HAR-CAESar Documentation
HAR-CAESar (Heterogeneous Autoregressive Conditional Autoregressive Expected Shortfall) is a Python framework for forecasting joint Value at Risk (VaR) and Expected Shortfall (ES).
This library implements the models developed for the MSc Thesis: “Forecasting Tail Risk with Long-Memory: A Heterogeneous Extension of the CAESar Model”. _.
Key Features
Long-Memory Modeling: Captures volatility cascades using HAR dynamics at daily, weekly, and monthly horizons
Asymmetric Slope Effects: Separates positive and negative return impacts to capture leverage effects
Joint Estimation: Simultaneous VaR and ES forecasting using the Fissler-Ziegel loss function
Comprehensive Backtesting: Implements Kupiec, Christoffersen, McNeil-Frey, and Acerbi-Szekely tests
Benchmarks: Includes CAESar, CAViaR, GAS1, and GAS2 implementations for comparison
Quick Start
Install the package:
git clone https://github.com/shawcharles/HAR-CAESar.git
cd HAR-CAESar
pip install -e .
Basic usage example:
import numpy as np
from har_caesar import HAR_CAESar
# Generate data
y = np.random.normal(0, 1, 2000)
# Fit model and predict
model = HAR_CAESar(theta=0.025)
results = model.fit_predict(y, ti=1500, seed=42)
# Access forecasts
var_forecasts = results['qf']
es_forecasts = results['ef']
Documentation Contents
User Guide:
API Reference: