Stochastic Calibration of JPY-Denominated Assets

Overview

Japanese Yen-denominated assets have exhibited anomalous volatility clustering since Q3 2025, driven by Bank of Japan policy divergence and shifting carry trade dynamics. This dispatch details our Bayesian approach to calibrating JPY exposure within the refined 18-position portfolio.

The Problem

Standard Black-Scholes implied volatility models were producing systematically biased hedging ratios for our JPY positions. Realized volatility consistently exceeded model predictions by 15-22%, resulting in under-hedged exposure during key risk events.

Bayesian Calibration Framework

We implemented a hierarchical Bayesian model that incorporates:

1. Prior Distribution: Informed by 10 years of JPY/USD realized volatility data
2. Likelihood Function: Real-time options market data (1-minute granularity)
3. Posterior Update: Sequential Monte Carlo with 10,000 particle filter

# Bayesian Vol Calibration — JPY Module
class JPYVolCalibrator:
    def __init__(self, prior_params: dict):
        self.prior = GammaDistribution(**prior_params)
        self.particle_count = 10_000
        
    def update(self, market_data: MarketSnapshot) -> Posterior:
        particles = self.prior.sample(self.particle_count)
        weights = self._compute_likelihood(particles, market_data)
        return self._resample(particles, weights)
    
    def hedge_ratio(self, posterior: Posterior) -> float:
        expected_vol = posterior.mean()
        return black_scholes_delta(vol=expected_vol)

Calibration Results

After 30 trading days of live calibration:

Parameter	Standard Model	Bayesian Model	Improvement
Vol Prediction Error	18.4%	4.2%	-77.2%
Hedge Ratio Bias	-0.08	-0.01	-87.5%
P&L Volatility	¥2.4M daily	¥0.9M daily	-62.5%
Max Drawdown	-¥18.2M	-¥6.1M	-66.5%

Risk Considerations

The Bayesian model introduces model risk through prior specification. We mitigate this through:

• Weekly prior sensitivity analysis
• Parallel shadow-running of 3 alternative prior specifications
• Automatic fallback to standard model if posterior divergence exceeds 2σ

“In the presence of regime uncertainty, the honest model is the one that admits what it doesn’t know.”

Current Status

The system is LIVE on Node_TKY_01 and processing real-time JPY options data. Alpha signal is currently NEUTRAL pending full calibration convergence (estimated: 15 additional trading sessions).

---

Dispatch Classification: LIVE — Active monitoring on Tokyo node. Calibration ongoing.