Paper: arXiv 2505.05646
Authors: Xin Tian
Abstract
This report presents a comprehensive evaluation of three Value-at-Risk (VaR) modeling approaches: Historical Simulation (HS), GARCH with Normal approximation (GARCH-N), and GARCH with Filtered Historical Simulation (FHS), using both in-sample and multi-day forecasting frameworks. We compute daily 5 percent VaR estimates using each method and assess their accuracy via empirical breach frequencies and visual breach indicators. Our findings reveal severe miscalibration in the HS and GARCH-N models, with empirical breach rates far exceeding theoretical levels. In contrast, the FHS method consistently aligns with theoretical expectations and exhibits desirable statistical and visual behavior. We further simulate 5-day cumulative returns under both GARCH-N and GARCH-FHS frameworks to compute multi-period VaR and Expected Shortfall. Results show that GARCH-N underestimates tail risk due to its reliance on the Gaussian assumption, whereas GARCH-FHS provides more robust and conservative tail estimates. Overall, the study demonstrates that the GARCH-FHS model offers superior performance in capturing fat-tailed risks and provides more reliable short-term risk forecasts.
Complexity vs Empirical Score
- Math Complexity: 7.0/10
- Empirical Rigor: 6.5/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced statistical models like GARCH and filters like FHS, which involve complex time-series mathematics, but also implements empirical backtesting with breach frequency analysis and visual diagnostics, making it both mathematically dense and data/implementation-heavy.
Research Flowchart
flowchart TD A["Research Goal: Comparative Evaluation of VaR Models<br>HS, GARCH-N, GARCH-FHS"] --> B["Data Input<br>Daily Returns for In-sample & Forecasting"] B --> C["Methodology: VaR Computation (5%)<br>1. Historical Simulation (HS)<br>2. GARCH-Normal (GARCH-N)<br>3. Filtered Hist. Sim. (GARCH-FHS)"] C --> D["Assessment: Accuracy & Breach Frequency<br>Compare Empirical Breaches vs 5%"] D --> E["Multi-Period Analysis (5-day)<br>Simulate Cumulative Returns<br>Compute VaR & Expected Shortfall"] E --> F["Key Findings & Outcomes"] F --> F1["HS & GARCH-N: Severe Miscalibration<br>High Breach Rates"] F --> F2["GARCH-FHS: Superior Performance<br>Aligns with Theoretical Levels"] F --> F3["GARCH-N: Underestimates Tail Risk<br>(Gaussian Assumption)"] F --> F4["GARCH-FHS: Robust Tail Estimates<br>Recommended for Risk Management"]
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 using the GARCH‑FHS method for risk management and 1‑week ahead forecasting.
Source sentence
We recommend the FHS method for practical applications in risk management, particularly in volatile market environments and high‑confidence settings.
Assumes 3
- GARCH‑N relies on a Gaussian (normal) distribution for standardized residuals. model
Source sentence
GARCH‑N underestimates tail risk due to its reliance on the Gaussian assumption
- Historical Simulation assumes a static historical return distribution. model
Source sentence
the assumption of a static historical distribution becomes invalid
- Return volatility is heteroskedastic, making identically distributed returns within the estimation window problematic. statistical
Source sentence
if return volatility is heteroskedastic—as is often the case in financial markets—then assuming identically distributed returns within the estimation window is itself problematic
Data
No dataset named in the text.
Costs
Not applicable
Finds 5
- Historical Simulation and GARCH‑N show severe miscalibration with empirical breach rates far exceeding theoretical 5% and 1% levels. performance
Source sentence
Our findings reveal severe miscalibration in the HS and GARCH‑N models, with empirical breach rates far exceeding theoretical levels at both 5% and 1% confidence thresholds.
- Filtered Historical Simulation consistently aligns with theoretical expectations and shows desirable statistical and visual behavior. performance
Source sentence
In contrast, the FHS method consistently aligns with theoretical expectations and exhibits desirable statistical and visual behavior.
- GARCH‑N underestimates tail risk because of its Gaussian assumption. negative
Source sentence
Results show that GARCH‑N underestimates tail risk due to its reliance on the Gaussian assumption
- GARCH‑FHS provides more robust and conservative tail estimates than GARCH‑N. performance
Source sentence
whereas GARCH‑FHS provides more robust and conservative tail estimates
- GARCH‑FHS outperforms GARCH‑N for multi‑day VaR and Expected Shortfall estimation at extreme quantiles. performance
Source sentence
Overall, the GARCH + FHS method outperforms GARCH + Normal in terms of accuracy and robustness for estimating multi‑day VaR and ES at extreme quantiles.
Checked by others
Supported by
- Portfolio Stress Testing and Value at Risk (VaR) Incorporating Current Market Conditions
Compare the findings
This paper: GARCH‑N underestimates tail risk because of its Gaussian assumption.
That paper: Gaussian‑parametrized VaR is lower than Historical Simulation VaR for the 97.5% confidence level because the Gaussian distribution does not capture heavy tails.
Qualified by
- Reliability-Aware ETF Tail-Risk Monitoring
Compare the findings
This paper: Historical Simulation and GARCH‑N show severe miscalibration with empirical breach rates far exceeding theoretical 5% and 1% levels.
That paper: The 252‑day historical VaR benchmark is conservative (breach rate 4.42%), whereas the EWMA‑normal benchmark remains under‑covered (5.75%).
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 (1)
- A LaTeX package to create plots. — link
0 of 1 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.