Efficient market hypothesis says prices fully reflect all available information | meaning returns are essentially unpredictable. Forty years of empirical research says: mostly true, but with well-documented exceptions. Those exceptions are what factor investing is built on.

This module maps the five key statistical properties of equity return time series that systematic strategies exploit, hedge against, or must account for. Each one has extensive academic documentation and observable behavior in Indian markets.

Nifty 500 daily returns | what the series actually looks like
Sample window FY04-05 to FY24-25. All figures computed on adjusted closes. Source NSE, RupeeCase research.
5185
trading days in sample
NSE calendar
0.42
lag-1 autocorrelation in absolute returns
volatility clustering marker
-0.06
lag-1 autocorrelation in raw returns
near zero as expected
7.8
excess kurtosis
fat tails vs normal 0
Raw returns look random. Squared or absolute returns do not. That gap is what every time series model tries to capture.
1
Stationarity check
ADF test, p less than 0.05 needed
2
Difference or log
price to log return series
3
Fit model
ARIMA for mean, GARCH for variance
4
Residual audit
Ljung-Box, ARCH-LM tests
5
Walk-forward
rolling out-of-sample, no peeking
Most retail backtests skip steps 1 and 4. That is how survivor-bias dressed up as alpha gets into pitch decks.
Model class share in systematic desks India
GARCH family 38%
ARIMA + exog 27%
State space / Kalman 19%
ML hybrids 16%
Nifty 50 regime composition FY04-05 to FY24-25
Low vol trend 46%
High vol trend 24%
Low vol range 18%
Crisis / gap 12%
Model choice should follow the regime share. If 58% of days are range-bound or crisis the trend-follower sits idle most of the sample.
One-month-ahead volatility forecast R squared, walk-forward 2010 to 2025
GARCH(1,1)
0.42
EGARCH
0.44
HAR-RV realised vol
0.51
Rolling 21d stdev
0.34
Naive last month
0.20
LSTM single layer
0.15
Realised volatility beats every ML headline. The boring model with economic structure wins on Indian monthly data. Source RupeeCase research, CMIE Prowess adjusted close.
TK | 2015 first Nifty GARCH fit
My first GARCH(1,1) on Nifty 50 daily returns printed alpha 0.08 and beta 0.90. Persistence 0.98. It forecast a 14.2% annualised vol cone for the next month. Actual realised for that month came in 13.9%. I thought I had cracked it. Three months later the August 2015 yuan devaluation hit, realised vol spiked to 26.4%, and my forecast cone had capped at 17%. Lesson | stationarity is a lie during regime breaks. Run a structural break test every quarter, refit after every 3-sigma residual week, and never quote one vol number without a regime label.

The five properties

1. Volatility clustering
HEDGE AGAINST

Large moves tend to be followed by large moves | in either direction. Calm periods cluster together; turbulent periods cluster together. This is the basis of GARCH models and the observation that "market stress arrives in waves." In Indian markets, the pattern is clear: the 2008 crash, 2011 correction, 2018 small-cap selloff, and 2020 COVID crash all featured days of extreme moves clustered together, not scattered randomly.

Implication: Risk management should be dynamic | position sizes should reduce during high-volatility regimes, not remain fixed.

2. Short-term mean reversion
EXPLOIT

Over horizons of 1 to 4 weeks, individual stocks that have fallen significantly tend to bounce, and vice versa. This is the basis of short-term reversal strategies. The mechanism is microstructure-driven: market makers and liquidity providers absorb large directional flows, then prices revert as the temporary imbalance unwinds.

Implication: This is why momentum strategies skip the most recent month (the 12-1 construction). The last month's return predicts reversal, not continuation.

3. Medium-term momentum
EXPLOIT

Over 3 to 12 month horizons, winners continue to outperform and losers continue to underperform. This is the momentum factor | documented by Jegadeesh & Titman (1993) in US markets, subsequently replicated in Indian markets. The mechanism is behavioral: investors underreact to fundamental information, so strong earnings or business improvement takes months to be fully priced in.

Implication: The 12-1 momentum signal (12-month return, skipping last month) is the canonical implementation. NSE Momentum Index tracks this systematically.

4. Long-term reversal
EXPLOIT

Over 3 to 5 year horizons, prior losers outperform prior winners | the opposite of momentum. De Bondt & Thaler (1985) documented this in the US. The mechanism is overreaction: investors become overly pessimistic about long-term losers and overly optimistic about long-term winners, creating mean reversion over multi-year horizons. This is partly why value investing works: buying long-term underperformers captures the reversal premium.

Implication: Momentum and contrarian strategies can coexist | they operate at different time horizons.

