Paper: arXiv 2506.17549

Authors: Sourish Das

Abstract

This paper develops a Bayesian Generalised Pareto Regression (GPR) model to forecast extreme losses in Indian equity markets, with a focus on the Nifty 50 index. Extreme negative returns, though rare, can cause significant financial disruption, and accurate modelling of such events is essential for effective risk management. Traditional Generalised Pareto Distribution (GPD) models often ignore market conditions; in contrast, our framework links the scale parameter to covariates using a log-linear function, allowing tail risk to respond dynamically to market volatility. We examine four prior choices for Bayesian regularisation of regression coefficients: Cauchy, Lasso (Laplace), Ridge (Gaussian), and Zellner’s g-prior. Simulation results suggest that the Cauchy prior delivers the best trade-off between predictive accuracy and model simplicity, achieving the lowest RMSE, AIC, and BIC values. Empirically, we apply the model to large negative returns (exceeding 5%) in the Nifty 50 index. Volatility measures from the Nifty 50, S&P 500, and gold are used as covariates to capture both domestic and global risk drivers. Our findings show that tail risk increases significantly with higher market volatility. In particular, both S&P 500 and gold volatilities contribute meaningfully to crash prediction, highlighting global spillover and flight-to-safety effects. The proposed GPR model offers a robust and interpretable approach for tail risk forecasting in emerging markets. It improves upon traditional EVT-based models by incorporating real-time financial indicators, making it useful for practitioners, policymakers, and financial regulators concerned with systemic risk and stress testing.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced Bayesian statistical methods and specialized survival function derivations (high math complexity), and includes a detailed simulation study, multiple volatility estimators, and backtest-ready metrics like RMSE, AIC, and BIC (high empirical rigor).

Research Flowchart

  flowchart TD
  A["Research Goal<br>Predict Stock Market Crashes<br>using Extreme Value Theory"] --> B{"Data & Inputs"};
  
  B --> B1["Nifty 50 Index<br>(Daily Returns)"];
  B --> B2["Covariates<br>VIX India, S&P 500 Vol, Gold Vol"];
  B --> B3["Threshold Selection<br>Exceedances > 5% Loss"];
  
  A --> C["Methodology<br>Bayesian Generalised Pareto Regression GPR"];
  
  C --> D["Computational Process<br>Prior Comparison & Fitting"];
  
  D --> D1["Priors: Cauchy, Lasso, Ridge, g-Prior"];
  D --> D2["Simulation for Model Selection"];
  
  D1 --> D3["Model Metrics:<br>RMSE, AIC, BIC"];
  D2 --> D3;
  
  D3 --> E{"Key Findings & Outcomes"};
  
  E --> E1["Optimal Model:<br>Cauchy Prior"];
  E --> E2["Risk Drivers:<br>High Volatility increases Crash Risk"];
  E --> E3["Global Spillover:<br>S&P 500 & Gold Volatility significant"];
  E --> E4["Outcome:<br>Robust Tail Risk Forecasting Framework"];

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes a Bayesian Generalised Pareto Regression (GPR) model that links the GPD scale parameter to market volatility covariates via a log‑linear function to forecast extreme losses in the Indian equity market.

Source sentence
This paper develops a Bayesian Generalised Pareto Regression (GPR) model to forecast extreme losses in Indian equity markets, with a focus on the Nifty 50 index.

Assumes 4

  • A common shape parameter is assumed across all observations; only the scale parameter varies with predictors. model
    Source sentence
    we assume a common shape parameter across observations and model only the scale parameter as a function of predictors.
  • The scale parameter follows a log‑linear relationship with covariates. model
    Source sentence
    our framework links the scale parameter to covariates using a log-linear function, allowing tail risk to respond dynamically to market volatility.
  • Analysis is restricted to daily log‑returns where the return is less than –2%, representing the left tail. data
    Source sentence
    We work with a cleaned dataset comprising daily log-returns of the NSE Nifty 50 index, with all missing entries removed. Denoting the daily return at time t as rt, we extract a subset of observations where rt < −2%.
  • Volatility covariates (empirical and Garman–Klass) for Nifty 50, S&P 500 and gold are standardized to mean zero and unit variance. data
    Source sentence
    We construct covariates based on (i) empirical volatility and (ii) Garman–Klass (GK) volatility for the Nifty 50, S&P 500, and gold. These volatility variables are standardised to have mean zero and unit variance.

