The Perimeter Grew Quietly
The U.S. export-control system has reached a much larger share of export value than the phrase “restricted technology” might suggest. This chapter examines what changed between 2010 and 2024, why EAR99 shipments are central to the story, and how the data distinguish a controlled shipment from one that requires an explicit license. The payoff is a more careful reading of the perimeter: broad coverage signals broad exposure to regulatory decisions, but it does not by itself measure a firm’s compliance cost or prove economic harm.
When one-third becomes three-fifths
The Census working paper measures controlled exports using public transaction data that include the product classifications used to administer U.S. export controls. On that measure, the share of total U.S. export value subject to the Export Administration Regulations rose from about 33 percent in 2010 to about 60 percent in 2024. The abstract rounds the starting figure to about 30 percent, but the main text reports the more precise comparison of roughly one-third to three-fifths.
The increase did not come mainly from a sudden concentration of highly sensitive goods. The share of export value associated with an explicit ECCN averaged about 13 percent of all exports across the period and reached roughly 15 percent in 2016. By contrast, the share designated EAR99 rose from about 23 percent of all export value in 2010 to about 45 percent in 2024. The controlled perimeter expanded largely through goods that were subject to the EAR but not listed on the Commerce Control List.
That result changes the mental picture of regulation. An ECCN often suggests a product with a clearly identified technical reason for control. EAR99 is less visually dramatic: the product is within the EAR but does not appear on the CCL. Yet EAR99 does not mean “outside the law.” The exporter still has to determine whether the destination, end user, end use, or another restriction changes the authorization question. A broad rise in EAR99 coverage can therefore enlarge the population of transactions that need classification and screening, even if it does not produce a matching rise in formal license applications.
The geographic pattern reinforces the point. Group B countries—most of the ordinary partner destinations in the paper’s classification—accounted for about 77.3 percent of the value of exports subject to the EAR in 2024, compared with 76.3 percent in 2010. Group D destinations, which include countries receiving more restrictive treatment, accounted for 18.9 percent in 2024, down modestly from 21 percent. Group E destinations under comprehensive embargo represented only 0.06 percent in 2024, compared with 0.25 percent in 2010.
Those shares are not a ranking of national risk, and they are not a claim that Group B transactions are routine in every legal sense. They show where the value of EAR-subject trade sits. The controlled system is not confined to the most visibly prohibited destinations. Much of the value moves to countries that receive relatively favorable treatment but still participate in a system of classifications, exceptions, screening, and conditions.
The paper’s Appendix D supplies another necessary distinction. In 2024, at least 95 percent of controlled export value did not require an explicit BIS license across the country groups, with Group E as the exception. This is compatible with the overall rise in controlled coverage. “Subject to the EAR” is a broad category; “requires a license” is a narrower conclusion reached after considering the item and the transaction.
The difference matters because a license is only one visible event in the compliance process. A shipment that does not require a license may still need an accurate classification, a check of the destination and parties, documentation of the decision, and monitoring for changes before release. If analysts count licenses as a proxy for all control-related work, they will miss much of the regulated perimeter. If firms treat no license as no responsibility, they may miss the conditions that make an otherwise ordinary shipment problematic.
The composition of controlled exports also changed in ways that are easy to overlook in an aggregate percentage. In the paper’s 2024 data, EAR99 accounted for 75.6 percent of the value within the set of shipments associated with an ECCN or EAR99 designation, up from 68.7 percent in 2010. Among the explicit CCL categories, aerospace and propulsion represented the largest share, followed by telecommunications and information security and then electronics. The categories are not interchangeable, and their shares do not tell us how difficult any particular transaction is to review. They do show that the expanding perimeter contains different types of exposure.
Why the perimeter is difficult to measure
The data become harder to interpret when we ask what exactly is being classified. A Schedule B code is a ten-digit export code based on a good’s physical characteristics. An ECCN is based on technical specifications, intended use, and reasons for control. One answers a customs-reporting question; the other helps answer a regulatory-control question. The same object can therefore have both codes without the codes carrying the same meaning.
There is no one-to-one bridge between them. One Schedule B code may appear on shipments associated with several ECCNs, while one ECCN may be used alongside many Schedule B codes. The same Schedule B category can also appear on a shipment with no listed ECCN. A researcher who labels every shipment in a familiar physical category as controlled may overstate exposure; one who searches only a narrow set of physical categories may miss technically controlled shipments.
