Margin trading in detail: methods
Updated 21 October 2026 following a source-definition and calculation audit.
Code: compute/mtf_detail.py, mtf_positions.py, mtf_identity.py,
trading_liquidity.py, mtf_insights.py and their publishers. Page:
/leverage/mtf/. Audit evidence: docs/status/mtf-audit.md,
docs/status/trading-liquidity.md, audits/mtf/, audits/liquidity/.
Source and reporting regimes
NSE’s daily public security-wise MTF file reports symbol, security name, quantity financed and amount financed in ₹ lakh, plus summary balances, fresh exposure and exposure liquidated. The file contains no client count, purchase lots, collateral balances or client loan balances. The amount is published as reported MTF amount; quantities are reported financed shares.
Earlier wording on this site asserted that the public amount was the gross purchase value of all funded positions. That was too definite. The 2021 member reporting validation package describes funded quantities/amounts after margin adjustments. The revised member format distinguishes gross funded cost values, maintenance-margin shares and net funded exposure at cost; net quantities can be rounded upward. These rules support a funded-unit cost interpretation but the plain public header does not identify its gross/net mapping. It is also incorrect to assume the published amount is an observed net broker loan or current market value.
Sources checked:
- Public report, 16 October 2026.
- Original public format, 2004.
- 2021 reporting and validations, COMP50280: Annexure C, “Previous data checks”, B2:B3.
- Revised member format, COMP66522: Annexure C fields 14–19; Annexure D validations.
- Implementation date, COMP69153: revised reporting applies to trade date 1 October 2025, reported on 3 October.
- NSE MTF FAQs.
The October 2025 step in gross openings/closings coincides with a documented reporting regime change. This does not establish that every change was administrative, nor permit a clean comparison of holding periods or trading behaviour across the break. The 2017 regime change is separately marked: SEBI framework, 13 June 2017.
Modelled estimates and why the exact figure is unknown
Reference price = reported amount in rupees / reported financed shares. This is a security-level funded-unit reference. It is not a reconstruction of individual acquisition lots or a verified average entry price for the whole gross client position. Changes in reported quantity and amount aligning with traded prices provide supporting diagnostics, not proof of that interpretation.
A security below its reference can contain clients in profit, and one above it can contain clients at a loss. Interest, charges, collateral and each client’s margin rules are unavailable. Neither the stock count nor reported-amount share estimates the proportion of losing clients, losing lots or margin calls.
The underlying calculation is descriptive: close relative to the reference. Stock count share gives each eligible stock one vote. Reported-amount share weights each stock by its reported MTF amount. These have different meanings and denominators. Counts reconcile as below + above + exactly at reference = analysed stocks. Positive financed quantities, positive amounts and valid same-day prices are required; ETFs are excluded.
The existing screen retains close/reference ratios from 0.5 to 2 inclusive.
It can remove real extremes as well as corporate-action/reporting mismatches;
it is not a verified error detector. The page now shows both unfiltered and
screened comparisons, the excluded stocks and amounts, missing prices and
invalid references. Each variant uses its own sample denominator. Legacy JSON
field names containing cost preserve the contract; they do not establish a
client purchase-cost measurement.
value_vs_cost_pct is mechanically
100 × (Σ reported financed shares × close / Σ reported MTF amount − 1)
within the screened sample. It is labelled financed-unit market value relative
to reported amount. It is not account P&L, an equity cushion, leverage ratio
or distance to a margin call.
The reference history starts in June 2022, where sampled amount/quantity changes align more closely with traded prices. Adjusted-price corrections can be retrospective. These historical descriptions are not point-in-time backtests.
Historical identity and AMFI classifications
mtf_identity.py resolves the latest known price-panel identity on or before
each report date. It never takes a later identity backward. Match historical
AMFI lists by ISIN first, then an unambiguous NSE-symbol fallback. Ambiguous
keys remain unmatched. Missing NSE-symbol cells no longer discard ranked
companies; MCX and securities with renamed symbols were affected by that bug.
Use the latest list with period end + 15 calendar days on or before the report. This is an assumed publication lag; release timestamps have not been verified for every historical list. Large caps are ranks 1–100, mid caps 101–250, small caps 251 on. AMFI rank bands are not actual NSE index membership.