Data

  • Nifty 50 daily closing prices and log‑returns · National Stock Exchange of India (NSE) market-prices public
    17 September 2007 to 13 June 2025 · Nifty 50 index constituents
    Source sentence
    We analyse the Nifty 50 index over the period from 17 September 2007 to 13 June 2025.
  • Realised volatility measures for Nifty 50, S&P 500 and gold alternative unknown
    Source sentence
    Volatility measures from the Nifty 50, S&P 500, and gold are used as covariates to capture both domestic and global risk drivers.

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • The Cauchy prior yields the best predictive performance, achieving the lowest RMSE (1.58), AIC (298.06) and BIC (322.20) among the four priors considered. predictive-accuracy
    Source sentence
    Among all models, the Cauchy prior delivers the best fit, achieving the lowest AIC and BIC values, along with the smallest RMSE, suggesting strong predictive accuracy and robust in-sample performance.
  • In simulation, the Cauchy prior also attains the lowest RMSE for y (11.36) and the lowest AIC (162.67) and BIC (176.97) compared with Lasso, Ridge and g‑prior. statistical
    Source sentence
    Table 1: Simulation comparison of GPD regression models with different priors
  • Tail risk increases significantly with higher market volatility, with expected losses rising from about 2.4 at low volatility to over 7 at high volatility and crash probabilities rising from below 5% to above 65%. statistical
    Source sentence
    Results show a nonlinear increase in both quantities with rising volatility. At the lower end

Checked by others

Supported by

  • NFT Bubbles (2023)
    On the question: Does higher market volatility increase the likelihood of a crash?
    Compare the findings

    This paper: Tail risk increases significantly with higher market volatility, with expected losses rising from about 2.4 at low volatility to over 7 at high volatility and crash probabilities rising from below 5% to above 65%.

    That paper: Crash likelihood is higher when ex‑ante volatility and price acceleration are high, and lower when turnover is high.

    Both find crash risk rises with volatility: A links high ex‑ante volatility/price acceleration to higher crash likelihood; B shows tail risk up from 2.4 to >7 and crash prob from <5% to >65% as volatility rises. A uses N

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (21)

  1. The 2020 global stock market crash: Endogenous or exogenous? (2021) — link
  2. Statistical properties of the aftershocks of stock market crashes revisited: Analysis based on the 1987 crash, financial-crisis-2008 and COVID-19 pandemic (2020) — link
  3. Preventing crash in stock market: The role of economic policy uncertainty during COVID-19 (2020) — link
  4. The impact of COVID-19 on the stock market crash risk in China (2020) — link
  5. Inside the Mind of a Stock Market Crash (2020) — link
  6. Understanding the economic determinants of the severity of operational losses: A regularized generalized Pareto regression approach (2018) — link
  7. Understanding extreme stock trading volume by generalized Pareto distribution (2016)
  8. Detecting structural breaks in tail behaviour – from the perspective of fitting the generalized Pareto distribution (2013) — link
and 13 more
  1. Hyper-$g$ priors for generalized linear models (2010) — link
  2. Analysis of Extreme Drinking in Patients with Alcohol Dependence Using Pareto Regression (2010) — link
  3. A weakly informative default prior distribution for logistic and other regression models (2008) — link
  4. The Bayesian Lasso (2008) — link
  5. On the power of generalized extreme value (GEV) and generalized Pareto distribution (GPD) estimators for empirical distributions of stock returns (2006) — link
  6. Statistical Decision Theory and Bayesian Analysis. Second Edition (James O. Berger) (1987) — link
  7. Maximum likelihood estimation in a class of nonregular cases (1985) — link
  8. Computational Finance with R (2023) — link
  9. COVID-19 and March 2020 Stock Market Crash. Evidence from S&P1500 (2020) — link
  10. Statistical modelling of financial crashes. (2008)
  11. The Elements of Statistical Learning: Data Mining, Inference, and Prediction (2001) — link
  12. On assessing prior distributions and bayesian regression analysis with g-prior distributions (1986)
  13. On the Estimation of Security Price Volatilities from Historical Data (1980) — link

0 of 21 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.