Efficient Monte Carlo Valuation of Corporate Bonds in Financial Networks

Valuing corporate bonds in systemic economies is challenging due to intricate webs of inter-institutional exposures. When a bank defaults, cascading losses propagate through the network, with payments determined by a system of fixed-point equations lacking closed-form solutions. Standard Monte Carlo methods cannot capture rare yet critical default events, while existing rare-event simulation techniques fail to account for higher-order network effects and scale poorly with network size. To overcome these challenges, we propose a novel approach – Bi-Level Importance Sampling with Splitting – and characterize individual bank defaults by decoupling them from the network’s complex fixed-point dynamics. This separation enables a two-stage estimation process that directly generates samples from the banks’ default events. We demonstrate theoretically that the method is both scalable and asymptotically optimal, and validate its effectiveness through numerical studies on empirically observed networks.

💡 Research Summary

The paper tackles the problem of pricing corporate bonds issued by firms that are embedded in a systemic financial network with inter‑bank liabilities, common asset holdings, and fire‑sale externalities. In such a setting, the value of a bond depends on the equilibrium clearing payments that result from a cascade of defaults, which is defined implicitly by a fixed‑point system and cannot be expressed in closed form. Standard Monte‑Carlo simulation is inefficient because the events that drive credit risk—rare, system‑wide default cascades—are observed only with vanishing probability. Existing rare‑event techniques either require an explicit description of the rare set or scale exponentially with the number of institutions, making them unsuitable for realistic networks containing hundreds or thousands of banks.

To overcome these obstacles, the authors propose a novel two‑stage estimator called Bi‑Level Importance Sampling with Splitting (BLISS). The first stage (outer level) samples shocks to the external assets of all non‑target institutions using exponential tilting, thereby biasing the simulation toward large systemic shocks that are more likely to trigger contagion. The second stage (inner level) samples the conditional default of a particular target institution given the outer‑level shocks. Crucially, the authors derive a deterministic “critical threshold” for the target’s external assets by temporarily decoupling the target from the network, solving a linear‑time problem that captures the amplified losses fed back from the rest of the system. If the target’s external assets fall below this threshold, default occurs. This threshold enables the inner‑level event to be sampled directly, and the inner‑level importance distribution is chosen as the theoretically optimal change of measure for this conditional problem.

The paper provides rigorous asymptotic analysis in two rare‑event regimes: (i) a large‑asset regime where external asset levels increase, and (ii) a low‑volatility regime where asset return volatility shrinks. In each regime the authors construct a deterministic surrogate for the second moment of the estimator, identify its minimizer, and show that the resulting BLISS estimator achieves logarithmic variance that remains bounded (i.e., asymptotically optimal) as the rare event becomes rarer. The analysis demonstrates that the variance reduction factor does not deteriorate with network size, establishing scalability.

Empirical validation uses a network calibrated to European Banking Authority data, comprising several hundred banks with realistic inter‑bank exposures and common asset holdings. The authors compare BLISS against plain Monte‑Carlo, single‑level importance sampling, and conditional Monte‑Carlo methods. Results show that BLISS attains the same confidence interval width with only 1–2 % of the simulation budget required by the alternatives. Moreover, the estimator remains stable when default probabilities fall below 10⁻⁴, a regime where other methods either fail to converge or demand prohibitive computational effort. Computational time scales roughly linearly with the number of institutions, confirming the method’s practicality for large‑scale systemic risk applications.

Key contributions of the work are: (1) a provably efficient decoupling technique that yields a linear‑time computable default threshold for any target institution; (2) a bi‑level importance sampling framework that simultaneously addresses higher‑order contagion effects and the curse of dimensionality; (3) theoretical guarantees of asymptotic optimality in multiple rare‑event regimes; and (4) extensive numerical evidence of scalability and accuracy on a realistic, high‑dimensional financial network.

In summary, BLISS provides a powerful, theoretically grounded, and computationally tractable tool for pricing corporate bonds and, more broadly, for evaluating any contingent claim that depends on systemic network dynamics. Its ability to handle non‑linear contagion channels, large networks, and extremely rare default events makes it a significant advancement over existing Monte‑Carlo and rare‑event simulation techniques in the field of systemic risk and network‑based financial engineering.