5.4.6 Advanced Extensions

Basic Prerequisites: 05 · Failure Modes and Tangency & CAPM.

---

1. Intuition & Practical Objective

Page 05 showed risk parity works only if you can lever the safe leg. This page asks the two questions everything else has avoided: Why should equal-risk allocation beat the market at all? And when is it honest to claim that? The objective is to (a) give the equilibrium theory - leverage aversion - that explains why safer assets historically earn higher risk-adjusted returns, (b) quantify the levered-parity story on Asness-style numbers, and (c) state the uncomfortable critique: risk parity is a view about expected returns, not a free lunch. It closes by routing to Hierarchical Risk Parity (HRP), the matrix-inversion-free cousin that shares ERC’s love of risk balance but drops the covariance-sensitivity.

The one-sentence essence. “Leverage-averse investors bid up risky assets, so safe assets cheapen and earn more per unit of risk; risk parity harvests exactly that premium by over-weighting the safe leg, then borrowing - and the honest formulation admits this is a view, not a mathematical law.”

---

2. Mathematical Ground Truth & Derivations

Why risk parity can earn a premium (Black 1972; Frazzini & Pedersen 2010). The CAPM says everyone holds the market, levered to taste, so the market is mean-variance efficient. But many investors cannot or will not use leverage (mutual-fund and pension rules, margin constraints, funding costs, outright aversion). Such an investor who wants more risk than the tangency portfolio must move up the hyperbola past the tangency point by buying the riskier asset - concentrating in stocks instead of levering bonds. Enough of these investors bid equity prices up (expected returns down) and leave the safe asset underpriced (high expected return). The equilibrium implication:

$$\frac{{\mu }_B-r_f}{{\sigma }_B}>\frac{{\mu }_S-r_f}{{\sigma }_S},$$

i.e. the Security Market Line is too flat - safer assets deliver higher risk-adjusted returns than riskier ones. An investor willing to borrow can then hold the safe-asset-heavy (risk-parity-flavored) portfolio, lever it, and capture the flatness - the “Betting Against Beta” (BAB) intuition of Frazzini & Pedersen (2010). Risk parity is, in this light, the cross-asset incarnation of low-beta investing: overweight the low-$\beta$ leg, lever up.

The mean-variance logic of the split. With Qian (2005)‘s argument, risk parity is mean-variance optimal under two assumptions - equal Sharpe ratios across assets and zero/low cross-correlation:

$$\frac{{\mu }_i-r_f}{{\sigma }_i}=\text{SR}\forall i\Rightarrow w_i\propto {\sigma }_i^{-1}(=\text{ERC}),$$

i.e. equal-risk weights are exactly the Markowitz-optimal weights if every asset is priced to its risk. This is the “parity is efficient” theorem - and it is also its own audited condition, because it exposes exactly which assumption the practitioner is betting on (the flat-SML/equal-Sharpe view).

Leveraging to match a benchmark. From a risk-balanced (low-vol) book with volatility ${\sigma }_p$ and excess return ${\mu }_p$, scaling by $L$ gives

$${\sigma }_L=L{\sigma }_p,{\mu }_L=L{\mu }_p,\text{Sharpe unchanged}={\mu }_p/{\sigma }_p.$$

So leverage raises return at fixed Sharpe ratio - the entire premise for using it. The catch from page 05: $L$ is a loan, and during forced de-leveraging the realized Sharpe is far from the model’s.

---

3. Computational Implementation - the flat-SML story, reconstructed

Reproduce Asness, Frazzini & Pedersen’s (2012) illustrative figures on their Long-Sample summary statistics (U.S. stocks 10.8% / 18.9% vol; Treasuries 5.2% / 3.4%; risk-free 3.6%; $\rho =0.2$): build the tangency portfolio, confirm it over-weights bonds (~88/12, as the paper reports), then lever the parity portfolio to match stock volatility and compare Sharpe.

Two numbers tell the whole story. First, the max-Sharpe (tangency) portfolio splits ~88% bonds / ~12% stocks - even the unconstrained mean-variance answer over-weights the safe asset, matching Asness et al.’s reported ex-post tangency. Second, levered parity earns 10.39% excess at 18.9% volatility vs 7.20% for holding equities at that same 18.9% - a ~3.2% per-year edge, consistent with the “risk-parity premium” the leverage-aversion theory and Asness’s long-sample evidence describe. (These are a stylized reconstruction from the paper’s summary statistics, not the paper’s monthly-rebalanced backtest; the mechanism, the 88/12 tangency split, and the flat-SML premise are the takeaway.)

The critique, stated plainly. Asness, Frazzini & Pedersen’s own warning is the sentence above the code: risk parity is not riskless, and it is not “always better diversification.” The equal-Sharpe (or flat-SML / low-beta) assumption is precisely the assertion “equities are not paid enough to justify their risk share.” If you believe the equity premium is large, a stock-heavy 60/40 is efficient and risk parity is the wrong bet. Parity buys safety + leverage, and the leverage is the fragile half (page 05).

---

4. Failure Modes & First-Principles Breakdowns

1. The free-lunch trap. “Risk parity outperforms” is a claim about the premium, which depends on the flat-SML assumption staying true and on cheap leverage staying available - neither is a law of nature, and both reverse in inflation/rate shocks.
2. Leverage is a loan, and loans are callable. The 4.2× in the example is a model 4.2×; real forced de-leveraging (margin) converts a mild drawdown into a realized loss and breaks the parity promise. Asness et al. App. B quantify the cost of even higher financing rates.
3. Concentration masquerading under “parity.” Because parity over-weights the low-$\beta$ leg, a levered parity book can carry enormous capital concentration in bonds - a single-rate shock to the safe leg is a shock to the whole (levered) book. Risk-balance in shares ≠ capital-balance in dollars.
4. HRP is a robust alternative, not a silver bullet. López de Prado (2016) Hierarchical Risk Parity climbs a correlation tree and rebalances within/between clusters without ever inverting $\Sigma$ - robust when $N\gtrsim T$, which ERC is not. But HRP abandons the clean budget meaning (you cannot express $R{C}_i=b_i\sigma$ inside it), so the choice is a robustness-vs-interpretability trade, not a clear win.

---

5. References

• Asness, Frazzini & Pedersen: Leverage Aversion and Risk Parity, Financial Analysts Journal 68(1):47–59 (2012)
• Black, Fischer: Capital Market Equilibrium with Restricted Borrowing, Journal of Business 45(3):444–455 (1972)
• Frazzini, Andrea & Pedersen, Lasse H.: Betting Against Beta, (2010)
• Qian, Edward (2005)
• López de Prado, Marcos: Building Diversified Portfolios that Outperform Out-of-Sample, Journal of Portfolio Management 42(4) (2016)

---

6. Connected Graph Bridges

• Back: 05 · Failure Modes · Index Hub
• Forward: Hierarchical Risk Parity (HRP) · Covariance Shrinkage & RMT
• Base: Tangency & CAPM · Convex Optimization