ETF

Cross-Asset Momentum: The Version That Held Up Out of Sample

Saral Team · 1 August 2026

Momentum works within stocks, within bonds, within currencies, within commodities, and — more usefully — across them. Asness, Moskowitz and Pedersen’s “Value and Momentum Everywhere” (2013) documented both premia across eight markets and asset classes and found the momentum returns were positively correlated across classes, suggesting a common driver rather than eight coincidences.

The cross-asset version has one practical advantage over the single-stock version that matters more than the academic result: it has held up better since publication.

Why the cross-asset version survived

Single-stock momentum is crowded. It is implementable by any quant fund with a data feed, the universe is enormous, and capacity is high. Decades of arbitrage capital have compressed the premium.

Cross-asset trend is harder to crowd for structural reasons. The universe is small — a few dozen liquid instruments — so capacity is limited relative to the pool of capital that would want it. And the underlying driver is slow macro flow: a currency depreciating over eighteen months, a commodity cycle turning, a rate regime shifting. These trends are driven by real economic reallocation, not by information diffusing through analysts, and they do not disappear because someone published a paper about them.

For an Indian retail investor there is also an access argument. Getting exposure to gold, government bonds, and multiple equity segments used to mean separate accounts, separate custody, and in gold’s case physical storage. ETFs collapse all of it into one demat account and one order type.

Equal weight is wrong here

The single most consequential design choice in a cross-asset rotation is sizing, and the intuitive answer is actively harmful.

A gold ETF, a Nifty ETF and a liquid or gilt ETF have annualised volatilities that differ by close to an order of magnitude. A liquid fund ETF might run at 0.5% annualised; a sector equity ETF at 25% or more.

Allocate 20% of capital to each and the risk allocation is nothing like 20% each. The equity holdings will contribute nearly all portfolio variance and the debt holding essentially none. You have written down a diversified allocation and are running a concentrated equity bet with a rounding error attached.

Risk parity fixes the accounting by sizing positions so each contributes equally to portfolio variance rather than equally to portfolio rupees. In practice this means large positions in low-volatility holdings and small ones in high-volatility holdings — which looks wrong on a holdings table and is right on a risk report.

The trade-off is honest: risk parity without leverage produces lower expected returns than an equity-heavy allocation, because it holds more of the low-return, low-volatility assets. Institutional risk-parity funds lever the whole portfolio back up to a target return. A retail implementation without leverage is accepting the lower return in exchange for the smoother path.

Absolute, not just relative

The template requires a positive six-month trend before holding anything, not merely the best six-month trend.

The distinction is the difference between relative and absolute momentum, and it decides the strategy’s drawdown behaviour. A purely relative screen ranks the universe and buys the top five, which means in a market where every asset is falling it holds the five falling least — fully invested through a decline.

An absolute filter asks a prior question: is this asset trending up at all? If nothing passes, the strategy holds cash. Gary Antonacci’s dual momentum work makes the case that the absolute filter, not the relative ranking, is what does most of the drawdown reduction in trend systems.

Six months is a deliberate middle. Three-month trends are noisy and generate excessive turnover in a universe this small. Twelve-month trends are slow to recognise a regime change — and in cross-asset rotation, the regime change is the event you are trying to catch.

The Indian ETF caveats

Three frictions specific to this market that a NAV-based backtest will understate.

Premium and discount to NAV. An ETF’s market price can drift meaningfully from its underlying NAV when the authorised participant mechanism is not actively arbitraging it. Indian gold ETFs and some thematic ETFs have traded at premiums of several percent in thin conditions. A backtest running on traded prices captures this correctly; one running on NAV does not. Either way, buying at a premium and selling at a discount is a real cost the strategy will pay.

Liquidity is extremely uneven. A handful of Indian ETFs trade in genuine size daily. Many trade a few lakh rupees a day, and some go days without a print. The liquidity filter in this template is doing more work than it appears — without it, the screen will select illiquid ETFs whose stale prices produce fictitious trend signals.

Tracking difference. ETFs charge fees and incur costs, so they lag their index by more than the stated expense ratio. Over a rotation strategy’s holding period this is small relative to the signal, but it is not zero and it is always negative.

Where it fails

Whipsaw in choppy markets. Trend systems lose money when trends do not persist. A market that oscillates without direction generates a sequence of entries near local highs and exits near local lows. This is the trend-follower’s structural cost and it cannot be engineered away — it is the premium paid for the trends that do persist.

Correlations converge in crises. Diversification across asset classes works until it does not. In March 2020, equities, gold, credit and most other assets fell together as investors liquidated whatever they could sell. Cross-asset diversification is a normal-times benefit and a partial one in a genuine liquidity event.

Gold is a rupee bet as much as a gold bet. For an Indian investor, gold ETF returns combine the dollar gold price and USD/INR. Rupee depreciation lifts rupee gold returns independent of gold itself, which is one reason Indian gold and Indian equity have sometimes been positively correlated in exactly the periods where you wanted them not to be.

Try it

The ETF Core-Satellite Rotation template screens Indian ETFs for a liquidity floor, requires a positive six-month trend, holds the five strongest, and sizes them by risk parity with a monthly rebalance.

The instructive modification: switch the weighting to equal weight and re-run. Returns may well improve — you have concentrated into equity. Then look at the drawdown. The gap between those two curves is the entire argument for thinking in risk contribution rather than rupee contribution, and it applies well beyond ETFs.