India GDP NowCast — Reference Guide.

01 What this is

India’s official GDP is published by the Ministry of Statistics (MoSPI) roughly two months after a quarter ends — the first estimate for April–June, for instance, lands around the end of August. A nowcast fills that gap: it reads a panel of timely monthly indicators and translates them into an early estimate of the current quarter’s GDP growth, updated as new data arrives.

This tool nowcasts both real GDP (volume, inflation-stripped) and nominal GDP (value, including prices), year-on-year. It is a research read for getting ahead of the print — not a replacement for the official figure.

02 The method — combination of bridge equations

A bridge equation is a simple regression that “bridges” a quarterly target (GDP) to higher-frequency monthly indicators, aggregated up to the quarter, plus a lag of GDP itself. A single large regression with many indicators overfits badly on a short sample like India’s — so instead the model estimates every one- and two-indicator bridge and averages their forecasts. That combination is far more robust when the sample is short and relationships shift, which is exactly the setting here.

Three further pieces make it work in real time:

• Ragged-edge extrapolation. Within the current quarter, indicators that haven’t printed yet are extrapolated, so a nowcast can be produced before all three months are in. The report’s “hard data in hand” figure tells you how much is actual versus extrapolated.
• COVID handling. The 2020–21 shock distorts any model estimated across it. The tool tests three treatments — no adjustment, a dummy variable, and deleting the COVID quarters — and keeps whichever performs best in backtest.
• Confidence bands. The shaded band on each chart is ± the mean absolute error (MAE) of the model’s own past nowcasts — roughly a 57% range — following the Federal Reserve’s approach to uncertainty (Reifschneider & Tulip, 2019).

03 The indicators

Seven monthly indicators feed the nowcast, split into two clusters. The volume cluster drives real GDP; the value cluster drives nominal GDP.

* PMI is licensed data (S&P Global); the published headline is read and entered by hand.  † NPCI blocks automated access, so UPI is entered by hand each month. All other series update automatically from their primary sources.  ‡ PV & 2W retail (FADA / VAHAN registrations) are monitored, not modelled — collected each month and shown for context, but deliberately excluded from the bridge equations. On this sample they carry no measurable predictive signal for real GDP (correlation +0.14 and +0.08 respectively, versus +0.77 for ICI), and adding them lowered like-for-like backtest accuracy. An apparent skill gain from including them proved to be an artifact of their shorter history: two columns of random noise, given the same missing-data pattern, reproduced it exactly.

04 Reading the output

Median nowcast

The central estimate — the dashed line and labelled point on each fan chart.

Central band (±MAE)

The shaded fan: the ~57% confidence range around the median, from the model’s own past error.

Drop-a-group range

How far the nowcast moves if any single indicator is removed — a sensitivity check. Tight = robust; wide = leaning on specific series.

Hard data in hand

Share of the quarter’s three months already published, versus model-extrapolated. Higher = less rests on extrapolation.

R² (out-of-sample)

Share of the variation in actual GDP the nowcast explains in backtest, versus a naive running-average forecast. Higher = more genuine skill.

Correlation

How closely the backtested nowcasts track the eventual actuals (1.0 = perfect co-movement).

Directional hit-rate

How often the nowcast gets the direction right — whether growth rose or fell versus the prior quarter.

RMSE

Root Mean Square Error — the typical size of a miss, penalising large errors more heavily.

Z-score panel

How far each indicator sits from its own long-run average, in standard deviations — a model-free read on the raw data.

YoY

Year-on-year: change versus the same quarter one year earlier.

05 A note on the two reads

Real GDP is inherently hard to nowcast. It moves in a narrow band quarter to quarter, so even a good model struggles to beat a simple average on level — the skill table shows a modest R². What it does well is direction (high hit-rate). Treat the real read as “near trend, leaning this way,” not a precise point.

Nominal GDP carries more skill (higher R²), and its three value indicators — GST, bank credit and UPI — now point the same way. An earlier version of this guide warned that GST was dragging the nominal headline down; that turned out to be a data artifact, not a real signal. The gst.gov.in workbook’s frozen monthly sheets never absorbed the government’s later downward revisions to the prior-year base, so GST year-on-year read falsely negative. The pipeline now rebuilds GST on the most-revised base automatically, and on that footing GST is firmly positive and in line with bank credit and UPI — so nominal reads clean, not a headline you have to look through. The September 2025 rate cut still weighs modestly, but that residual softness laps out once it annualises in October 2026.

06 Caveats

• The backtest is pseudo real-time — it uses today’s data with historical publication lags but not true data vintages — so the out-of-sample error reads slightly optimistic.
• MoSPI revises prior quarters and periodically rebases, so the target itself moves.
• The nowcast is only as good as its inputs; a single mis-keyed manual figure (PMI, UPI) flows straight through. Sanity checks catch gross errors, not subtle ones.
• This is a research tool, not investment advice.

07 References