Data literacy

India's trade data: what's in it, what's missing

Every trade intelligence product is built on the same few public and semi-public record sets. Knowing which one you are looking at, and what it structurally cannot contain, is the difference between evidence and confident nonsense.

TradeLala · 15 August 2026 · 7 min read

"Trade data" gets sold as though it were one thing. It is at least three. They carry different fields, they were collected for different reasons, and they disagree with each other. Most bad conclusions in trade research start with not knowing which one is on the screen.

The three sources

Alongside these sit the official aggregate statistics, from DGCI&S and the Ministry of Commerce's monthly releases. Those are the figures quoted in the press: national totals by commodity and country, with no company names in them at all. They are authoritative for "how big is this market". Transaction-level data is a different animal, and the two will never reconcile exactly, because they are compiled on different bases with different cut-offs and revisions.

What is structurally missing

None of these are defects in any one provider. They are properties of the record sets themselves, and no amount of money buys around them.

1. Coverage is partial, and uneven

   No commercial dataset is a complete census of every consignment through every Indian port, in every year, in both directions. Coverage varies by port, by period and by trade direction, and import-side coverage is generally thinner than export-side. One hard consequence follows from that: a nil result never proves a company does not trade. It proves the record is not in this data.

   Any provider implying otherwise is either not reading their own pipeline or hoping you will not ask.

2. Whole modes of trade produce no record

   Land borders. Intra-EU movements, which are not customs events at all. Courier and postal consignments below thresholds. Much of e-commerce. Services, licensing and software, which are a growing share of India's export earnings and leave no shipment behind. If your product moves any of these ways, shipment data is simply the wrong instrument.

3. There is a lag, and it is not uniform

   Records reach a commercial dataset weeks to months after the event, and the lag differs by source and by port. Practical effect: the most recent month always looks like a collapse in volume, because it is incomplete. Never read the last data point as a trend. Compare complete periods only.

4. Identity is legally fuzzy

   Negotiable bills carry TO ORDER instead of a consignee. Master bills name forwarders at both ends. Group buyers import through a single procurement subsidiary in a third country, so the shipment lands in the Netherlands while the decision was made in Germany. Traders sit in the middle deliberately, sometimes using switch bills of lading. Each of those is a real company relationship that the record represents incorrectly.

5. The fields are dirty in specific, predictable ways

   The goods description is free text typed by a shipping clerk. The same company appears as "ABC EXPORTS PVT LTD", "ABC EXPORTS PRIVATE LIMITED", "ABC EXPORT'S P LTD", and a version with the city stuck on the end. Quantity units switch between pieces, kilograms and square metres inside the same HS line. Values are INR here and USD there. HS codes are sometimes just wrong, because the filer optimised for clearance speed rather than accuracy.

   Cleaning all of that is most of the engineering work in any credible product, and it is where products differ from each other most.

How dirty data becomes a wrong answer

Every row there is a mistake I have seen in a real market report. All of them are avoidable, and none of them are avoided by buying more rows.

Nine questions to ask a provider

Ask these before you pay. The answers, and how willing someone is to give them, tell you more than any demo.

• Which sources is this built on: customs, manifests, mirror data, or a blend, and can I tell which record I am looking at?
• What is your coverage by port and by year, for the direction I care about? You want a specific answer, not "comprehensive".
• What is the lag, by source, and how do you handle the incomplete most-recent period in charts?
• How do you resolve company names? Ask them to show you a company with several spellings.
• What do you do with "TO ORDER" and with forwarders? If they do not immediately know what you mean, you have your answer.
• How do you handle mixed units and currencies when computing unit values?
• Can I get from any statistic to the individual bills behind it? A number you cannot drill into is a number you cannot defend.
• Where do contact details come from: attributed from the shipment records themselves, or scraped and matched? The difference matters legally and practically.
• What does your data not contain? This is the most useful question on the list, and a provider with a real answer is a provider who knows their own pipeline.

Reading it well: five habits

• Rank, do not total. Relative comparisons survive partial coverage. Absolute market-size claims do not.
• Always drill to the bills. Before acting on an aggregate, open three of the records underneath it and check they are what you think.
• Compare complete periods. Twelve months against the prior twelve, never a part-month against a full one.
• Treat absence as a question. "No record" starts an investigation rather than ending one.
• Cross-check with a second source for anything going in front of a bank, a board or a buyer. Partner-country mirror data is the natural check on India-side numbers.

Used this way the data is genuinely powerful. It is the only evidence base that names who bought what, from whom, when, and roughly at what price. Used carelessly it produces a deck full of numbers that nobody can defend the moment a buyer pushes back.

The methods built on top of it, finding buyers, choosing markets and positioning a price, all assume you have read this page first. They are only as sound as the reading underneath them.