●

Methodology

How the India Trade Terminal sources, validates, calculates and flags India's official merchandise trade data. Every formula on this page is read directly from the code that produces the numbers you see elsewhere on the site — nothing here is a separate, possibly-drifting description of it.

Live data status · methodology v1.0.0 (2026-09-19)
Data through
AUG 2026
Latest period
PROVISIONAL
Coverage
APR 2018 → AUG 2026
Records
70.6M RECORDS
Signal engine
v1.0.0
Signal status
READY
Build
b20261102T0654
Last refresh
2026-11-02

Provisional indicates the latest available reporting period and may be revised in subsequent source releases.

Data source & coverage

The permanent analytical dataset — permanent historical analytics, screener results, signals, anomaly detection and backtests — comes from DGFT TradeStat, published by the Directorate General of Foreign Trade, Government of India — the official record of India's merchandise (goods) exports and imports. Two faster, provisional government releases are shown alongside it, clearly labelled and never mixed into that history. The DGCI&S Selected-Country Quick Estimates support only limited, clearly-labelled provisional calculations for the newest country-level month, with no HS-level detail; the India Topline Estimate is a separate provisional all-India layer. See the data layers below.

Coverage runs from 2018-04 through AUG 2026, monthly, across the full HS (Harmonized System) hierarchy DGFT publishes:

  • HS2 — Chapter (e.g. Chapter 85, electrical machinery)
  • HS4 — Heading
  • HS6 — Subheading
  • HS8 — Tariff line, the most granular published level

Both flows (exports and imports) are covered, by commodity and by trading partner country. The database currently holds 70.6M RECORDS across APR 2018 → AUG 2026.

Validated vs. provisional. The most recent one or two months are typically still provisional in the source release and can be revised upward or downward once DGFT finalizes them. This site marks that on screen wherever it applies, and never silently treats a provisional month as final. Right now, AUG 2026 is provisional.

Currency. DGFT publishes trade values in US dollars, and every calculation on this site — growth rates, CAGR, shares, concentration, signal detection — is computed in USD. An INR display toggle exists purely for reading convenience: it converts a USD figure using that specific month's own average USD/INR exchange rate (sourced from the U.S. Federal Reserve's published daily rate, averaged per month), not one current rate applied across the whole 2018–2026 history, because the rupee has moved substantially over that span. This conversion happens only at display time and never changes an underlying figure or a percentage — a growth rate or CAGR is a ratio, so it is identical in either currency.

Indian financial year. Where the site shows fiscal-year figures, an Indian FY runs April–March and is named after the calendar year it ends in (e.g. April 2025–March 2026 is “FY2026”). An in-progress fiscal year is never compared against a completed one — a fiscal-year comparison only happens between two periods covering the same number of months.

Detailed vs. provisional data layers

Three government sources publish India's trade at different speeds and different levels of detail. The site keeps them as separate layers, and every page says which one you are looking at and for which month (e.g. DETAILED · JUL 2026, COUNTRY QE · AUG 2026 · PROVISIONAL).

1. Detailed layer DGFT TradeStat country x HS (chapter -> tariff line), both flows, monthly since 2018. The permanent analytical dataset: historical analytics, screener, signals, anomaly detection and backtests use this layer. Newest month is provisional in DGFT's own release (~45-75 day lag). 2. Selected-Country Quick Estimate DGCI&S Provisional country-level totals for a SELECTED list of partner countries. No HS/commodity breakdown. Only the one month right after the detailed layer is shown live (dashed point on Country and Group charts); superseded automatically once detailed data for that month lands. Supports only limited provisional calculations for that month (value, YoY, MoM, FYTD, share, covered-member group totals, member drivers). A group trade balance is shown only when the export and import releases cover exactly the same members. Never fed into signals, backtests or aggregated periods. 3. India Topline Estimate Department of Commerce (monthly press release) India-level totals only (~15 day lag), shown on the home page. No country and no HS breakdown. Separate from the Selected-Country Quick Estimate.

