Ziyuan Wang
School Project

HKU | STAT3609

Examining Lunar New Year Effects on the Hong Kong Stock Market (2015-2025)

Team-based statistical study on whether the Lunar New Year (LNY) period creates measurable return and risk anomalies in the Hang Seng Index. The project focused on pre-LNY, post-LNY, and baseline windows.

What We Did

  • Collected HSI market data from Yahoo Finance and aligned 2015-2025 LNY windows.
  • Tagged each day as Pre-LNY, Post-LNY, or Baseline.
  • Computed simple/log returns and risk metrics: volatility, VaR(95%), CTE(95%), Sharpe ratio.
  • Ran statistical tests: two-sample t-tests (means), Levene's test (variances), Jarque-Bera normality, and dummy-variable regression.
  • Compared findings against prior LNY literature in Hong Kong and East Asian markets.

Dataset and Scope

  • Period: Sep 1, 2015 to Sep 1, 2025.
  • Core series: Hang Seng Index daily price data.
  • Sample sizes: Pre-LNY 39, Post-LNY 17, Baseline 2402.

Key Results

Risk/Return Profile

  • Pre-LNY average return (0.00271) was above baseline (0.00013).
  • Post-LNY showed lower return and higher volatility versus baseline.
  • Combined LNY window had higher average returns but deeper downside tail risk than baseline.

Sharpe Ratio

  • Pre-LNY: 0.1939 (best risk-adjusted performance).
  • Post-LNY: -0.0071 (risk not compensated by return).
  • LNY Avg: 0.1163, Baseline: 0.0005.

Hypothesis Testing

  • Mean return tests (t-tests): no pair was statistically significant at 5%.
  • Variance tests (Levene): no pair was statistically significant at 5%.
  • Regression dummy coefficients for Pre/Post-LNY were not significant.

Interpretation

  • Descriptive/economic patterns align with LNY holiday effect literature.
  • Formal significance was weak mainly due to very small LNY window samples.
  • Conclusion: evidence is suggestive, but not statistically conclusive in this setup.

Limitations and Next Steps

  • Severe sample imbalance reduced statistical power (Type II error risk).
  • Baseline returns were strongly non-normal, weakening parametric test reliability.
  • Simple tests/regression may miss dynamic effects (autocorrelation, volatility clustering, asymmetry).
  • Next step: use GARCH/TGARCH/EGARCH frameworks for conditional volatility effects.
Back to Projects