Risk Aggregation Across Asset Classes

Summing up risk aggregation across asset classes—adding VaR from equities to VaR from bonds to VaR from derivatives—looks straightforward on paper but stumbles on the real-world complications: different risk models, hidden correlations, liquidity mismatches, and the fact that diversification breakdowns exactly when you need it most.

Why Firms Need Aggregation

A mid-sized financial institution might run ten trading desks: equities, fixed income, foreign exchange, commodities, derivatives, structured products, and more. Each desk has its own risk models and daily VaR numbers. The equity desk reports a $2M VaR; the bond desk reports $1.8M; the derivatives desk reports $0.9M. The chief risk officer needs to know: What is the firm’s total risk exposure?

The simplest answer—$2M + $1.8M + $0.9M = $4.7M—is wrong. It assumes every desk’s loss happens simultaneously. In reality, some assets move together; others move in opposition. Diversification means the firm’s VaR is lower than the sum of the parts. The real answer might be $3.2M, reflecting that equity, bond, and commodity moves are not perfectly correlated.

But here’s the catch: that $3.2M number is only as good as the correlation assumptions, and those assumptions explode in a crisis.

The Aggregation Methods

Simple summation (worst-case stacking). Add all desk VaRs without any correlation adjustments. Used only for gross exposure limits and as a conservative sanity check. Overstates true risk but ensures no hidden leverage.

Correlation-based VaR aggregation. The standard method: treat each desk’s P&L as a variable, estimate the correlation matrix between desks (or between asset classes), then compute the firm-wide VaR using portfolio theory.

Formula (simplified): Firm VaR² = Σ(VaR_i²) + 2 × Σ(ρ_ij × VaR_i × VaR_j)

The first term is the sum of squared individual VaRs. The second term captures correlations: positive correlation increases firm VaR; negative correlation reduces it (hedging effect). Estimate ρ_ij from daily returns over the past year or two, and you get an aggregate number.

Factor model aggregation. Instead of correlating desks, decompose each desk’s P&L into common factors (equity beta, interest rate duration, credit spread, volatility). Then aggregate the factor exposures.

Example:

• Equity desk: $5M exposure to market beta, $0.5M to small-cap factor
• Bond desk: $(3)M (short) to interest rate factor, $1.2M to credit factor
• Commodity desk: $2M to oil price, $(0.3)M to agricultural prices

A central risk function can then compute the firm’s total exposure to each factor, and apply a single sensitivity model across the firm. This is more transparent but requires each desk to break down its risk in a consistent taxonomy.

Stress-based aggregation. Instead of relying on VaR or correlations, apply a scenario (e.g., “equities fall 20%, credit spreads widen 200bp, implied volatility spikes to 50%”) and compute the firm’s loss. Aggregate across multiple scenarios. Stress testing avoids correlation assumptions but is labor-intensive.

Core Challenges

Different asset class models. Equity models use beta and volatility; bond models use duration and convexity; derivatives models use Greeks and volatility surface assumptions. There is no single “VaR” methodology that applies cleanly to all. An equity desk might use a historical VaR model; a bond desk uses a duration/convexity model; a derivatives desk uses a full revaluation (Monte Carlo). Aggregating these numbers is mixing apples and oranges.

Correlation instability. Historic correlations are reliable in calm markets. Equities and bonds are often negatively correlated over a year—a classic diversifier. Then a crisis hits, yields spike, and both equities and bonds fall together. Correlations spike to 0.7 or 0.8. The firm thought it was diversified; suddenly it is not. Regulatory stress tests try to account for this (assuming spike scenarios), but static correlation matrices do not capture the regime shift in real time.

Hidden leverage. Desk-level VaR can hide leverage. A derivatives desk might hold a large short volatility position that carries low delta and duration exposure (thus low desk-level VaR) but enormous vega and tail risk. When implied volatility explodes, the desk loses a fortune. If no one is aggregating vega across desks (checking that the firm’s overall short volatility is not too large), the risk goes undetected.

Cross-desk correlations and netting. A major challenge is determining what correlations matter between desks. If the equity desk is long cyclicals and the commodities desk is short oil, there is a natural hedge (recessions hit both). But this hedge is hidden if you are aggregating desk-level VaR numbers. Only detailed factor models surface these cross-desk offsets. Many firms discover (too late) that they have large correlated exposures in seemingly unrelated desks.

Liquidity mismatches. VaR assumes you can exit a position at yesterday’s prices. But in a stress, bid-ask spreads widen, and some assets become illiquid. A bond desk’s $1.8M VaR is worthless if the bonds cannot be sold quickly. Equity desks can often exit quickly; some structured product desks cannot. A firm-wide VaR that ignores liquidity differences is misleading.

Non-linear exposures. VaR is linear: it assumes P&L moves proportionally with market moves. But options, credit default swaps, and mortgages are non-linear. A large market move can push a position from low-risk to high-risk (or vice versa). A linear aggregation model misses these threshold effects.

Practical Governance

Most financial institutions run a three-layer risk review:

Daily desk limits. Each desk has a VaR limit (e.g., $2M equity desk, $1.8M bond desk). Desk-level reports are automated and checked continuously.

Weekly enterprise risk review. Risk aggregation is computed by correlating desks or using a factor model. The enterprise risk committee reviews the firm-wide VaR, stress scenarios, and exposure to key factors. If firm-wide VaR breaches a threshold or a single factor exposure exceeds a limit, the committee escalates.

Quarterly stress testing. The firm runs scenarios: equity crash, credit spread widening, yield curve inversion, volatility spike, etc. Each desk’s P&L is modeled under each scenario, then summed. The goal is to identify concentrated bets and correlation fragilities that daily VaR might miss.

Regulatory Requirements

Post-Dodd-Frank, Basel III, and Volcker Rule, financial institutions must calculate and report firm-wide VaR and stress losses to regulators (the Federal Reserve, CFTC, etc.). The regulations require:

• Daily VaR at 99% confidence level (1% tail loss)
• Stress testing under historical scenarios (2008, 1987, etc.)
• Incremental risk capital (IRC) for credit-sensitive instruments
• Comprehensive risk measure (CRM) for correlation and basis risk

These mandates have driven firms to invest in centralized risk infrastructure, though the systems remain imperfect.

The Enduring Paradox

A well-aggregated VaR model says “the firm can lose up to $3M on a bad day.” That is useful until the bad day comes and the firm loses $8M. The gap usually reflects either:

• A correlation or exposure that was omitted (a desk held an unmodeled position)
• A tail event outside the historical distribution (a new correlation regime emerged)
• A liquidity crisis (the model assumed exit prices; markets froze)

The best firms handle this by using VaR as a baseline and stress testing as a check. They also cultivate a culture where desk heads know that “the VaR model said it was okay” is not a defense after a blowup.

See also