The 60/40 portfolio is the default recommendation in personal finance: 60% equities for growth, 40% bonds for stability. It sounds balanced. In risk terms it is not remotely balanced, and the gap between how it sounds and what it is explains a lot of disappointed investors.
The arithmetic
Take reasonable long-run numbers: equities at roughly 16% annualised volatility, bonds at roughly 4%, low correlation between them.
Portfolio variance contribution scales with the square of the weighted volatility. The equity leg contributes on the order of (0.6 × 16)² = 92; the bond leg (0.4 × 4)² = 2.6. Even before accounting for correlation, equities account for well over 90% of the portfolio’s variance.
The bonds are not diversifying the portfolio in any meaningful sense. They are reducing its size. A 60/40 portfolio behaves, day to day and in every crisis, essentially like an equity portfolio scaled down — which is why 60/40 investors are consistently surprised by how much their “balanced” portfolio falls in an equity bear market.
What risk parity changes
Risk parity allocates so that each holding contributes equally to portfolio variance, rather than equally to portfolio rupees.
Mechanically this means large positions in low-volatility assets and small ones in high-volatility ones. A risk-parity portfolio across the same equity and bond sleeves might hold 25% equities and 75% bonds — which looks absurd on a holdings table and is correct on a risk report, because that is where the two legs contribute equal risk.
The obvious objection is that this must produce lower returns, since it holds more of the low-return asset. That is exactly right, and it is why institutional risk-parity funds apply leverage: get the balanced risk profile first, then scale the whole thing up to a return target. Bridgewater’s All Weather is the best-known implementation of this idea.
A retail implementation without leverage cannot do the second step. It accepts a lower expected return in exchange for a smoother path. Whether that trade is worth it depends entirely on whether path matters to you — and for anyone who might need the money, or who might sell at the bottom, it does.
Why naive risk parity is not enough
Here is where most simple implementations break.
Suppose your universe is twelve ETFs: eight equity (Nifty, Sensex, Next 50, Midcap, Bank, IT, Pharma, Auto), two gold, two debt.
Naive risk parity treats these as twelve independent risks and equalises variance contribution across all of them. But the eight equity ETFs are not eight risks. They are one risk — Indian equity — expressed eight ways, with pairwise correlations frequently above 0.8. Equalising contribution across twelve holdings therefore hands roughly eight-twelfths of the risk budget to a single underlying factor.
The portfolio looks diversified. It is a concentrated equity bet with a naming convention.
Hierarchical risk parity
Marcos López de Prado’s “Building Diversified Portfolios that Outperform Out-of-Sample” (2016) addresses this with a three-step procedure:
- Tree clustering. Convert the correlation matrix into a distance measure and hierarchically cluster the assets. The eight equity ETFs form a tight cluster; gold sits apart; debt sits further apart.
- Quasi-diagonalisation. Reorder the covariance matrix so similar assets sit adjacent, concentrating large covariances near the diagonal.
- Recursive bisection. Split the tree top-down, allocating between clusters by inverse cluster variance at each split before allocating within them.
The result is that the equity cluster receives a budget as a cluster, which is then subdivided among its members. Adding a ninth equity ETF does not increase the portfolio’s equity exposure; it splits the existing equity budget nine ways instead of eight.
That property — insensitivity to how many ways you happen to slice the same underlying exposure — is what makes HRP the right choice for a universe assembled from whatever instruments are liquid rather than from a clean set of independent asset classes.
The second reason: numerical stability
There is a less-discussed but equally practical argument.
Classical portfolio optimisation, including minimum variance and naive risk parity, requires inverting the covariance matrix. When assets are highly correlated, that matrix is close to singular, and inverting a near-singular matrix produces enormous, unstable weights that swing wildly with small changes in the input data. This is the well-documented reason mean-variance optimisation performs badly out of sample despite being optimal in-sample.
HRP never inverts the covariance matrix. It only uses it to compute distances and cluster variances. López de Prado’s simulations found HRP delivered lower out-of-sample variance than both minimum-variance and inverse-variance allocation, despite not being optimal in-sample — a good illustration of the general point that a robust method beats an optimal one when the inputs are estimated.
For a twelve-asset Indian ETF universe where nearly everything is correlated with the Nifty, this is not a theoretical concern.
The Indian caveat that matters
The standard argument for gold in an Indian portfolio is diversification. It is weaker than usually presented, for a currency reason.
An Indian investor’s gold return combines the dollar gold price and the USD/INR rate. Rupee depreciation raises rupee gold returns independently of what gold does in dollars.
Rupee depreciation is also frequently associated with foreign portfolio outflows, which pressure Indian equities. So there are episodes where the rupee weakens, gold rises in rupee terms, and equities fall — diversification working. But there are also risk-off episodes where global investors sell everything, dollar gold falls, and Indian equities fall too. In March 2020 gold sold off alongside equities before recovering, as investors liquidated whatever was liquid.
The general point: correlations converge in crises, and cross-asset diversification is a normal-times benefit that partially fails in precisely the scenarios you most wanted it. That is not a reason to skip it. It is a reason not to treat it as insurance.
Practical notes
Liquidity screening is doing heavy lifting. Many Indian ETFs trade a few lakh rupees a day. An illiquid ETF’s price series is stale, and a stale series has artificially low measured volatility — which risk parity rewards with a larger allocation. The screen is not garnish; without it the weighting scheme is actively pointed at the wrong holdings.
Rebalance quarterly. Correlation and volatility estimates move slowly, and rebalancing on estimation noise generates cost without changing the risk profile.
Estimation windows matter more than the algorithm. Both the clustering and the variances depend on a covariance estimate from a finite sample. Any of these methods is a sophisticated way of extrapolating from that sample, and no amount of algorithmic elegance fixes a window that does not contain the regime you are about to enter.
Try it
The All-Weather Risk Parity template screens Indian ETFs for liquidity, holds the twelve most liquid, and sizes them by hierarchical risk parity with a quarterly rebalance.
The comparison that makes the case: run it with equal weighting instead, then with plain risk parity, then with HRP. Returns will likely be highest for equal weight — you have concentrated into equity. Look instead at the maximum drawdown and at the composition of each book. Equal weight will be an equity portfolio wearing a multi-asset label, and seeing that on your own screen is more convincing than any argument about variance contribution.
Further reading
- Volatility targeting: sizing is a more reliable lever than timing applies the same volatility estimate across time rather than across holdings
- Cross-asset momentum: the version that held up out of sample is a rotation built on the same set of Indian ETFs
- Why blending factors beats picking one makes the equivalent argument one level down, inside the equity sleeve
Glossary: risk parity, inverse volatility weighting, minimum variance, volatility, rebalancing.