4.4.5 Failure Modes & Real-World Practice

Basic Prerequisites: 04 · Reduced-Form & CDS and 03 · Distance-to-Default & PD.

---

1. Intuition & Practical Objective

The Merton model is mathematically exact and empirically wrong in four specific ways. This page names them precisely, so a practitioner knows which assumption to distrust and how the failure shows up in money terms. The objective is not cynicism - it is knowing exactly where credit risk modelling becomes an approximation, so the residual can be measured and managed.

The four failures, in one line each:

1. Firm value is unobservable - the model inverts equity into $(V,{\sigma }_V)$; that inversion is ill-conditioned and amplifies input error into PD.
2. No jumps → no jump-to-default - a pure diffusion cannot produce a sudden collapse, so it badly understates short-horizon default.
3. The credit-spread puzzle - the model’s own spreads are orders of magnitude too small at short maturities, so it cannot be trusted for spread quoting.
4. Defaults are correlated - every single-name model is silent about the joint event that actually kills a portfolio.

---

2. Mathematical Ground Truth & Derivations

2.1 Where the assumptions live

Every Merton result rests on four interlocking assumptions:

• (A1) Observable, lognormal firm value - one diffusion with constant ${\sigma }_V$; the model’s inputs $(V_0,{\sigma }_V)$ must be inferred, not observed.
• (A2) Debt is a single zero-coupon bond; default only at $T$. No interim defaults, coupons, covenants, or rollover.
• (A3) Continuous, frictionless markets; no jump-to-default, no illiquidity. Else the no-arbitrage replication underlying the call price fails.
• (A4) Constant $r$, known and constant recovery. Recovery is a fraction of firm value in the model ($R_V=V/D$ relationship), not of face.

2.2 Failure 1 - the ill-conditioned inversion

DD and PD are defined through $(\ln (V/D),{\sigma }_V)$, but we solve for $V,{\sigma }_V$ from equity observables whose levels are noisy. Differentiating the vol identity ${\sigma }_{E}E=N(d_1){\sigma }_{V}V$ shows $\partial {\sigma }_V/\partial {\sigma }_E>0$ with an elasticity greater than one near the money - equity-vol error is amplified into asset-vol error, and PD is convex in ${\sigma }_V$. Page 03’s firm shows the effect: a $+1\%$ relative error in ${\sigma }_E$ moves PD by $\approx 9\%$.

2.3 Failure 4 - correlation is the whole portfolio story

For a portfolio of $n$ loans with PDs $p_i$ and pairwise default correlation $\rho$, the unexpected loss satisfies (Bluhm 2010, eq. 1.13)

$${\mathrm{UL}}^2=\sum _i{p}_i(1-p_i)+2\rho \sqrt{p_1(1-p_1)p_2(1-p_2)}(\text{2 loans}),$$

and in general ${\mathrm{UL}}^2={\sum }_{i,j}{\rho }_{ij}\sqrt{p_i(1-p_i)p_j(1-p_j)}$. When $\rho =0$ risk diversifies to $\sum p_i(1-p_i)$; when $\rho =1$ the portfolio behaves as one obligor with $n$-fold intensity - pure concentration risk. A single-name model has no $\rho$ in it at all; that is why portfolio credit risk needs the factor model of page 06.

---

3. Computational Implementation - the failures, measured

Experiment 1 - PD sensitivity and the credit-spread puzzle. Stdlib only.

Two first-principles failures in numbers: PD rises super-linearly with equity vol (a $1\%$ input error becomes a $\sim 9\%$ PD error), and the model’s credit spread is under $1$ bp for a firm the market would charge $100+$ bp - Merton (1974) famously under-predicts short-maturity spreads because it forces default to the maturity date.

Experiment 2 - jump-to-default. Add an independent Poisson(λ) collapse on top of the Merton diffusion. Default now arrives either as a jump or as a diffusion breach at $T$, giving the closed form

$${\mathrm{PD}}_{\text{jump}}=(1-e^{-\lambda T})+e^{-\lambda T}N(-d_2).$$

A mere $1\%$ annual jump intensity nearly quintuples the default probability, because the jump term dominates at short horizons where the diffusion is nearly default-free. This is the structural model’s missing catastrophe, and precisely why reduced-form/intensity models dominate practice for short-dated credit (page 04).

---

4. Failure Modes & First-Principles Breakdowns (numbered)

1. Unobservable asset value (A1). The model’s inputs are inferred, not observed; the inversion is ill-conditioned and PD is convex in the inferred ${\sigma }_V$ - $1\%$ in, $9\%$ out (Experiment 1).
2. No jump-to-default (A3). A diffusion cannot collapse; adding the smallest realistic jump channel multiplies PD several-fold (Experiment 2). This is the single most important structural defect and the reason markets are priced with intensities.
3. The credit-spread puzzle (A2). Merton spreads are far below market at short maturities (Experiment 1), because default is forced to $T$ and the firm cannot default between coupons. Black–Cox first-passage barriers and jump-diffusion partially repair this.
4. Correlation (A4). Single-name models contain no $\rho$; portfolio risk is entirely a correlation story (Bluhm eq. 1.13), and the Gaussian-copula correlation that credit markets used was not a physical quantity - its mis-estimation underpriced mezzanine tranches in 2008.
5. Recovery and LGD are random. Recovery is correlated with default rates and varies with seniority; treating it as a constant biases every derived intensity and CVA.
6. Model risk per se. Swapping $D$ for the KMV default point $D^{\ast }=\mathrm{ST}+\frac{1}{2}\mathrm{LT}$, or changing the horizon, moves the answer without changing the data - exactly the “right model, wrong inputs” failure Derman (1996) warned about, and a required item under SR 11-7.

---

5. References

• Hull, Options, Futures, and Other Derivatives
• Merton (1974)
• Gregory, Jon - The xVA Challenge (5th ed., 2025)
• Bluhm, Overbeck & Wagner - Introduction to Credit Risk Modeling (2010)
• Derman, Emanuel - Model Risk (1996)

---

6. Connected Graph Bridges

• Back: 04 · Reduced-Form & CDS · 03 · DD & PD · Index Hub
• Forward: 06 · Portfolio Credit & Vasicek
• Siblings: EVT & Fat Tails (the jump/thin-tail problem) · VaR & ES · Liquidity Risk