5. Calendar effects
BE AWARE

Certain calendar periods show systematically different returns | the "January effect" (small caps outperform in January) in the US, budget-season volatility in India, FII rebalancing around quarter-end. Some of these are robust; others are data-mined artifacts. Indian-specific: pre-budget volatility (January to February), dividend stripping effects in August to September, and year-end FII position squaring in March.

Implication: Calendar effects are real but small. They're worth being aware of for execution timing but too unstable to build primary strategies around.

GARCH: modelling volatility clustering

GARCH (Generalised Autoregressive Conditional Heteroskedasticity) models are the standard tool for modelling time-varying volatility. The core idea: today's volatility depends on both yesterday's volatility and yesterday's return shock.

GARCH(1,1) Model
σ²t = ω + αε²t-1 + βσ²t-1
σ²t = today's variance. ε²t-1 = yesterday's squared return shock (news impact). σ²t-1 = yesterday's variance. α captures how quickly volatility responds to new information; β captures how persistent high volatility is. For Indian equities, β is typically high (0.85 to 0.92), meaning volatility shocks are persistent.

For practical systematic investing, GARCH models are most useful for dynamic position sizing | reducing exposure when estimated volatility is high, increasing it when volatility is low. This is not market timing (you're not predicting returns), just volatility-adjusted sizing.

India-specific note: NSE data from 2003 gives roughly 20 years of usable daily price history | about 5,000 trading days. This is sufficient for robust GARCH estimation, but means you have at most 3 to 4 complete market cycles (bull + bear) in the sample. Structural regime changes (2010 liquidity growth, 2018 SEBI re-categorisation) should be accounted for.

Mean reversion vs trend following: the regime question

Strategies built on mean reversion (contrarian, value) and strategies built on trend following (momentum) both have empirical support. They don't always work simultaneously | there are periods where one dominates.

Time series tools on RupeeCase

The RupeeCase terminal shows rolling 252-day volatility for each strategy and the underlying index | allowing you to see when you're in a high-volatility regime. The backtester uses 20-day rolling volatility for inverse volatility weighting. Strategy return autocorrelation is computed and displayed to help identify when a strategy is behaving differently than its historical pattern. Available at invest.rupeecase.com.

Glossary

Key terms from this module
Volatility clustering
The empirical tendency for periods of high market volatility to cluster together. Captured formally by GARCH models.
Short-term reversal
The tendency for stocks with extreme returns over 1 to 4 weeks to reverse direction. Driven by microstructure effects. The reason momentum skips the last month.
GARCH
Generalised Autoregressive Conditional Heteroskedasticity. A family of models for time-varying volatility that captures persistence and mean reversion in variance.
Long-term reversal
Over 3 to 5 year horizons, prior losers outperform prior winners. Related to overreaction and mean reversion in fundamentals. Partly explains why value investing works.
Non-stationarity
A time series whose statistical properties change over time. Most financial price series are non-stationary; return series are closer to stationary but still exhibit regime changes.
TK
A note from the author
Why this matters

Time series analysis was the first lens that made Indian market data truly speak to me. Once you understand autocorrelation, seasonality, and stationarity in the context of Nifty and sector indices, you stop guessing and start modelling. Every rupee I have deployed systematically began with a time series insight.

TK
Tanmay Kurtkoti
Founder & CEO, RupeeCase · 17 years systematic trading · QC Alpha
Want to put this into practice? RupeeCase is the systematic investing terminal built around everything you're learning here, factor scores, strategy backtests, portfolio construction for Indian markets.
Explore the terminal →

Sources & further reading

  • → Engle, R.F. (1982). Autoregressive Conditional Heteroscedasticity. Econometrica. (Original ARCH paper)
  • → Jegadeesh, N. & Titman, S. (1993). Returns to Buying Winners and Selling Losers. Journal of Finance.
  • → De Bondt, W. & Thaler, R. (1985). Does the Stock Market Overreact? Journal of Finance.
  • NSE Nifty Momentum Index methodology

Quick check, Module 5.2

0 correct · 0 answered
🎉
Module 5.2 complete
3 correct. Continue to Module 5.3 when ready.
Research Lab Qualifier
Path 5, Module 2 of 5 done, complete all 5 + path test to unlock
✅ 5.1 Statistics 📍 5.2 Time Series 5.3 ML for Alpha 5.4 Alt Data 5.5 Out-of-Sample
← Previous
Previous, Module 5.1
Statistical Foundations
Calculator

Rolling Correlation Regime Detector

Indian equity correlation jumps in stress. A jump from baseline to current is a regime signal.

Quick check, Module 5.2

3 questions. Get 2 right to mark this module complete.

0 of 3 answered
Up next, Module 5.3
Machine Learning for Alpha
What ML can and can't do in systematic equity investing | signal-to-noise, effective sample size, which tools actually work in practice.
Continue →
PRACTICE WHAT YOU LEARNED
Try systematic strategies on RupeeCase | free paper trading.
Get Started Free →