Risk Parity and the All Weather Logic

In 1996, Bridgewater Associates launched the All Weather fund with a thesis that sounded almost heretical: a portfolio balanced by risk would outperform one balanced by capital over a full market cycle. Not in every quarter. Not in every year. Over decades, through recessions, expansions, wars, and disinflations.

The thesis was right. All Weather has compounded at roughly the same rate as the S&P 500 with about a third of the volatility. It’s now one of the largest hedge funds in the world, sitting on over $100 billion in part because of this approach.

This article walks through the math and the logic. The core idea is “equal risk contribution”: allocate capital so that each asset contributes equally to total portfolio risk. Not 60/40 between stocks and bonds. Allocate based on how much volatility each asset brings to the table.

The Problem With 60/40

A traditional pension allocation is 60% stocks, 40% bonds. Sounds balanced. Let’s check.

Suppose stocks have 16% annualized volatility and bonds have 5%. With 60/40 weights and a correlation of around 0.1, portfolio volatility is:

So 9.9% vol. The contribution to risk from stocks:

Wait, let me redo this. The marginal contribution to risk (MCR) for asset $i$ is $(\Sigma w)_i / \sigma_p$. The component contribution to risk is $w_i \cdot (\Sigma w)_i / \sigma_p$. Sum over all assets equals $\sigma_p$ (a fundamental identity).

For the 60/40 example:

• $$(\Sigma w)_{\text{stocks}} = 0.6 \cdot 0.0256 + 0.4 \cdot 0.0008 = 0.0157$$
• $$(\Sigma w)_{\text{bonds}} = 0.6 \cdot 0.0008 + 0.4 \cdot 0.0025 = 0.0015$$
• $$\sigma_p = \sqrt{0.6 \cdot 0.0157 + 0.4 \cdot 0.0015} = \sqrt{0.0100} = 0.10$$

Component contributions:

• Stocks: $0.6 \cdot 0.0157 / 0.10 = 0.094$
• Bonds: $0.4 \cdot 0.0015 / 0.10 = 0.006$

Stocks contribute 0.094 to total portfolio vol of 0.10. That’s 94% of portfolio risk from the “60%” allocation to stocks. The “balanced” 60/40 is, in risk terms, a 94/6 bet on stocks. Calling this diversified is marketing.

Equal Risk Contribution

The risk parity portfolio is defined by:

Each asset’s dollar-volatility contribution to the portfolio is the same. Total risk equals the sum of equal contributions: $\sigma_p = \sum_i w_i (\Sigma w)_i / \sigma_p$. With $N$ assets, each contributes $\sigma_p / N$.

In the two-asset case:

Solving:

With $w_b = 1 - w_s$, this is one equation in one unknown. Closed-form:

For the 60/40 example with $\sigma_s^2 = 0.0256$, $\sigma_b^2 = 0.0025$, $\sigma_{s,b} = 0.0008$:

Whoa. To get equal risk contribution with these volatilities, you need 6.4% in stocks and 93.6% in bonds. That’s not a typo. If you naïvely think 60/40 is balanced, you haven’t done the math.

In practice, no one runs an actual 6/94 portfolio. Instead, they lever up the bond allocation to a target volatility. If you want 10% portfolio vol, lever bonds by 1.6x or so, keep stocks at 6.4% unlevered, and you get 10% vol with equal risk contribution. This is exactly what All Weather does.

The General Case

For $N$ assets, the equal-risk-contribution condition is a system of $N-1$ equations:

There’s no closed-form solution in general. You solve it numerically, typically with a root-finding algorithm or a constrained optimization.

Spinu (2013) gave a clean algorithm:

1. Start with $w = \mathbf{1}/N$.
2. Compute risk contributions $RC_i = w_i (\Sigma w)_i / (w^\top \Sigma w)$.
3. Adjust weights: $w_i \leftarrow w_i \cdot \alpha / RC_i$ where $\alpha$ normalizes to $\sum w_i = 1$.
4. Iterate.

This converges quickly for most covariance matrices. For degenerate cases (highly correlated assets, near-singular $\Sigma$), you may need regularization or a trust-region method.

