---
title: "Method: Flows, rates and credit"
url: https://tipsheet.markets/methods/flows_rates/
description: "How tipsheet.markets computes flows, rates and credit: sources, start dates, checks and caveats."
updated: 2026-10-30
publisher: tipsheet (https://tipsheet.markets/)
cite: "tipsheet, “Method: Flows, rates and credit”, 30 October 2026, https://tipsheet.markets/methods/flows_rates/"
---

[Methods](https://tipsheet.markets/methods/)

# Flows, rates and credit

**Code:** `pipeline/tipsheet/compute/flows.py`, `publish/flows.py`.

**Bundles:** `flows/*`, `rates/*`.

All inputs are India Data Hub series in my data archive. The series IDs are listed in `SERIES`, `SPREADS` and `CURVE_TENORS` in the code. Values are converted to ₹ crore. Yields are in per cent and spreads in basis points.

## Investor flows

- **Net purchases in the cash market by investor category,** daily:
  
  - categories: FPI, mutual funds, insurers, banks, AIFs, PMS, retail and others
  - NSE from 2023-01-31; NSE plus BSE from 2023-12-04
  - before 2023-12-04 only the NSE figure is shown, labelled as such
- **DII net purchases:** daily from 2007-04.
- **FPI equity (NSDL, primary plus secondary):** daily from 2021.
- **Daily bundles:** `flows/investors_daily_1y` (every category, the last year) and `flows/investors_daily_5y` (from 30 Oct 2026: `dii_daily` and `fpi_equity_daily` only, the last five years, for the front page’s ranking of a day’s flow against its own days). Five years, not DII’s full history back to 2007: a crore of DII buying in 2008 is a far larger share of the market than one today, so older days would make today’s flows look rarer than they are. The weekly sums from 2007 are in `flows/dii_fpi_weekly_full`. Checked on 30 Oct 2026: the 5-year bundle equals the 1-year one on all 248 shared sessions.
  
  - NSDL dates trades one day after the exchanges. Same-day correlation with the exchange FPI figure is 0.41; shifted one day it rises to 0.87–0.91. So never compare the two on the same date.
- **Check:** mutual funds + insurers + banks + AIFs + PMS together track the DII series (daily correlation 0.96, ratio about 0.9).

## Flows scaled by market cap

Crore figures inflate as the market grows (NSE’s listed companies were worth about ₹43 lakh crore in mid-2009 and ₹451 lakh crore at the end of September 2026), so `/flows/` also shows each month’s net buying as a share of the market. Code: `pipeline/tipsheet/compute/flows_scaled.py`; bundles `flows/scaled_monthly`, `evidence/flows_scaled`.

- **Denominator:** total market cap of NSE-listed companies at the end of the **previous** month, so it is known before the flow.
  
  - Quarter-ends: implied by PRIME Database’s ownership table for all NSE-listed companies, FII value ÷ FII share by value. It matches the filings-based total (`shp_aggregate_quarterly`) within 0.5–2.0% at the five quarters where both exist.
  - Between quarter-ends: the latest quarter-end value rolled forward with the Nifty 500 TRI. The TRI includes dividends, so the roll-forward overstates price growth by about 0.1 points a month; it resets every quarter.
  - Starts June 2009 (PRIME’s first quarter).
- **FPI:** NSDL’s fortnightly equity net investment (primary and secondary market) summed into calendar months; a month counts once its month-end fortnight is in. From January 2012.
- **DII:** the exchanges’ daily DII figure summed into months. The daily series misses 33 sessions against the Nifty 500 trading calendar (9 in April 2007, 24 in 2025-26, up to four in a month). A month missing up to five sessions is **pro-rated** by sessions traded ÷ sessions observed and listed in the bundle as `prorated_dii_months`; a month still running is left out. The pre-registered tests give the same answers without the pro-rated months.
- **Twelve-month sums** need twelve complete months.
- **Pre-registered tests on these series:** C30 (persistence), C31 (coincidence with falls and DII absorption), C32 (what followed heavy foreign selling), in `docs/research/site_calculations_spec.md`.

