Methodology

How a report is built

"Traceable to a source document" is a claim, and claims should be checkable. This page sets out what it means in practice: the order sources are trusted in, how figures get verified against each other, and what is deliberately left out.

1. The source hierarchy

Every figure is traced to the highest-ranking source available. Where a lower-ranked source is the only one available, the report says so at the point the figure is used.

  1. Audited financial statements. Annual accounts with their schedules and notes. The notes are usually where the answer actually sits.
  2. Statutory filings. Documents companies are legally required to file, including MCA filings and their attachments.
  3. Regulatory disclosures. Filings made to a sector regulator, plus the regulator's own published aggregates and circulars.
  4. Scheme or product-level disclosures. Where an industry publishes at instrument level, that level gets used in preference to any industry total.
  5. Company-published material. Investor presentations and press releases, used for context only, never as the basis of a headline number.

What is not used

  • Syndicated market-research databases whose own sourcing cannot be inspected.
  • Market-size figures cited in news coverage or industry decks without a traceable origin.
  • Estimates from parties with a commercial interest in the number being large.
  • Any figure that cannot be reconstructed from a document a reader could retrieve.

2. Measuring instead of inheriting

Where an industry total is commonly cited, that total does not get taken on trust. It gets rebuilt from the bottom by measuring the units that compose it, and the coverage achieved is reported. If 39% of an industry has been measured directly, the report says 39%. It does not quietly gross the figure up to 100% and present the result as measured.

Where grossing up is unavoidable, the extrapolation is stated separately from the measured base, so a reader can discard it and keep the measurement.

Issue 01 labels this explicitly: figures are measured at scheme and AMC level, directional at industry level. The three-AMC sample is measured. The industry pool implied from it is an extrapolation, and it assumes the relationship between passive share and effective rate holds across the whole industry. Both numbers are in the report, presented as the different things they are.

3. Cross-verification

Every measured aggregate gets reconciled against an independent path to the same number, usually a disclosed total from a different document, produced by a different party for a different purpose. The gap between the two gets reported, not buried.

A worked example from Issue 01. Each scheme's investor-education charge was extracted two independent ways from the AMC scheme annual reports. The first uses the related-party disclosure of the scheme's contribution to AMFI. Since 50% of the levy is transferred by mandate, doubling it recovers the total charge. The second uses the scheme-level investor education and awareness expense reported under other expenses.

For SBI Mutual Fund the two methods produced ₹175.25 crore and ₹174.96 crore, a gap of 0.17%, with 61 of 64 tested schemes agreeing within 1%. Those figures are printed in the report so you can check the reconciliation yourself. A gap that size supports the measurement. A gap of 15% would have meant the method was wrong, and the report would have said so.

4. Assumptions are visible and separated

Analysis requires assumptions. The rule here is that they are never buried in a calculation. Each report distinguishes three things explicitly:

  • Measured. Read directly from a source document, and cited.
  • Derived. Calculated from measured figures, with the calculation shown.
  • Assumed. A judgement call, stated as one, with the reasoning given and a sensitivity so you can see what happens if it is wrong.

Delivered models work the same way. Assumptions live in one labelled block, never hard-coded into formulas, and every output traces back to them.

5. Sensitivity, not false precision

Where a result depends on an uncertain input, the report gives a range and says which variables actually move the number. A single-point estimate carried to three decimal places implies a confidence the underlying data does not support.

6. Corrections

If a figure is wrong, the right response is to say so in public. Errors get published as dated amendments rather than silently edited, so anyone who downloaded an earlier version can tell what changed. Corrections can be sent to rajat@thebasispoint.in and are read.

7. What you receive

The same standard applies to commissioned work, and you get the working, not just the conclusion:

  • The report or model itself.
  • A sources appendix listing every document used, with enough detail to retrieve each one.
  • The assumptions block, separated as described above.
  • A walkthrough call, where the numbers get defended directly.

See it applied in Issue 01