What happened to the NIFTY Financial Services Index around the Union Budget and RBI MPC announcements?

Deepak Kambala
JNTUK University College of Engineering Narasaraopet

Question

This page explains what the paper and its notebook calculate when they compare the NIFTY Financial Services Index with its own recent return history around selected Reserve Bank of India (RBI) Monetary Policy Committee (MPC) decisions and Union Budgets. It is a compact, reproducible case study in reading an event study: what was measured, how the comparison works, and what the results can and cannot establish.

The largest event-day abnormal return in the six analysed events is −2.264% for Union Budget 2026. The notebook’s event-day statistic is t = −4.159, with a two-sided p = 0.0003. This says the event-day move was large relative to the notebook’s estimated recent variability under its test assumptions. It does not isolate the Budget’s causal effect from other information or market forces.

Read this first. An abnormal return is a comparison with a modelled baseline, not a directly observed “effect” of a policy. The index moved; the study subtracts its average return in a preceding window to describe how unusual that movement was against this specific benchmark.

Data and events

The dataset contains daily open, high, low, close, volume and turnover fields for the NIFTY Financial Services Index. The paper describes the sample as 31 March 2025 to 31 March 2026 with 247 trading days; the CSV and notebook data run from 1 April 2025 to 30 March 2026. Returns are arithmetic percentage changes in the closing level.

Events analysed. Event day t₀ is the announcement date or, if that is not a trading day, the next trading day.
Event	Announcement date
RBI MPC, June 2025	6 Jun 2025
RBI MPC, August 2025	6 Aug 2025
RBI MPC, October 2025	1 Oct 2025
RBI MPC, December 2025	5 Dec 2025
Union Budget 2026	1 Feb 2026
RBI MPC, February 2026	6 Feb 2026

The paper identifies seven relevant policy events, but analyses six. The April 2025 MPC decision is excluded because there are not enough earlier observations for the 30-day estimation window.

What is in the index data?

The CSV contains one daily row with date, open, high, low, close, shares traded and turnover. The notebook sorts rows chronologically and calculates returns from consecutive closing levels. The paper describes the index as tracking 20 large NSE-listed financial-services companies, weighted by free-float market capitalisation. This is an index-level study: it does not estimate separate reactions for individual banks, insurers or other constituents.

Why these announcements?

The paper selects scheduled fiscal and monetary announcements because they can convey information relevant to financial firms. RBI policy can affect borrowing costs and interest-rate expectations; budget measures can affect taxes, public spending and financial-market rules. These are possible transmission channels described in the paper, not mechanisms separately tested by this dataset.

Method

The design follows the standard event-study framework described by Brown and Warner (1985) and MacKinlay (1997). It uses a mean-adjusted return model: the expected return during an event window is the index's mean return in a preceding estimation window. No separate market benchmark is used.

Figure 1. Window layout used by the notebook. In notation, L=[t_0-35,t_0-6] supplies 30 returns and W=[t_0-5,t_0+5] contains 11. The windows are adjacent and do not overlap.

Event day and windows

For each announcement, the notebook locates the announcement date in the trading calendar; if it is not a trading day, it uses the next available trading day. Relative day 0 is that selected date. Day −1 and +1 mean the preceding and following observations in the dataset, not necessarily calendar days.

t_0 = \min\{\, t \in \mathcal{T} \mid t \ge t_{\text{announced}} \,\}

W = \{t : t_0-5 \le t \le t_0+5\}, \qquad L = \{t : t_0-35 \le t \le t_0-6\}

\mathcal{T} is the set of trading days in the dataset and t_{\text{announced}} the official date. The rule prevents the event day from preceding the release. The event window contains 11 observations: five before, the event day, and five after. Its pre-event portion can include anticipation or other news; the post-event portion can include follow-up information.

Abnormal and cumulative abnormal returns

R_t = \frac{\text{Close}_t - \text{Close}_{t-1}}{\text{Close}_{t-1}} \times 100, \qquad \bar R_i = \frac{1}{30}\sum_{t\in L} R_{it}

AR_{it} = R_{it} - \bar R_i, \qquad CAR_i(-5,+5) = \sum_{t=-5}^{+5} AR_{it}

A negative AR_{it} means the index returned less than its estimation-window average. CAR_i is the sum across all 11 trading days, including the five days before the event.