A more modern approach treats it as a constrained optimization:

This is a quadratically-constrained quadratic program (QCQP). It’s convex if you’re careful about formulation. Most production systems use a smoothed version.

Bridgewater’s All Weather

Ray Dalio’s specific implementation (described in Dalio (2017) and various Bridgewater research papers) goes like this:

1. Identify the major economic “drivers”: inflation growth, deflationary growth, reflationary growth, deflationary recession. (Or sometimes expressed as growth, inflation, and the cross of those two.)
2. For each environment, identify asset classes that historically do well: equities (growth), nominal bonds (deflation), inflation-linked bonds (inflation), commodities (reflation), gold (extreme environments).
3. Allocate roughly equal risk to each of these environments.
4. Use leverage to hit a target portfolio vol (typically 10-12%).

The resulting portfolio has a strong risk-parity flavor. The exact asset weights depend on the covariance matrix and the leverage target, but the philosophy is risk-balanced, not capital-balanced.

Key empirical observation: through the 2008 financial crisis, the S&P 500 was down 37%. The 60/40 was down roughly 25%. All Weather was down about 5%. Over the full cycle 1996-2020, All Weather returned comparable to the S&P with about 1/3 the vol. Risk-adjusted, it dominated.

Why Does It Work?

Three reasons, in increasing order of importance.

First, diversification does something. Most of 60/40’s risk is in stocks. If stocks drop 40%, you drop 25%. If you balance risk, you don’t drop 25% in a stock crash; you drop something more like the weighted-average of asset returns in that scenario. Which is much better if the assets are genuinely uncorrelated in the crash.

Second, leverage is a tool. Bonds are less risky per dollar, so you lever them. The leverage is funded by borrowing at short rates, capturing the term premium and the credit premium. This works as long as the term premium is positive on average — which it has been, more or less, for the last 200 years.

Third, rebalancing is alpha. Every quarter, you rebalance back to target weights. When an asset has outperformed, you sell it; when it has underperformed, you buy more. This is a systematic trend-following / mean-reversion overlay that picks up roughly 1-2% per year on a risk-parity book.

The third reason is the one nobody talks about but matters most. Risk parity without rebalancing is just an oddly-weighted portfolio. Risk parity with rebalancing is a dynamic strategy. Bridgewater’s edge is in the second part.

Common Criticisms

Risk parity has critics. Here are the main objections.

“Risk parity lost in 2013-2016.” True. In a low-rate environment with rising bond prices, leverage on bonds was a drag. Risk-parity funds underperformed 60/40 by several hundred basis points. Bridgewater’s response: this is one regime in one cycle; over a full cycle, risk parity dominates. Reasonable people disagree on how to weight regime evidence.

“Risk parity is implicitly short volatility.” True. Levered bond allocations blow up if rates spike. The Taper Tantrum of 2013 and the 1994 bond rout both hurt risk-parity funds. The leverage amplifies mark-to-market drawdowns.

“Risk parity underestimates correlation in crises.” True. Covariance matrices computed from daily data underestimate tail correlation. In a crisis, everything becomes correlated, and risk parity stops being diversified. (This is the same critique that applies to all mean-variance approaches.)

“It only works because of falling rates.” Partially true. Risk parity has historically benefited from the secular decline in interest rates since 1980. If rates rise structurally, the term premium on bonds compresses or inverts, and the strategy suffers.

Despite these, risk parity has survived every crisis since its inception. The strategy is robust enough that criticisms have to be specific. The general point, balance by risk not by capital, is mathematically correct.

