NIFTY 50 Seasonal Analysis by Industry & SARIMA

Indian large-cap equities exhibit recurring calendar structure that is not uniform across industries. The NIFTY 50 benchmark aggregates fifty liquid NSE-listed names, yet cyclical banks, energy producers, and consumer discretionary stocks often respond differently to the same fiscal calendar than defensives such as pharmaceuticals, utilities, and information technology. This report quantifies those differences using a transparent, reproducible pipeline: equal-weight industry baskets, month-by-month return calendars, cyclical-versus-defensive cycle spreads, and seasonal autoregressive integrated moving-average (SARIMA) diagnostics on the index itself.

The central empirical question is whether the Indian fiscal-year rhythm — particularly the contrast between the December–March window (fiscal Q4) and the April–September stretch spanning fiscal Q1 and Q2 — leaves a measurable imprint on both the benchmark and its constituent industries. We test the null that neither the index nor any industry bucket exhibits structured seasonality across these windows, against the alternative that cycle-sensitive sectors rally more strongly in April–September while showing relative weakness or consolidation in December–March. The analysis spans a long history of daily adjusted prices, maps each constituent to a research industry label, and reports both aggregate sector behaviour and ticker-level seasonality for drill-down.

Calendar effects in equity markets have been documented globally since at least Rozeff and Kinney (1976). In India, several institutional features plausibly reinforce industry-level seasonality rather than a single market-wide drift. Union budget announcements in February, advance-tax and year-end portfolio rebalancing by domestic institutions, foreign portfolio investor flows tied to global risk appetite, and monsoon-linked sentiment for agriculture-adjacent sectors all create recurring shocks whose impact may concentrate in particular months and industries.

Unlike a single-factor market model, this study treats each industry as an equal-weight basket of Nifty constituents assigned to research sectors (Financials, IT, Energy, Consumer, Materials, Industrials, Healthcare, Utilities, Telecom, and Conglomerate). Formally, if $r_{i,t}$ denotes the total return of stock $i$ on day $t$, calendar seasonality at the industry level implies that the conditional expectation $\mathbb{E}[r_{s,t}\mid \mathrm{month}(t)=m]$ varies with calendar month $m\in \{1,\dots ,12\}$ for sector basket $s$, even when the unconditional daily mean is near zero. The benchmark index provides the reference calendar ${\bar{r}}_{I,m}$ against which each sector's monthly profile is compared.

We evaluate three complementary propositions throughout the report. Industry heterogeneity (H₁): equal-weight sector baskets display materially different twelve-month calendar profiles; seasonality dispersion ${\sigma }_s^{\mathrm{season}}=\mathrm{std}({\bar{r}}_{s,1},\dots ,{\bar{r}}_{s,12})$ differs across sectors. Cyclical amplification (H₂): macro-sensitive cyclical industries exhibit a larger April–September minus December–March return spread than defensive industries: $\mathbb{E}[S_s\mid \mathrm{cyclical}]>\mathbb{E}[S_s\mid \mathrm{defensive}]$, where $S_s={\bar{r}}_{s,Q1Q2}^{\mathrm{ann}}-{\bar{r}}_{s,Q4}^{\mathrm{ann}}$. Benchmark co-movement (H₃): sectors whose business models track broad domestic growth (Consumer, Financials) show higher correlation between their monthly calendar vector and the Nifty calendar than idiosyncratic industries.

For the index time-series block, the stationarity null is $H_0:\Delta x_t$ contains a unit root, tested via the Augmented Dickey–Fuller statistic on first-differenced log prices $x_t=\ln P_t$. Rejection supports modelling $\Delta x_t$ with SARIMA dynamics. The seasonal structure null remains $H_0$: $\mathbb{E}[r_{I,t}\mid t\in Q4]=\mathbb{E}[r_{I,t}\mid t\in Q1Q2]$ for the benchmark; the alternative allows asymmetric drift across Indian fiscal windows.

Returns and sector baskets. For constituent $i$ with adjusted close $P_{i,t}$, the daily simple return is $r_{i,t}=P_{i,t}/P_{i,t-1}-1$. Constituents are grouped into sector $s$ with membership set ${\mathcal{N}}_s$. The equal-weight sector return on day $t$ is:

$$r_{s,t}=\frac{1}{|{\mathcal{N}}_s|}\sum _{i\in {\mathcal{N}}_s}{r}_{i,t}$$

Sector wealth indices compound these daily returns; calendar-month total returns aggregate daily compounding within each month.

Monthly calendar statistics. For each sector $s$ and calendar month $m$, the unconditional sample mean and hit rate are:

$${\bar{r}}_{s,m}=\frac{1}{N_m}\sum _{t:\mathrm{month}(t)=m}{r}_{s,t},H_{s,m}=\frac{1}{N_m}\sum _{t:\mathrm{month}(t)=m}1\{r_{s,t}>0\}$$

Relative performance versus the Nifty benchmark isolates industry rotation:

$${\Delta }_{s,m}={\bar{r}}_{s,m}-{\bar{r}}_{I,m}$$

Calendar-profile correlation measures how closely a sector's twelve-month rhythm tracks the index:

$${\rho }_s=\mathrm{corr}(({\bar{r}}_{s,1},\dots ,{\bar{r}}_{s,12}),({\bar{r}}_{I,1},\dots ,{\bar{r}}_{I,12}))$$

Fiscal window spreads. Following Indian market convention, Q4 comprises December through March and Q1/Q2 spans April through September. For any return series, the annualized return over window $W$ with $T_W$ trading days uses compound scaling:

$${\bar{r}}_W^{\mathrm{ann}}={\left(\prod _{t\in W}(1+r_t)\right)}^{252/T_W}-1$$

The headline cycle spread for sector $s$ is $S_s={\bar{r}}_{s,Q1Q2}^{\mathrm{ann}}-{\bar{r}}_{s,Q4}^{\mathrm{ann}}$. Cyclical and defensive aggregates report the equal-weight mean of $S_s$ within each macro group.

