How does the Basel backtesting traffic light system work for validating VaR models?

Question

AcadiFi · Accepted Answer

The Basel backtesting framework is the primary method regulators use to validate banks' internal VaR models. It compares the model's predicted losses (VaR) against actual realized losses over a 250-day window. The traffic light system classifies model performance and determines the regulatory capital multiplier.

**What is a VaR Exception?**
An exception occurs when the actual daily loss exceeds the predicted VaR. For a 99% confidence VaR model, you expect approximately 2.5 exceptions in 250 trading days (250 x 1% = 2.5). Significantly more exceptions suggest the model underestimates risk.

```mermaid
graph TD
  A[Count VaR Exceptions in 250 Trading Days] --> B{Number of Exceptions?}
  B -->|0 to 4| C[GREEN ZONE]
  B -->|5 to 9| D[YELLOW / AMBER ZONE]
  B -->|10 or more| E[RED ZONE]
  C --> C1[Capital Multiplier: 3.0x]
  C --> C2[Model Accepted - No Action Required]
  D --> D1[Capital Multiplier: 3.4x to 3.85x]
  D --> D2[Regulatory Inquiry - Bank Must Explain]
  D --> D3[Possible causes reviewed]
  E --> E1[Capital Multiplier: 4.0x]
  E --> E2[Model Rejected or Major Remediation]
  D3 --> D3a[Bad luck / market regime shift?]
  D3 --> D3b[Model deficiency?]
  D3 --> D3c[Intraday trading not captured?]
  D3 --> D3d[Inaccurate data or IT error?]
  style C fill:#1a3c2a,color:#fff
  style D fill:#c9a84c,color:#0a0a0f
  style E fill:#8b2020,color:#fff
```

**Traffic Light Zones Detailed:**

**Green Zone (0-4 exceptions)**
The model is performing as expected. Under a correctly specified 99% VaR model, the probability of 4 or fewer exceptions in 250 days is approximately 89%. The regulatory capital multiplier stays at the minimum **3.0x**.

Exceptions in this zone are consistent with normal statistical variation — even a perfect model will occasionally produce exceptions.

**Yellow/Amber Zone (5-9 exceptions)**
This is the ambiguous zone. The model may have problems, or the bank may have experienced unusual market conditions. Each additional exception increases the multiplier:

| Exceptions | Multiplier | Cumulative Probability |
|---|---|---|
| 5 | 3.40 | 95.88% |
| 6 | 3.50 | 98.63% |
| 7 | 3.65 | 99.60% |
| 8 | 3.75 | 99.89% |
| 9 | 3.85 | 99.97% |

The bank must provide an explanation to its supervisor. Acceptable explanations include:
- Market regime shift (e.g., sudden volatility spike during a geopolitical crisis)
- Specific event not in historical data window
- Known model limitation being remediated

**Red Zone (10+ exceptions)**
The model is almost certainly misspecified. The probability of 10+ exceptions from a correct 99% VaR model is approximately 0.01% — essentially impossible by chance alone. The multiplier increases to **4.0x**, and the regulator may require the bank to revert to the Standardized Approach or undertake a complete model rebuild.

**Worked Example:**

Crestfield Securities runs a 99% 1-day VaR model. Over the past 250 trading days:
- VaR model predicted daily loss threshold: average $14.2 million
- Number of days actual loss exceeded VaR: **7 exceptions**

Classification: **Yellow zone** with a capital multiplier of **3.65x**.

If Crestfield's base VaR is $14.2 million:
- Green zone capital charge: $14.2M x 3.00 x sqrt(10) = **$134.7M**
- Yellow zone capital charge: $14.2M x 3.65 x sqrt(10) = **$163.9M**

The seven exceptions cost Crestfield an additional **$29.2 million** in required market risk capital.

**Statistical Foundation:**
The number of exceptions follows a binomial distribution B(n=250, p=0.01). The expected number is 2.5 with standard deviation sqrt(250 x 0.01 x 0.99) = 1.57. The zone boundaries are set to balance Type I error (penalizing a correct model) against Type II error (accepting a bad model).

**Exam Tip:** Know the exact zone boundaries (0-4, 5-9, 10+), the minimum and maximum multipliers (3.0x and 4.0x), and the fact that the sqrt(10) scaling converts 1-day VaR to 10-day VaR for capital purposes. Also understand that backtesting alone cannot identify whether the model systematically over- or under-estimates risk in specific tails — it only counts exceptions.

Practice backtesting calculations in our FRM Part II Market Risk question bank on AcadiFi.

How does the Basel backtesting traffic light system work for validating VaR models?

Master Part II with our FRM Course

Related Questions

Practice Questions