A country missing from the Quick Estimate is not zero. The Selected-Country release lists only some partners. A country outside it is shown as “not covered in the latest Selected-Country Quick Estimate” and its chart simply ends at the latest detailed month. For a Country Group, the estimate is a sum over the covered members only, always labelled with how many counterpart members are covered (e.g. PARTIAL · 8/9 COUNTERPART MEMBERS) and never scaled up to a full-group total.

How a data build is validated

Every monthly release goes through the same conceptual pipeline before it is ever served:

DGFT source release -> ingestion into the historical fact table -> analytical marts rebuilt from that fact table -> automated validation of the new marts -> signals recomputed against the validated marts -> served to the site (only once validation passes)

Validation checks include row-count sanity (no shrinkage month over month), month-to-month continuity, coverage across every HS level, internal consistency between rollup levels, and a reconciliation of the latest month's totals and top codes/partners back against the underlying fact table. A structural integrity check runs against the database file itself before anything is activated. New marts are built alongside the ones currently serving the site and only replace them in one atomic step once every check passes — if a check fails, the previous validated build keeps serving and the site says so, rather than showing a partially-built or inconsistent state.

If the underlying source has published a newer month than what has passed validation, the site reports “validation pending” for that month instead of serving unvalidated figures.

See /about for how this fits into the overall system, and what is deliberately left out of that description.

Calculations

Every calculation on this site — on every page — runs through one shared module. Two pages can never disagree about what a “YoY” or a “CAGR” means because there is exactly one implementation of each. A percentage with no mathematically valid base (division by zero or a negative prior value) is always shown as N/A, never as 0%.

Month-over-month (MoM) and year-over-year (YoY)

Both use the same generic percent-change formula, just against a different prior period:

pct_change = (current - previous) / previous x 100 MoM: previous = the immediately preceding month YoY: previous = the same month one year earlier

Undefined (shown as N/A) whenever the prior value is missing, zero, or negative.

Trailing twelve months (TTM)

The sum of the most recent 12 validated months for a given flow. TTM is the basis for CAGR, rank, and share below, and is refreshed every month as the trailing window rolls forward.

3-year and 5-year CAGR

Compound annual growth rate between the current trailing-12-month total and the trailing-12-month total ending 3 (or 5) years earlier:

CAGR = (TTM_now / TTM_N_years_ago) ^ (1/N) - 1, as a percent Undefined (N/A) when the earlier TTM is zero or negative, the later TTM is negative, or N is not positive -- a CAGR through a value that went to zero or negative is not a meaningful real number under this formula.

Trade balance

Exports minus imports for the same period, positive meaning a surplus.

Export / import share and rank

A commodity's (or country's) share of India's total trade for a flow is its TTM value divided by the TTM total across all codes at that level (or all partners), as a percent. Rank orders every code at the same HS level by TTM value for that flow, over the same window.

Concentration — HHI and bands

Partner (or commodity) concentration uses the standard Herfindahl-Hirschman Index: the sum of each participant's squared percentage share.

HHI = sum(share_pct ^ 2), share_pct on a 0-100 scale A single-partner monopoly (100% share) gives HHI = 10,000; N equally-sized partners give HHI = 10,000 / N. Band (standard DOJ/FTC merger-guideline thresholds): HHI < 1,500 Low concentration 1,500 <= HHI <= 2,500 Moderate concentration HHI > 2,500 High concentration

Top-1/3/5/10 share figures are simply the sum of the largest 1, 3, 5 or 10 shares.

Volatility

Measured as the coefficient of variation — population standard deviation divided by the mean of a monthly-value series, a plain ratio rather than a percent. Chosen over YoY-volatility or log-change-volatility for being scale-independent and reproducible from a single series with no extra machinery. Undefined for fewer than two observations or a non-positive mean (a volatility of exactly 0 would misleadingly imply “perfectly stable” for a series that averages to zero or below).

Indexed (base-100) comparisons

When multiple series of different absolute scale are plotted together, each is rebased so its first valid value becomes 100 and every later value is shown relative to that anchor — standard practice for comparing shape/trajectory rather than absolute size.

