The world and India: method
Written 2026-10-07, before any outcome below was computed. Page: /global/india/. Spec: docs/plans/world_and_india.md.
Code: pipeline/tipsheet/compute/world_india.py. The readings, thresholds and verdict rule below are fixed here and are
not to be tuned after seeing results.
The five readings
All monthly, taken at month end (the month’s average for monthly series).
| Row | Reading | Measured as | Source | From |
|---|---|---|---|---|
| US interest rates | US 10-year real (inflation-protected) Treasury yield | level | US Treasury, FRED DFII10 | 2003 |
| The dollar and the rupee | The rupee’s value in dollars, change on 12 months earlier; the broad dollar’s change beside it | change | Federal Reserve H.10, FRED DEXINUS, DTWEXBGS | 2000 (dollar index 2006) |
| Oil | Brent in rupees, change on 12 months earlier | change | World Bank Pink Sheet, times DEXINUS | 2000 |
| Fear in global markets | OFR Financial Stress Index | level | US Office of Financial Research | 2000 |
| World growth | Share of economies whose OECD leading indicator is rising (3 months) | level | OECD, via the Monitor | 2005 |
Levels are used for readings that swing around an average. Changes are used for readings that trend, because a trending level is always near its own record and says nothing.
States, without look-ahead
Each month, a reading is high when it is at or above the 75th percentile of all its readings up to and including that month, low at or below the 25th, and normal in between. Only data available at the time is used (an expanding window), and a reading needs 60 months of history before it is classified. For the rupee, “high” means the rupee rose most; for stress, “high” means most stress.
The outcome
The Nifty 50 total return index in rupees, the return over the next three months (from the end of month t to the end of month t + 3). For each reading and state: the median, the middle half (25th to 75th percentile), the number of months, and the number of separate spells (runs of consecutive months in the state, since the months inside a spell overlap and are not independent). The same figures for all months are the reference.
The verdict rule
A row is called a headwind (or tailwind) only when both hold:
- The median three-month return in its high state and in its low state differ by more than 3 percentage points.
- Both states have at least four separate spells.
Otherwise the page says: “Since 2003, Indian returns looked much the same whatever this reading was.” The direction word follows the sign: if returns were lower when the reading was high, high is the headwind.
Limits
About 280 months from 2003, and far fewer independent spells. Overlapping three-month windows overstate the evidence, which is why spells are counted. This is a description of the past, not a forecast. FPI flows are shown as an outcome on the page, not a cause: earlier tests on this site found that FPI flows follow Indian returns rather than lead them.