Small example. If a day’s observed return is +0.50% and the preceding-window average is +0.10%, its abnormal return is +0.40 percentage points. It does not mean the index itself rose by 0.40%; that was +0.50%. Because this is a mean-adjusted model, the “normal” return is a historical average for this index, not a forecast based on the whole market or a control index.

Volatility

\sigma_{i,\text{pre}} = \operatorname{std}\{R_t : t\in[t_0-35,\,t_0-6]\}, \quad \sigma_{i,\text{post}} = \operatorname{std}\{R_t : t\in[t_0,\,t_0+5]\}, \quad \Delta\sigma_i = \sigma_{i,\text{post}}-\sigma_{i,\text{pre}}

The post-event volatility uses the event day and the following five trading days. The paper also plots a rolling 30-observation sample standard deviation, s_{30,t}=\operatorname{sd}(R_{t-29},\ldots,R_t), across the full sample. Each plotted value summarizes the most recent 30 daily returns, so it changes as the window advances.

Significance tests

Test 1 is a one-sample t-test on the 11 daily abnormal returns in W. It tests whether their sample mean is zero; it does not test the full-window CAR directly:

t_1=\frac{\overline{AR}_{i,W}}{s_{AR,i,W}/\sqrt{11}},\qquad p_1=2\Pr\!\left(T_{10}\geq |t_1|\right)

Test 2 is called a Patell test in the paper, but the notebook’s implemented calculation is the event-day abnormal return divided by the estimation-window standard deviation:

t_2=\frac{AR_{i0}}{s_{i,\text{pre}}},\qquad p_2=2\Pr\!\left(T_{29}\geq |t_2|\right)

The notebook uses a two-sided Student’s t reference distribution with 29 degrees of freedom, based on 30 estimation-window returns. For the Budget, −2.264 / 0.544 is approximately −4.16. This page reports the notebook’s implemented statistic and p-value. The paper’s “Patell” label should be treated cautiously: the code shown does not implement a market model or a broader causal test.

Results

Table 1. Event-day and window results as reported in the paper (Tables 6.1 and 6.4) and reproduced in the notebook output. All values in percent except test statistics and p-values.
Event	Baseline \bar{R}	AR_0	CAR(-5,+5)	\sigma_{\text{pre}}	\sigma_{\text{post}}	\Delta\sigma	T1 p	T2 t	T2 p
MPC Jun 2025	+0.161	1.589	−2.278	1.133	1.010	−0.123	0.3916	1.402	0.1714
MPC Aug 2025	+0.034	−0.042	−2.099	0.697	0.750	0.053	0.2841	−0.061	0.9518
MPC Oct 2025	+0.052	1.332	0.078	0.620	0.669	0.049	0.9743	2.149	0.0401
MPC Dec 2025	+0.138	0.842	−2.484	0.606	0.701	0.095	0.2251	1.389	0.1754
Union Budget 2026	−0.047	−2.264	3.015	0.544	1.798	1.253	0.5374	−4.159	0.0003
MPC Feb 2026	−0.005	0.430	3.018	0.625	0.695	0.070	0.5193	0.688	0.4969

Table guide. \bar{R} is the mean daily return across the 30-day estimation window. AR_0 and CAR(-5,+5) are percentage returns; volatility columns are standard deviations of daily returns, also in percent. \Delta\sigma=\sigma_{\text{post}}-\sigma_{\text{pre}} is therefore a percentage-point difference. T1 p is the window test’s p-value; T2 t and T2 p are the notebook’s event-day statistic and p-value.

At the conventional 5% threshold, Test 2 rejects its zero-abnormal-return null for the Budget (p = 0.0003) and October 2025 MPC (p = 0.0401). Only the Budget result is below 1%. Test 1 does not reject for any event. These are nominal p-values: the notebook reports 12 tests (two for each of six events) and does not adjust for multiple comparisons.

For the Budget, post-event daily-return standard deviation is 1.798%, compared with 0.544% in the estimation window: \Delta\sigma=+1.253 percentage points. This is a descriptive comparison of two short windows; the notebook does not run a formal test of the volatility difference. Its 11-day CAR(-5,+5) is +3.015% even though event-day AR_0 is −2.264%. The other ten days therefore sum to about +5.279 percentage points of abnormal return. This arithmetic describes the window; it does not establish that the Budget caused a recovery.