Index stationarity and SARIMA. Let $x_t=\ln P_t^{\mathrm{Nifty}}$. First differences $\Delta x_t=x_t-x_{t-1}$ are tested for stationarity via ADF; if $H_0$ is rejected, seasonal ARIMA models are fit separately on Q4 and Q1/Q2 subsamples using specification SARIMA$(1,1,1)\times (1,1,0{)}_{12}$:

$$(1-\varphi B)(1-B)(1-\Phi B^{12})(1-B^{12})y_t=(1+\theta B)(1+\Theta B^{12}){\epsilon }_t$$

where $B$ is the backshift operator, $y_t=\Delta x_t$, and ${\epsilon }_t$ is white noise. Models are compared by Akaike (AIC) and Bayesian (BIC) information criteria alongside root mean squared error (RMSE) on each subsample.

The investable universe is the current NIFTY 50 membership screened for sufficient price history so that monthly and seasonal statistics are estimated on stable samples. Each surviving constituent maps to exactly one research industry label; names with fragmented listing history are excluded rather than forward-filled, preserving a clean point-in-time panel.

The empirical pipeline proceeds in four stages. First, daily adjusted closing prices for all constituents and the benchmark index are aligned on a common trading calendar from September 2007 onward, producing the return panel $\{r_{i,t}\}$. Second, equal-weight sector baskets $\{r_{s,t}\}$ are constructed and their monthly calendars, hit rates, and benchmark-relative returns ${\Delta }_{s,m}$ are tabulated. Third, pairwise correlation of sector monthly calendars quantifies which industries move together through the year. Fourth, log-differenced Nifty prices undergo ADF testing; subsamples restricted to Q4 and Q1/Q2 each receive an independent SARIMA fit so that information criteria reveal whether the spring–summer rally window is more linearly predictable than the winter window.

Cyclical sectors (Financials, Energy, Materials, Industrials, Consumer, Conglomerate) are contrasted with defensives (Healthcare, Utilities, IT, Telecom) by averaging Q4 and Q1/Q2 annualized returns within each group. Ticker-level tables replicate the sector metrics for individual constituents, enabling inspection of whether a sector's aggregate pattern is driven by one dominant name or broad participation.

The industry seasonality table ranks sectors by the Q1/Q2 minus Q4 spread $S_s$, alongside calendar correlation ${\rho }_s$ with the Nifty monthly profile. A high ${\rho }_s$ indicates the sector tends to rise and fall in the same calendar months as the broad index; a large positive $S_s$ signals historically stronger April–September performance relative to December–March.

The monthly return heatmap visualizes ${\bar{r}}_{s,m}$ across all sectors simultaneously. Scanning rows reveals industry-specific peaks (for example, budget-adjacent months for Financials) while scanning columns highlights market-wide month effects shared across industries.

Sector drill-down charts plot ${\bar{r}}_{s,m}$ and ${\Delta }_{s,m}$ for a selected industry, making rotation versus the benchmark visible month by month. The constituent table extends the same metrics to individual tickers within the sector.

The correlation matrix reports $\mathrm{corr}({\bar{r}}_{s,m},{\bar{r}}_{s^{\prime },m})$ across sector pairs — useful for identifying clusters of industries whose calendar cycles move in tandem. The benchmark price chart and seasonal subsample overlays contextualize where Q4 and Q1/Q2 windows sit in the long index history. The SARIMA scorecard compares AIC, BIC, and RMSE across windows; lower information criteria on the Q1/Q2 subsample suggest the upward seasonal segment is more parsimoniously modelled, though economic narratives (budget cycles, global liquidity) remain essential for interpretation.

All reported figures are unconditional sample means over the estimation window. They describe historical calendar drift, not risk-adjusted alpha, and should not be treated as live trading signals without independent validation.

Seasonal splits follow the Indian fiscal-quarter convention (Q4 = December–March). Structural breaks — demonetisation, the COVID shock, sharp rate cycles, or geopolitical dislocations — can temporarily invert or erase calendar patterns that held in earlier decades. Unconditional monthly means conflate multiple macro regimes; stability should be assessed with rolling or sub-period analysis before deployment.

SARIMA models assume linear, stationary dynamics after differencing. Policy surprises, liquidity stress, or index reconstitution events may violate these assumptions for short intervals. Equal-weight sector baskets treat all constituents equally; they do not reflect free-float or liquidity weights used by the official index.

Industry labels are research classifications, not exchange GICS codes. A conglomerate with diversified operations may be assigned a single bucket, understating cross-industry exposure. Constituent membership changes over time; the panel uses current index names with full available history, introducing a mild survivorship bias relative to a point-in-time backtest.

This document is research output for education and investment-committee discussion. It is not investment advice, and past calendar regularities do not guarantee future performance.

The interactive section below reproduces the full estimation output: headline cycle metrics, industry seasonality tables, monthly heatmaps, sector drill-down charts, constituent-level seasonality, inter-sector correlation, benchmark price history, Q4 versus Q1/Q2 subsample overlays, and SARIMA diagnostics. Dynamic summary paragraphs reflect the latest estimation date and sample composition.