation: The data set appears to be moderately spread out, with a significant range between the minimum and maximum monthly values (from $60-264million). The mean and median are relatively close ($146-144million), suggesting a fairly symmetric distribution of export values.

Seasonality

The decomposition of the export time series shows observed data, trend, seasonal, and random (or residual) components. The charts are plotted over a time period from around 2002 to 2025.
Observed (Top Chart): This chart represents the raw time series data as it was recorded. The y-axis is labeled “observed” with monthly values ranging from approximately $60 mil to $240 mil. The data shows fluctuations over time with noticeable peaks and troughs. There appears to be a general upward trend until 2008, followed by a decline, and then a more stable but fluctuating pattern from 2015 onward.
Trend (Second Chart): This chart isolates the long-term trend in the data, smoothing out short-term fluctuations and seasonal effects. The y-axis is labeled “trend”. The trend shows a gradual increase from 2002 to around 2008. After 2010, the trend declines slightly and then levels off, remaining relatively stable from 2015 to 2025.
Seasonal (Third Chart): This chart captures the repeating, cyclical patterns in the data that occur at regular intervals. The seasonal component shows a clear, consistent oscillatory pattern with regular peaks and troughs. The amplitude of these oscillations appears relatively constant over time, indicating a stable seasonal effect throughout the period.
Random (Bottom Chart): This chart represents the residual or random component, which is the remaining variation in the data after the trend and seasonal components have been removed. The random component shows irregular fluctuations around zero, with no discernible pattern. These fluctuations represent noise or unpredictable variations in the data.
Summary: The observed data is a combination of a long-term trend, a repeating seasonal pattern, and random noise. The trend indicates a rise until 2008, followed by a slight decline and then stabilization. The seasonal component shows a consistent cyclical pattern, likely indicating a regular repeating event (e.g., annual cycles). The random component captures the leftover noise, which appears to be relatively small compared to the trend and seasonal effects, suggesting that the trend and seasonal components explain most of the variation in the data.

Moving average

References

U.S. Census Bureau. (2025). USA trade online. U.S. Census Bureau. https://usatrade.census.gov/