What the numbers mean

Use the measures together. AR_0 describes one day relative to the estimated baseline; CAR(-5,+5) describes the sum across the complete 11-day window; volatility compares the spread of daily returns in two windows. None is a direct measure of investor beliefs or policy success.

How to read one row

Take the October 2025 MPC row. The index’s event-day return exceeded its preceding 30-day mean by 1.332 percentage points (AR_0). Across all 11 days, the sum of abnormal returns was +0.078%, close to zero (CAR(-5,+5)). The event-day test gives p_2=0.0401, but the 11-day mean test gives p_1=0.9743. So this analysis flags a relatively unusual event-day observation under Test 2, while it finds no evidence that the average AR across the whole window differs from zero under Test 1. The tests address different summaries of the data.

What the Budget row says

The event-day AR_0 is negative and unusually large relative to the notebook’s baseline. The full-window CAR(-5,+5) is positive because gains in abnormal returns on other days outweigh the event-day decline. Post-event volatility is also markedly higher than the pre-event estimate. Together these values describe a sharp one-day move and a choppy following window; they do not identify which news drove either pattern.

Event-day significance and window significance disagree

For the Budget, Test 2 is significant (p = 0.0003), while Test 1 is not (p = 0.5374). The tests answer different questions: whether the event-day abnormal return was unusual relative to the estimation-window variation, and whether the mean abnormal return across the full 11-day window differed from zero.

Closely spaced dates

The Budget date, 1 February 2026, and MPC date, 6 February 2026, fall within five calendar days. Their trading-day windows overlap substantially. Returns in that shared period contribute to both event summaries, so the two CARs are not independent observations and cannot cleanly separate the two announcements’ effects.

Limits of this analysis

Small event count: six analysed events, with only 11 daily observations in each event window. A non-significant result means this test did not reject its null; it does not prove there was no market response.
No control benchmark: the mean-adjusted model subtracts the index’s prior average return but does not remove market-wide or other sector-wide movements.
Announcements are not surprises: the data do not measure what markets expected beforehand or how much new information each announcement contained.
Other news may coincide: the design does not control for simultaneous company, economic or global news.
Overlapping February windows: Budget and MPC measurements share event-window observations.
Statistical assumptions: the notebook uses parametric t-tests on short daily-return samples. It reports no correction for the 12 comparisons. The volatility changes are not accompanied by a significance test.
One index and one year: results describe this index in the available sample, not every financial firm, future policy announcement or other market.

Source notes

For transparency, the paper and the executable notebook describe the inputs and method slightly differently in places:

Sample period. Paper: 31 Mar 2025 to 31 Mar 2026. Notebook output: 1 Apr 2025 to 30 Mar 2026, 247 trading days.
Event count. The paper says seven relevant events were identified and six were analysed. The notebook's event dictionary also includes the February 2025 Union Budget, which is outside the sample and produces no usable result.
Test 2 description. The paper labels Test 2 a Patell test. The notebook computes event-day abnormal return divided by the sample standard deviation in the estimation window, then uses a two-sided Student’s t p-value with n − 1 degrees of freedom. This implementation detail is stated explicitly here because the label alone does not describe the code’s calculation.
Rounding. The Budget's pre-event volatility appears as 0.544% in the results table and about 0.545% in the discussion; the difference is rounding.
Interpretation. The paper discusses possible explanations for the Budget reaction, but the event study does not separately test those explanations.

References

D. Kambala. An Event Study of Abnormal Returns and Volatility Patterns in India's NIFTY Financial Services Index Around Union Budget and RBI Monetary Policy Committee Announcements (FY2026). SSRN 6518198.
S. J. Brown and J. B. Warner (1985). Using daily stock returns: The case of event studies. Journal of Financial Economics 14, 3–31.
A. C. MacKinlay (1997). Event studies in economics and finance. Journal of Economic Literature 35, 13–39.
J. A. Patell (1976). Corporate forecasts of earnings per share and stock price behavior: Empirical test. Journal of Accounting Research 14, 246–276.

Primary materials: research paper PDF, analysis notebook, and CSV data. Numerical results above follow notebook output where the paper’s description and code differ. The added explanations clarify calculations and limitations; they do not add new empirical findings.