The aggregate market-cap benchmark covers all companies on the historical AMFI list across exchanges, including BSE-only companies, rather than only rows with populated NSE symbols. Ambiguous duplicate-ISIN rows and missing or nonpositive caps are excluded from this benchmark; excluded-row counts are published. Real source rank ties are allowed. The numerator remains the NSE public MTF disclosure, whose unmatched identifiers include BSE-style numeric codes. Its venue scope must not be assumed identical to NSE bhavcopy coverage. Do not add NSE and BSE MTF books without resolving possible overlap.
Company and group caps are six-month average full company market values, not current or free-float values. Reported MTF amount / average cap is a concentration comparison, not a loan-to-value ratio or exact ownership share. Unmatched exposure remains in full-book share denominators and is reported separately. Match-method shares are retained in the data.
Trading capacity and selling simulations
Volume equivalents compare reported financed shares with average total or deliverable quantity. Twenty market-session averages require at least 15 valid observations; the 60-session total-volume comparison requires 45. Missing bars are never zero-filled. Known ETF/SME holdings are excluded using their identity as of that date; past SME status does not permanently exclude a stock after main-board migration. Unknown holdings stay in the denominator and have no capacity estimate.
Applied splits/bonuses/consolidations rescale past quantities to current share units. Other applied actions invalidate affected windows. The underlying price pipeline includes inferred actions that use later closes for confirmation; these revised descriptive histories are not a point-in-time forecasting test.
Simulated trading days = reported financed shares × fraction sold / (average daily total quantity × participation × volume multiplier). Default: sell 20% at 10% participation; volume multipliers 1, 0.5 and 2. Controls alter the latest simulation table; history keeps these defaults. Medians are weighted by reported MTF amount among usable estimates. Shares of the book exceeding five simulated days retain unmeasured exposure in the full denominator, so they describe measured exposure rather than treating unknown holdings as liquid. Coverage is displayed; no usable estimates means a missing simulation reading, not zero days. The reported quantity may not represent the entire gross client position.
These are deterministic simulations, uncalibrated to real liquidation outcomes. They omit price impact, changing prices, circuits, competition for bids and simultaneous non-MTF selling. Double volume need not mean easier execution. Neither traded nor delivery volume establishes available bids or executable depth. Trading days mean exchange sessions, not calendar days.
Reference and capacity readings use each week’s last shared MTF/price date; newer MTF files are not paired with stale prices. Other reported-balance charts can have a newer date, explicitly shown. The superseded available-observation legacy delivery column is no longer displayed in the crowding table.
Reporting reconciliation and descriptive history
Within-report identity: opening + fresh exposure − closed exposure = closing, subject to source rounding. Separately, reported opening can differ from the previous report’s closing amount. Those discrepancies are not inferred flows. The reconciliation table shows the largest opening gaps and balance residuals. New same-day activity ratios use the reported opening book. Older closing and worst-day comparisons retain their explicitly labelled prior-close ratio denominator; the two need not agree.
NSE’s exposure-liquidated field does not identify voluntary versus forced closures. The observations around market falls do not establish trader motives, option-to-MTF switching, causation or when margin calls occurred. Full-month balance changes are reported differences, not clean cash flows.
Crowding-return test withheld
The old test removed stocks lacking a future exit price before forming groups,
introducing survivor/look-ahead bias and changing cohorts across horizons. It
has been removed from the page and its production bundles carry an explicit
withheld status, without a numeric verdict/history. The archived research
method records the defect: mtf_crowding_test.md. A valid rerun needs formation
using contemporaneous inputs, stable identities, consistent cohorts and
explicit handling of missing exits. A disclaimer is not a repair of this test.
Latest NSE impact-cost companion
The page now acquires the official
security-category and impact-cost file
through sources/nse_security_liquidity.py. The file header supplies its month,
year and detail-row count. Raw data, content hashes, retrieval time and parsing
provenance are retained. A 24-hour cache and explicit refresh-error/staleness
states prevent a failed fetch being represented as fresh data. File periods
older/newer than the retrieval month are flagged and not presented as current
impact readings.
