Should Gold Be Part of a Long-Term Portfolio?

Does gold belong in a long-term portfolio?

Kabir has been told several times that he should hold some gold. The reasons offered have varied — a hedge, a crisis asset, protection against inflation, something every Indian portfolio should have — and none of them has come with a number attached.

He is the reader this site's evidence articles are written for, and he has learned to notice when a recommendation arrives as a story. His question is the right one: not whether gold sometimes goes up, but whether adding it to a portfolio he already holds would have made that portfolio better, and by how much, after costs.

That is a portfolio question, and what follows runs most of that test for real, on a GoldBeES ETF NAV series covering 16.7 years. What it finds first: the roles a gold allocation could coherently play, and why the popular arguments for holding it are not arguments.

Why judge gold by its effect on the whole portfolio?

The first thing to fix is what the question even is, because it is usually asked in a form that cannot be answered.

"Has gold gone up?" is not the question. Any asset that has risen over some period can be presented attractively by choosing the period, and gold has had both long strong stretches and long flat ones. Nor is "does gold do well in a crisis?" sufficient on its own, because an asset that does well in the rare bad year and poorly in the many ordinary ones may still leave a portfolio worse off overall.

The question that matters is a portfolio question. Take the mix Kabir would otherwise hold, replace some fraction of it with gold, run both through the same history, and compare — not just the return but the depth of the falls, and not at one allocation but across a range of them. A diversifier earns its place by improving the whole, and it can only be assessed as part of the whole.

That framing also rules out the most common way this argument is made. Pointing at a single crisis in which gold rose is an anecdote about one episode, and it is exactly the reasoning data mining and backtest overfitting warns about: with enough episodes to choose from, some will support any asset.

What can a gold allocation be for?

Three roles are coherent, and they are not the same as each other. Which one is being bought is worth being able to say.

The first is diversification: gold's price is driven by different things from Indian company earnings, so it may move differently in some stressed periods and reduce how far the whole portfolio falls. This is the strongest of the three in principle and the one that most needs measuring, because "may move differently" is not a property that can be assumed.

The second is currency exposure. Because the rupee gold price contains the exchange rate, a holding carries a position that gains when the rupee weakens, as how gold prices work sets out. That is a real mechanism rather than a claim about sentiment, and it is the one part of the case that follows from arithmetic rather than from history.

The third is not an investment role at all: funding a future purchase of gold. If Meera intends to buy jewellery in three years, holding gold matches the asset to the liability, and the reasoning is completely different from the other two.

What is not a legitimate role: capital protection, guaranteed income, or a long-term growth engine. Gold produces nothing while held, so it has no internal source of return to compound.

What test would show whether gold helps?

The test is not complicated to specify, and setting it out is worth doing on its own, because it is also the standard for judging any study somebody produces. Six of the seven points below are now run, in the next section; the fifth — full transaction costs and the tax due on each rebalancing trade — is not, and that gap is disclosed there rather than folded into a number that looks more finished than it is.

1. Use total returns on both sides, in the same currency, over the same dates. Comparing gold's price change against an equity index's price change omits dividends from one side and understates the alternative by a margin that compounds.
2. Test a portfolio, not the asset. Several gold weightings against the zero-gold portfolio, so the answer is a curve rather than a verdict.
3. Run every available period, not a chosen one, and report the distribution. A diversifier that helps in the typical period and fails in the worst ones is not doing the job it was bought for.
4. Report the falls, not just the returns. The case for gold is mostly about the depth of drawdowns, so drawdown is the primary measure and return is the cost side.
5. Charge the costs a real person pays — the spread between buying and selling, storage or fund expenses, and the tax treatment of the specific form held, which differs between forms.
6. Include a control. Something else uncorrelated in its place, so that any benefit found is attributable to gold rather than to holding anything that is not equity. Without this, a positive result may only be rediscovering that mixing in a different asset reduces volatility.
7. Rebalance both portfolios by the same rule, because most of a diversifier's measured benefit comes from rebalancing into it rather than from holding it.