A Working Example

Three assets with different volatilities. Solve for the equal-risk-contribution portfolio.

import numpy as np
from scipy.optimize import fsolve

Sigma = np.array([
    [0.04,  0.005, 0.002],
    [0.005, 0.0225, 0.003],
    [0.002, 0.003,  0.01]
])

def risk_contributions(w, Sigma):
    sw = Sigma @ w
    port_var = w @ sw
    return w * sw / np.sqrt(port_var)

def erc_objective(w_free, Sigma):
    # Last weight is implied by budget constraint
    w = np.append(w_free, 1 - w_free.sum())
    rc = risk_contributions(w, Sigma)
    # Penalize deviations from mean RC
    return np.sum((rc - rc.mean()) ** 2)

from scipy.optimize import minimize
result = minimize(
    erc_objective,
    x0=np.array([1/3, 1/3]),
    args=(Sigma,),
    method='Nelder-Mead'
)

w_erc = np.append(result.x, 1 - result.x.sum())
print("ERC weights:", w_erc.round(4))
print("Risk contributions:", risk_contributions(w_erc, Sigma).round(4))

The output: low-vol assets get larger weights, high-vol assets get smaller weights. The risk contributions come out approximately equal.

What the Literature Says

• Dalio (2017). “Principles for Navigating Big Debt Crises.” Bridgewater Research. Free PDF on Bridgewater’s website. The All Weather philosophy.
• Qian (2005). “Risk Parity Portfolios: Efficient Portfolios Through True Diversification.” PanAgora Asset Management Research. The original academic-ish paper on risk parity.
• Maillard, Roncalli & Teïletche (2010). “The Properties of Equally Weighted Risk Contribution Portfolios.” Journal of Portfolio Management 36(4): 60-70. The math.
• Spinu (2013). “An Algorithm for Computing Risk Parity Weights.” Working paper. The iterative algorithm.
• Chaves, Hsu, Li & Shakernia (2012). “Risk Parity Portfolio vs. Other Asset Allocation Heuristic Portfolios.” Journal of Investment Management 10(4): 52-69.
• Asness, Frazzini & Israel (2020). “Measuring the Real Estate Exposure of Risk Parity Strategies.” Journal of Portfolio Management 46(6): 100-112. A key criticism: risk parity is implicitly long real estate.

The Spinu Algorithm in Detail

Spinu (2013) gave an iterative algorithm for solving the equal-risk-contribution system. The math is straightforward: at each step, adjust weights to drive risk contributions toward equality.

The iteration:

1. Initialize $w^{(0)} = \mathbf{1}/N$.
2. Compute portfolio variance $V = w^\top \Sigma w$.
3. Compute marginal contributions $m_i = (\Sigma w)_i$.
4. Compute component contributions $c_i = w_i m_i$.
5. Update $w_i^{\text{new}} = w_i \cdot \alpha / c_i$ where $\alpha$ normalizes to sum to 1.
6. Repeat until convergence.

The convergence criterion: $|c_i - V/N| < \epsilon$ for all $i$, where $V/N$ is the target equal contribution.

Why does this converge? Each step drives risk contributions toward equality. The updates are smooth (no jumps), and the system is well-conditioned for most covariance matrices. Typical convergence in 10-50 iterations.

For pathological covariance matrices (near-singular, large condition number), the iteration may diverge or converge slowly. Adding a small diagonal to $\Sigma$ before the iteration usually fixes this:

with $\epsilon$ around $10^{-8}$.

The Leverage Question

Risk parity requires leverage to hit a target volatility. This is where the strategy gets controversial.

Suppose you want 10% portfolio vol. Your asset universe has an average vol of 8% (mix of stocks and bonds). To hit 10% vol with ERC, you need to lever the portfolio by 10/8 = 1.25x. Where does the leverage come from?

Two options. (1) Borrow at the risk-free rate and invest the proceeds. This is “levered risk parity.” (2) Use derivatives (futures, swaps) to get exposure without explicit borrowing.

Option 1 is cleaner economically but introduces credit risk and term risk. Option 2 is operationally simpler but introduces counterparty risk and basis risk.

Bridgewater uses both. Their All Weather fund borrows short-term to fund long-term bond allocations. They also use futures for parts of the portfolio.

The leverage is the source of risk parity’s returns — and its risks. When rates rise and bond prices fall, levered bond positions lose money. This is the standard critique.

The Rebalancing Bonus

Every quarter (or month), risk parity portfolios rebalance back to target weights. This is a built-in systematic strategy: sell assets that have outperformed, buy assets that have underperformed.