Low-base treatment

A growth percentage computed off a very small prior value can be technically correct and practically meaningless (a rise from $200,000 to $2 million is a real move but a “900% surge” headline overstates its importance next to a chapter worth billions). Two independent checks flag this, and either one is enough:

Absolute check: prior value is between $0 and $1 million (USD) Relative check: current value is more than 10x the prior value (equivalent to growth beyond +900%)

A flagged figure is shown with a low-base indicator rather than being hidden or altered — the number itself is not changed, only its context. The absolute check alone is calibrated for reviewing one code at a time; screening across many codes at once additionally needs the relative check, since a coarser HS level's “typical” values are large enough that the absolute check alone would miss a similarly misleading multiplier.

Country Groups

A Country Group (BRICS, the EU, ASEAN, the GCC and others under /groups) is not a new dataset — it is India's existing bilateral trade with each member country, summed. A group's exports are the sum of its members' own exports figures shown on their individual Country Intelligence pages; nothing is computed independently of those numbers, so a group total can always be checked by adding up its Member Countries table.

Historical membership vs. current members through history

Group membership changes over time — the EU lost the UK in 2021, BRICS added five countries in 2024–2025, Bulgaria joined the euro area in 2026. Two different, equally valid questions can be asked of that history, and this site never mixes them:

Historical membership (default) Each month counts only the countries that actually belonged to the group in that month. A country's trade counts from the month it joined (or before it exited), never earlier or later. Current members through history Takes today's member list and applies it to every month in the database, including years before some members joined. Answers "what would this group's trade have looked like if today's roster had always applied?" -- a hypothetical, not a historical record.

The convention for WHEN a change takes effect: a membership change effective on date D is attributed from the first day of D's calendar month (this site's trade data is monthly, so day-level precision is not meaningful). Each member's exact accession or exit date, and its source, is recorded alongside its country code in this site's group configuration.

Groups overlap

A country can belong to several groups at once — the UAE is in both BRICS and the GCC, Brazil is in both BRICS and MERCOSUR. Group totals are therefore not mutually exclusive and should never be summed across groups (India's “total trade with country groups” is not a meaningful number, since it would double-count every country in more than one group shown here).

Membership statuses

Not every relationship is full membership. This site distinguishes MEMBER, PARTNER (e.g. BRICS Partner Countries, a distinct group from full BRICS membership), OBSERVER, SUSPENDED (e.g. Venezuela in MERCOSUR since 2016), ASSOCIATE, and INVITED (accepted an invitation but not yet formally acceded, e.g. Saudi Arabia and BRICS). Only MEMBER status counts toward a group's headline trade figures by default; the others are real relationships but are not silently treated as equivalent to full membership.

Sources

Membership and accession dates are sourced from each organization's own official website or another primary intergovernmental source (linked from every group's page), not from crowd-sourced references. Where a status is genuinely contested or still unfolding (Myanmar's partial RCEP implementation, Bolivia's exact MERCOSUR accession date, Afghanistan's disputed SAARC representation since 2021), that is stated on the group's page rather than resolved by guessing.

Explain This Month, for a group

A group's month-over-month or year-over-year move can be decomposed two ways: by member country (which members drove the change) or by commodity (which HS chapters drove it) — both reuse the exact same decomposition math as Explain This Month everywhere else on this site. A member joining or leaving the group between the two compared months appears as a “new” or “discontinued” driver, the same mechanism already used for a commodity that starts or stops being traded.

Signals

Signals flag commodities, countries, or trade-balance situations whose recent behaviour is unusual relative to their own history — they describe what changed, not what will happen next. Two families exist: event signals (a single month spikes or collapses) and structural signals (a slower shift in partner share, concentration, or trade balance over a full year).

Seasonality-adjusted anomaly detection

A month is compared against its own trend and its own seasonal pattern, not against a generic threshold, so that ordinary growth is not mistaken for an anomaly and a recurring seasonal peak (e.g. every December) is not flagged as unusual every year it happens:

baseline(t) = average of the 12 months immediately before month t ratio(t) = value(t) / baseline(t) history = ln(ratio) for the same calendar month in the 3-5 prior years (only years with a usable baseline and a positive value) z(t) = ( ln(ratio(t)) - mean(history) ) / max(stdev(history), 0.10) z is capped at +/-10. A month that falls to exactly zero is treated as the most extreme possible collapse (z = -10) since its ratio has no logarithm.