## Mutual funds, SIPs, NPS and demat

- **Monthly net inflows by fund category,** from AMFI via India Data Hub:
  
  - equity, total, balanced advantage, arbitrage, multi-asset and hybrid
- **SIPs:**
  
  - SIP inflow, SIP AUM, and new and closed SIP accounts
  - **SIP stoppage ratio:** accounts closed ÷ new accounts in the month
    
    - **January to April 2025 are left out** of the published ratio (`sip_stoppage_ratio_pct`). In those months the registrars closed old, inactive SIP registrations in bulk: 5.2 to 16.2 million closures a month against 3.9 to 4.8 million in the three months either side, and a ratio of 353% in April 2025. Those months measure a data clean-up rather than investors stopping, so the chart’s line breaks there and the stat’s record and usual range ignore them. The raw ratio stays in `flows/funds_monthly` as `sip_stoppage_ratio_raw_pct`. The months are set in `compute/flows.py` (`SIP_CLEANUP`).
- **NPS:**
  
  - total and equity AUM, and the equity share
  - **Estimated NPS equity flow:** the change in equity AUM, minus last month’s equity AUM times the Nifty 500 total return that month. This is labelled an estimate, because NPS equity is mostly large and mid caps, not exactly Nifty 500.
- **Demat accounts:** CDSL and NSDL, total and active.

## Rates and credit

- **G-sec curve:** FBIL par yields at 1, 2, 3, 5, 7, 10, 15, 20 and 30 years, weekly from 2018-04.
  
  - Snapshots for the latest date, one month ago and one year ago.
  - **Curve slope:** 10-year minus 1-year.
- **Corporate and NBFC bond spreads over G-secs,** from 2021:
  
  - ratings AAA, AA and A
  - 3- and 5-year tenors
- **Money market:**
  
  - call rate
  - 91-, 182- and 364-day T-bill yields (carried forward between auctions)
  - commercial paper yield range
  - **Term spread:** 10-year minus 91-day.

## Known limits

- **Investor-category data is short:** under three years.
- **The DII daily series has gaps** (see above); the monthly crore tables on `/flows/` sum what is there, while the scaled series pro-rates.
- **NSDL’s FPI figure counts primary-market buying** (IPOs, QIPs); the exchanges’ DII figure is secondary market only. Comparisons between the two carry that difference.
- **The NPS flow is an estimate,** not a disclosure.
- **Spreads are India Data Hub’s computed series.** I use them as published.

## Who buys the falls

**Code:** `compute/flows_absorb.py`, step `flows_absorb`. **Bundles:** `flows/absorb/weekly`, `flows/absorb/worst`. **Page:** `/flows/#falls`.

- NSE investor-category net purchases (“Retail” is individuals trading on their own account; FPI; mutual funds), summed by week (weeks to Friday) over the sessions that have them. Each week is set against the Nifty 500 TRI’s return over those same sessions. Weeks with fewer than three such sessions are dropped. The series is near complete from December 2023.
- **Result (to 16 October 2026, 149 weeks):** individuals were net buyers in 79% of the 61 down weeks and 44% of the 88 up weeks. The same-week correlation is −0.37 (circular-shift p below 0.01). With the next week’s return it is 0.00.
- In the 10 worst weeks, individuals bought in 9, each time while foreign investors sold. Mutual funds bought more in rupees than individuals did.
- **Limits:** same-week association only. It shows who takes the other side of selling, not who is right. NSE’s categories add up to zero across buyers and sellers, so one group’s buying is always someone’s selling.

## Foreign selling by era

**Code:** `compute/fpi_proof.py`, step `fpi_proof`. **Bundle:** `flows/fpi_proof/eras`. **Page:** `/flows/foreign/#proof`.

- Built on the scaled monthly flows of C30–C32 (`flows_scaled_monthly`): NSDL FPI equity flow and exchange-reported DII flow, each in basis points of the previous month-end market value, against the month’s Nifty 500 TRI return.
- **Eras:** 2012–2019 and 2020 on. “Heavy selling” is the fifth of months with the most negative scaled FPI flow, within each era. Domestic buying is counted only in months with complete exchange data.
- **Slope:** OLS of the month’s return on the FPI flow as a percentage of market value. The difference between eras gets a moving-block bootstrap (six-month blocks, 5,000 draws, seed 20261027).
- **Result (to September 2026):** in heavy-selling months domestic institutions bought 119% of the foreign selling in 2012–2019 and 125% since 2020. The market fell in 90% and 76% of those months, by 3.0% and 3.5% on average. The slope rose from 14.7 to 26.0 points per 1% of market value, a difference of 11.3 with a 90% interval of −0.5 to 18.6. There is no sign the market has become less sensitive.
- **Limits:** this is association, not price impact, because returns also drive foreign flows. Months are few: 20 heavy months in the first era, 17 in the second.

This note is the repository file `docs/methods/flows_rates.md`, rendered as-is.

---

Historical research, not investment advice. This page in HTML: https://tipsheet.markets/methods/flows_rates/. Every published dataset: https://tipsheet.markets/data/v1/manifest.json. A map of the site: https://tipsheet.markets/llms.txt