The 2024 mapping makes the ambiguity concrete. Of 9,393 unique ten-digit Schedule B codes in the paper’s analysis, 53.6 percent were associated with EAR99 but no ECCN, 14.4 percent with exactly one ECCN, 20.2 percent with two to ten ECCNs, and 5.4 percent with eleven or more. The remaining 6.5 percent were never associated with EAR99 or an ECCN in the observed shipments. The reverse mapping is just as dispersed: among 399 unique non-EAR99 ECCNs, 56.6 percent were associated with two to ten two-digit Schedule B chapters and 18.5 percent with eleven or more.
These figures do not mean that every code is equally uncertain. They mean that the code systems operate at different conceptual levels. A Schedule B description can be broad enough to contain multiple technical configurations, while an ECCN can apply to components or applications that appear in physically unrelated trade categories. The relationship also changes over time. A Schedule B code associated with an ECCN in 2024 was not necessarily associated with the same ECCN—or with any ECCN—in an earlier year.
That time variation creates a measurement problem with a legal consequence. A historical mapping cannot automatically settle the status of a current shipment. A classification decision has to use the applicable rules and the item’s actual technical facts, not only the fact that a code appeared in a previous dataset. This is one reason the BIS guidance places technical classification inside a documented authorization process rather than treating a trade code as a final answer.
The paper’s firm-level data make the study possible, but they also define what the study cannot observe. The Longitudinal Firm Trade Transactions Database includes goods shipments over $2,500 and records destination, product code, ECCN or EAR99 designation when present, firm identifiers, and related-party status. Those transactions can be linked to domestic employment and payroll data and to a survey used to identify known R&D performers.
The database does not identify the foreign buyer. That prevents the researchers from separately identifying transactions involving a particular foreign legal person on the Entity List. License information is also incomplete: it is sometimes included in the transaction data but not consistently enough to identify all licensed firms. The paper’s authors therefore identify EAR-subject shipments through their ECCN or EAR99 designation rather than claiming a complete census of every license decision.
Those limits are not merely technical caveats. They determine how the trend should be used. The data can show how much export value falls inside the observable control categories and can support comparisons across firms and destinations. They cannot reveal every conversation between an exporter and a customer, every internal classification memo, every screening result, or every hour spent resolving uncertainty. The measured perimeter is a window into the system, not a transcript of all compliance activity.
The diagram captures why a single number cannot answer every question. A shipment has a physical export classification and may also have an ECCN or an EAR99 designation. Its destination affects the authorization context, and a license may or may not be required. The mapping between the two code systems is many-to-many and time-sensitive. None of those relationships means that a Schedule B code alone determines the legal result.
Three perimeters, not one
It helps to separate three ideas that are often merged in public discussion. The first is the legal perimeter: the transactions to which the EAR and related restrictions can apply under specified facts. The second is the measured perimeter: the transactions that the available Census data classify as associated with an ECCN or EAR99. The third is the economic perimeter: the people, processes, products, customers, and domestic operations whose behavior changes when a firm responds to the rules.
The first two overlap but are not identical. The paper’s empirical categories are designed to identify exposure, yet the authors note that buyer information and complete license information are unavailable. The third is wider still. It includes outcomes such as a delayed order, a changed customer relationship, an engineering review, or a decision to avoid a sensitive destination—events that may not appear as a separate field in the export record.
This three-perimeter model explains why the 33-to-60 percent trend is important without making it do too much work. It is evidence that the observable regulatory perimeter expanded substantially. It is not evidence that 60 percent of firms faced the same burden, that 60 percent of trade required a license, or that the domestic cost doubled. Export value is a useful scale measure, not a direct measure of administrative effort.
The same caution applies to the country groups. The fact that most controlled export value goes to Group B destinations means that compliance exposure is not synonymous with embargo exposure. It does not imply that every Group B shipment is low risk or that every Group D shipment generates the same burden. Product characteristics, technical details, customer identity, intended use, and applicable exceptions still matter transaction by transaction.
Nor should the growth of EAR99 be read as proof that the government expanded controls by silently labeling ordinary commerce as dangerous. The designation means that the item is subject to the EAR but not listed on the CCL; its licensing outcome remains conditional. The trend shows that the system’s regulated category became larger. To determine whether that enlargement created wider economic effects, we need firm-level evidence and a comparison that can distinguish policy responses from ordinary changes in trade.
The important conclusion is therefore about interpretation. The export-control perimeter grew through a category that often does not end in an explicit license, and the data expose that growth only through classifications that do not map cleanly onto physical trade codes. Broad measured exposure is a reason to examine how firms organize decisions and absorb shocks. It is not a substitute for that examination, and it cannot by itself establish the size or direction of domestic harm.