What does the Indian record show about holding gold?

Two gold weightings — a tenth and a fifth of the portfolio — against a zero-gold control, rebalanced back to target every year, run over every rolling ten-year window in the 16.7-year overlap between this site's Nifty total-return series and its gold series: 1,642 windows, 4 January 2010 to 11 September 2026.

Allocation	Typical yearly return over 10 years	Worst 10-year run	Best 10-year run	Typical fall along the way
No gold (control)	+13.22%	+5.13%	+15.71%	11.35%
10% gold	+13.12%	+5.80%	+15.87%	6.96%
20% gold	+12.78%	+6.37%	+16.23%	4.28%

Show these numbers as a table

Strategy	Poor outcome	Typical	Good outcome
No gold (control)	+8.97%	+13.22%	+14.57%
10% gold	+9.35%	+13.12%	+14.25%
20% gold	+9.68%	+12.78%	+14.31%

Adding a tenth of the portfolio to gold moved the median ten-year outcome from +13.22% a year to +13.12%, moved the worst window from +5.13% to +5.80%, and roughly halved the typical drawdown from 11.35% to 6.96%. Doubling the allocation to a fifth pushed every one of those numbers further the same way: median +12.78%, worst +6.37%, typical drawdown 4.28%.

That is the shape a genuine diversifier produces: a small, consistent cost to the typical outcome, taken in exchange for a meaningfully better worst case and a shallower typical fall — not the shape a story produces, which is all upside and no cost. It is also not proof the same trade repeats in the next crisis. 1,642 overlapping ten-year windows drawn from one 16.7-year record are not independent observations, any more than the equity-only backtests elsewhere on this site are, and the record contains at most one full commodity cycle in gold specifically.

Three things this does not do. It does not charge the spread paid buying and selling gold, the fund expense already embedded in the ETF's own NAV notwithstanding, or the tax due each time the annual rebalance trims one side — both arms are rebalanced the same way, so this understates each of them by roughly the same amount rather than favouring one, but it is not charged. It does not separate a move in the metal from a move in the rupee, which needs the international US-dollar gold price this site does not hold — that decomposition stays with how gold prices work. And it tests one instrument, the ETF form; a physical holding or a sovereign gold bond carries different costs and, for the bond, a different tax treatment, neither modelled here.

What does this mean for a gold allocation?

If the decision is to hold gold, the measured result above is a starting point for sizing it, not a recommendation of one figure over another — 1,642 overlapping windows from one 16.7-year record is real evidence and a thin one at the same time. The target and the band still get decided before buying, and it is treated like any other allocation — reviewed on a date, rebalanced when it leaves the band, and sold down after a strong run even while the commentary is enthusiastic. An allocation that only ever grows is not an allocation. A larger allocation than the ones tested here is a bet that the measured relationship holds even more strongly than it has so far; a smaller one costs less if it does not.

If he has near-term spending that must not fall in value, that money should be protected directly, in something stable, rather than by relying on gold's reputation in a crisis. Gold has had deep and lengthy declines — 27.40% at worst in the record this site holds — and money needed on a fixed date cannot wait out a recovery.

What to take away

The question is whether gold improves the portfolio Kabir already holds, after costs, measured on falls as well as returns — not whether gold has ever gone up. Three roles are coherent — diversification, currency exposure, funding a future gold purchase — and capital protection and long-term growth are not among them.

Run on 16.7 years of real Indian data, a modest gold allocation — a tenth to a fifth of the portfolio, rebalanced annually against the Nifty — has cost a little of the typical return and bought a meaningfully better worst case and a shallower typical drawdown, which is what a genuine diversifier is supposed to do. That is a real, measured result, not a story, and it is also thinner evidence than it sounds: one overlapping record, no transaction costs charged, and no separation of the metal's own return from the rupee's. A written target and band turn a decision to hold it into a real allocation rather than a drift; and near-term money is protected somewhere else regardless, since gold has had declines too deep and long (27.40%, 4.5 years underwater) to wait out on a fixed date.