The rebalancing bonus is empirically around 1-2% annualized for risk parity portfolios (Harvey, Hoyle & Korgaonkar, 2018). It comes from two sources:

1. Mean reversion of asset returns: assets that outperform one period tend to underperform the next. Rebalancing captures this.
2. Volatility pumping: when portfolio vol drops below target, you lever up to restore it. This systematically buys volatility when it’s cheap.

Volatility pumping is the deeper source. The idea: when realized vol is low, option-implied vol is also low, and levering up captures the “carry” of volatility. When realized vol is high, you delever and avoid the worst of the drawdown.

Bridgewater’s research suggests about half the rebalancing bonus comes from mean reversion, half from volatility pumping. Either way, the bonus is real.

Regime Conditioning: A Natural Extension

Risk parity allocates the same way regardless of the macro regime. A natural extension: condition the allocation on the regime.

Define regimes by observable signals (yield curve slope, inflation trend, credit spreads). For each regime, compute the historical covariance matrix of asset returns. Solve ERC within each regime.

The regime-conditional covariance matrix is more conservative — it captures the covariance structure that’s relevant for the current environment. This typically results in more defensive allocations in late-cycle regimes.

Implementation requires (a) a regime classification model and (b) conditional covariance estimation. Both are non-trivial but feasible. The result is a portfolio that adapts to changing macro conditions.

This is what Bridgewater’s actual risk parity implementation does, more or less. Their “All Weather” framework explicitly conditions on the inflation-growth matrix.

The 2008 Stress Test

Risk parity’s signature stress test was 2008. The S&P 500 was down 37%. Long-only 60/40 was down 25%. Risk parity (Bridgewater All Weather) was down about 5%.

Why did risk parity survive when many diversified portfolios didn’t? Three reasons.

First, bond allocations (levered) gained value as rates fell in the flight to safety. The levered bond exposure was a hedge against the equity crash.

Second, the diversification across asset classes was genuine — correlations in 2008 stayed moderate. The risk parity portfolio wasn’t forced to liquidate.

Third, the rebalancing discipline worked: as equities fell and bonds rose, the portfolio sold bonds and bought equities. This was counterintuitive but correct; it captured the mean reversion of the cross-asset relationship.

The 5% drawdown was painful but recoverable. Investors who held through 2009 saw a full recovery by mid-2009.

The 2013-2016 Stress Test

Risk parity had a different problem from 2013-2016. The “taper tantrum” of 2013 and the gradual Fed tightening put upward pressure on rates. Levered bond positions lost money.

In 2013, the Bridgewater All Weather fund was reportedly down 3-4%. Other risk-parity funds were down more. The strategy underperformed 60/40 by several hundred basis points.

This is the regime where risk parity’s implicit short-volatility bet hurts. Rates rising means bond prices falling means levered bond positions losing money means volatility of bond returns increasing means the leverage amplifying losses.

Risk parity survived 2013-2016 but lost its shine. Investors who had embraced risk parity in 2008-2009 started questioning the strategy.

The honest assessment: risk parity is not a “beat the market” strategy. It’s a “do better than the market on risk-adjusted basis across regimes” strategy. In some regimes, it underperforms. In others, it dominates.

What to Remember

Risk parity allocates by risk contribution, not capital. In a typical 60/40 portfolio, stocks contribute 90%+ of risk despite being “only” 60% of capital. ERC fixes this.

The math is one equation per asset: equal dollar-volatility contributions. Closed form for 2 assets, iterative for N (Spinu’s algorithm).

It works because (a) genuine diversification survives most market regimes, (b) leverage on low-vol assets captures term premium, and (c) rebalancing is a free source of return.

It has real risks: levered bond exposure is short vol, and tail correlation breaks diversification. Critics are not wrong. But over the full sample, the strategy has delivered.

The leverage question is the source of both return and risk. The rebalancing bonus is real (1-2% annualized). Regime conditioning is a natural extension that adapts allocations to the macro environment.

The 2008 stress test was a win. The 2013-2016 stretch was a loss. Both are part of the strategy’s full-cycle behavior.

Next article: hierarchical risk parity. López de Prado’s 2016 attempt to combine the robustness of risk parity with the structure of MVO.