How aditya6969/4fpdv-calibration works
This repository implements a 4-factor path-dependent volatility model calibration pipeline for SPX/VIX options, VIX futures, and historical market series. The principal workflow loads and filters market data, optimizes parameters against a selected calibration objective, then reprices and saves diagnostics; a historical harness repeats that workflow by date. Separate objectives support time-series, likelihood, mixed, and futures fits, while pricing can use Monte Carlo or a VIX neural network. Edges reflect README statements and excerpted/indexed references; unsampled implementation details are not assumed.
Main entry points: calibrate.py Single-date engine · pricing_vix_neural_bnp.py VIX pricing · sampler_PDV_bnp.py Path simulator · mixed_loss.py Mixed objective
Components 16
Calibration runs
- Single-date engine
calibrate.py - Historical backtest
calibrate_historical.py
Calibration objectives
- Smile objectives
calibrate.py - VIX time-series fit
vix_ts_loss.py - Historical likelihood
mle_loss.py - Mixed objective
mixed_loss.py - Futures objectives
vix_fut_ts_mle_loss.py - Expiry extraction
times_to_maturity.py
Simulation and pricing
- Path simulator
sampler_PDV_bnp.py - VIX pricing
pricing_vix_neural_bnp.py
Neural model
- Neural model assets
neural_network/model_NN_batch64-7Bayesian - Scaling parameters
neural_network/model_NN_batch64-7Bayesian_scaling - Training data generation
train_vix_nn.py
Other components
- Market data
data - Operator
- Calibration results
Connections 27
- Operator → Single-date engine: runs
README.md - Operator → Historical backtest: starts backtest
README.md - Historical backtest → Single-date engine: runs per date
calibrate_historical.py - Market data → Single-date engine: supplies inputs
README.md - Single-date engine → Smile objectives: optimizes against
README.md - Single-date engine → VIX time-series fit: selects objective
calibrate.py - Single-date engine → Historical likelihood: selects objective
calibrate.py - Single-date engine → Mixed objective: selects objective
calibrate.py - Single-date engine → Futures objectives: selects objective
calibrate.py - Single-date engine → Expiry extraction: uses expiries
calibrate.py - Single-date engine → Path simulator: simulates paths
calibrate.py - Mixed objective → Historical likelihood: evaluates likelihood
mixed_loss.py - Mixed objective → VIX time-series fit: adds time-series loss
mixed_loss.py - Mixed objective → Path simulator: simulates paths
mixed_loss.py - Mixed objective → VIX pricing: prices VIX
mixed_loss.py - VIX time-series fit → Path simulator: computes VIX
vix_ts_loss.py - VIX time-series fit → VIX pricing: prices VIX
vix_ts_loss.py - Futures objectives → VIX time-series fit: uses time-series loss
vix_fut_ts_mle_loss.py - Futures objectives → Path simulator: simulates paths
vix_fut_ts_mle_loss.py - Futures objectives → VIX pricing: prices VIX
vix_fut_ts_mle_loss.py - VIX pricing → Path simulator: uses simulated factors
pricing_vix_neural_bnp.py - Single-date engine → Calibration results: saves diagnostics
README.md - Historical backtest → Market data: reads history
README.md - Training data generation → Path simulator: uses simulator
train_vix_nn.py - Training data generation → Neural model assets: generates training data
- VIX pricing → Neural model assets: loads model
- VIX pricing → Scaling parameters: uses scaling