NSE’s methodology uses order-book snapshots, a ₹1 lakh order and a six-month historical average. This is actual order-book-based historical-average impact cost, not current executable depth, and it must not be scaled directly to the MTF holding.
The companion matches ISIN and price series, preserving blank impact values. Ambiguous alias keys are quarantined rather than selecting the lower impact cost. The October 2026 source had 14,466 detail rows; 12 ambiguous keys/24 rows were quarantined, leaving 14,442 usable unique keys. The page states file month, retrieval time, MTF observation date and amount coverage separately.
This is a latest-file companion only. Historical archive availability and original release timestamps have not been established. The current file is never backfilled into historical charts or a return test.
Same-date current market-cap companion
mtf_current_cap.py compares reported MTF amount with the full company market
cap of the matched equity class in the same date’s NSE daily PR archive, acquired through the existing
owner-approved mtf.trading mirror. Eligible source rows are Listed or Permitted
ordinary equity in EQ/BE/BZ. This differs from the listed-only aggregate NSE
market universe on the market-cap page. Known ETFs and SMEs are excluded using
their identity on or before the holdings date; unresolved holdings remain in
coverage denominators. A cap symbol and series must identify an ordinary equity
on the same price date, then match the holding by unambiguous ISIN. Conflicting
keys/aliases and invalid values are not arbitrarily resolved. No stale cap is
carried. The published provenance retains report date, archive URL and the
cached cap-extract hash.
Separate equity classes remain separate rather than being summed into a whole-issuer value. The current full-cap ratio complements AMFI’s six-month average. Neither ratio is LTV, gross client ownership, free-float exposure or an execution-cost estimate. The table ranks the current ratios and retains the AMFI comparison separately.
Price bands and observed intraday hits
sources/nse_trading_constraints.py retrieves specifically dated NSE reports,
keeps raw hashes/retrieval times, validates file schemas, and caches each source
independently. A source failure does not become a zero count or substitute a
different date. These are latest dated companions, not historical event tests;
filename dates alone do not establish original publication availability.
The PR ZIP’s own readme.txt defines bhDDMMYYYY.csv as securities that hit
their price bands during the day. The H/L field supplies the direction.
The same archive’s normal-market pdDDMMYYYY.csv, with positive traded quantity,
defines the covered universe. Absence from a validated, complete hit report is
treated as no hit only for a covered traded symbol/series. Outside that universe,
status remains unknown. The source includes operating-range hits for securities
without fixed bands, so the total is not a count of locked circuit stocks.
The separate sec_list_DDMMYYYY.csv supplies percentage bands and No Band rows.
Its report date is retained; the effective session has not been independently
verified. Cross-referencing fixed-band assignments with official hits is shown
as a dated comparison with that limitation, not a definitive circuit-lock count.
All book percentages use the full non-ETF MTF amount, including unmatched
holdings. Counts refer to securities, not clients; overlapping rows must not be
added. Trade-to-trade identifies BE/BZ/ST/SZ in the verified traded universe.
Hits do not establish closing at a limit, an absent bid, lock duration, queue size or that MTF holders traded at that price. Closing-at-limit values are withheld unless same-session fixed rupee limits can be verified. The security-master limits inspected were centred on the report-day close and describe a subsequent session, so are not applied backwards. The computation can compare verified same-session fixed limits with raw high/low/close (half-paisa decimal tolerance, positive quantity, range validation); rounded +/-5% returns are never used as proof of circuit hits. Suspensions and surveillance stages remain unknown until dated effective coverage is verified; missing bhavcopy rows are not suspensions.
Public-share comparison withheld
Public-share data requires filings actually available on the holdings date,
stable equity-class identities, reconciled public/total share counts, and
corporate-action controls between the filing period and the MTF date. The audit
found incomplete coverage and share-count inconsistencies. More fundamentally,
NSE’s reported financed quantity has not been reconciled to gross client-held
MTF shares. A ratio of that quantity to public shares could imply an ownership
measure the inputs do not establish. It is withheld, with the evidence retained
in audits/mtf/public-share-feasibility-2026-10-21.json. Public holdings are also
not equivalent to investible free float or available bids.