Dividing by the trailing 12-month average removes trend (a series that has doubled in five years is not flagged just for being bigger); comparing against the same calendar month in past years removes seasonality. The floor on the standard deviation stops an unusually steady series from turning a tiny wobble into an outsized z-score.

Normal range. A signal's “normal range” is baseline × exp(mean ± 2.5 × stdev). Because the spread is measured on a log scale, a lumpy series (occasional near-empty months) gets an enormous band — a $4M tariff line can come out with an upper bound in the trillions. When the band runs more than 10× the expectation either way, the site shows HIGH UNCERTAINTY instead of printing meaningless bounds. The actual value, expectation and z-score are always shown, and detection is unchanged.

Spike / collapse classification

A month qualifies as a spike only when its seasonal z-score clears the threshold (2.5) and its year-over-year growth clears a minimum (currently +30% for exports, +25% for imports) and the month's value clears a minimum dollar floor (so a jump from near-nothing to a small number never qualifies). A collapse is the mirror image: z at or below −2.5, YoY at or below −30%, and the same month a year earlier having been large enough to make a collapse meaningful. Severity is low / medium / high based on how far past the threshold the z-score sits.

Structural signals: share, concentration, balance

These compare the trailing 12 months with the 12 months immediately before that — a full year against a full year, so a single seasonal month cannot drive the result.

  • Share gain / loss — a trading partner's share of a selection's trade moves by at least a minimum number of percentage points (a smaller move is enough to flag at India's all-partner level than within a single HS4 heading, where the bar is higher), with a floor on both the selection's and the partner's trade value so a shift among negligible flows is not reported as if it were meaningful.
  • Concentration change — the HHI across partners (or commodities) moves by at least a minimum number of points, or crosses a concentration band, between the trailing 12 months and the 12 before.
  • Trade balance reversal — the trailing 12-month balance changes sign (surplus to deficit, or the reverse) versus the prior 12 months, with both periods' balances clearing a minimum absolute size so a reversal between two near-zero balances is not flagged as significant.

Partner attribution

Where a signal has an identifiable driver, it is the trading partner whose own change (versus the same period a year earlier) is largest in the direction of the overall move — shown together with the share of the total move that partner accounts for. This describes what contributed most to a move already detected; it does not imply that partner will keep driving it going forward (see limitations below).

Parent/child suppression

A chapter- or heading-level (parent) event signal whose size is essentially explained by one specific child code's move at the next level down is marked as covered by that child, rather than shown twice. This keeps the feed showing the most specific code responsible for a move once, instead of the same underlying move appearing at every HS level that contains it.

Signal limitations

Signals are descriptive, not predictive. They report that a change in trade data has already happened and cleared a defined statistical bar — they are not investment recommendations, forecasts, or trading signals in the financial sense.

  • A signal describes a change that has already occurred in reported trade data, not a prediction of what happens next.
  • Rare, one-off events (a single large shipment, a base effect, a reporting anomaly) can trigger a statistically real signal that has no lasting significance.
  • Historical relationships between a signal and what typically follows it can change over time and are not guaranteed to repeat.
  • Nothing on this page or in the signal feed is investment advice, and no signal should be treated as one.

These are not just disclaimers — they are the honest reading of this site's own historical backtest. The full results are published at /signals/validation and are worth reading in full before relying on any signal type. The single most important finding: structural signals (partner share shifts, concentration changes, balance reversals) have historically persisted strongly and predictably for six to twelve months after firing, while one-off event signals (spikes in particular) persist much more weakly — a spike is more often a one-time occurrence than the start of a durable new normal, while a collapse tends to reflect a real, longer-lasting break. See the validation page for exact persistence rates by signal type, severity, and horizon, and for the sample-size rules applied before any rate is shown as reliable.

Access

Interactive use of this terminal — browsing commodities and countries, running the screener and comparisons, reading signals — is public and free, and that is exactly what it is built for. Automated bulk extraction or systematic replication of the underlying dataset is a separate matter and is restricted; see /terms for the actual policy. See also /about for how the system is built and how the DGFT source data relates to this site's own derived work.

Historical trade analytics on this site are for research and informational purposes. Signals describe unusual changes in reported trade data; they are not investment recommendations. See Terms & data usage.