Property, commodities, gold and cash

The idea behind all of them

Every forecast here starts with the same question: what does this asset actually pay you, and what happens to that payment over time?

A property fund collects rent. An infrastructure fund collects tolls and utility bills. Cash collects interest. Those three have something to compound and their expected returns follow from the income plus whatever that income does over time.

Property, or REITs

A REIT is a listed company that owns buildings and passes most of the rent to shareholders. So the forecast is the rent it pays now, plus what happens to rents over time, adjusted for interest rates.

The dividend yield is the actual MSCI US REIT index yield, around 3.5 percent. Rental growth is assumed to match inflation, which is reasonable because commercial leases are commonly linked to it.

There is now no valuation adjustment at all. Reinstating one would need a reliable long-run REIT yield spread to revert towards and no free source for one was found, so the current gap between the REIT yield and the Treasury yield is shown on the page as information rather than being fed into the answer. At the moment REITs yield less than Treasuries, which historically has meant property is not cheap.

Infrastructure

Toll roads, water companies, power networks, pipelines. The appeal is that the charges are often set by regulators with an explicit link to inflation, so the income tends to keep up with prices without anyone having to negotiate.

The yield is the actual MSCI World Infrastructure index yield, around 3.5 percent. It was previously assumed to be the 10-year Treasury plus half a point, which overstated it by more than a point and a half.

The 80 percent inflation linkage and the half point of real growth are assumptions rather than measurements. Regulated charges usually track inflation closely but not perfectly, and the assets need continual capital spending to keep working.

Commodities

Almost nobody holding commodities owns any. A commodity fund holds futures contracts and Treasury bills, and its return comes from three places: the interest on the bills, the change in commodity prices, and the cost or benefit of continually replacing expiring contracts.

That third piece, called roll yield, is where a correction was needed. When the price of a contract for delivery next year is higher than one for delivery next month, replacing contracts costs money. That situation, contango, has been far more common than the reverse since the mid 2000s, so rolling has generally been a drag rather than a benefit.

Gold

Gold pays nothing. It produces no rent, no interest and no earnings, so there is no income to compound. A ten-year forecast is therefore a view about what one ounce will be worth against something else, and the whole question is what that something else should be.

This model used to answer inflation, on the reasoning that gold holds its purchasing power. Tested against the record, that reasoning does not hold up. Across ten-year windows since 1975, gold’s correlation with realised inflation is about zero. Over the sixteen years to 2026 gold returned 8.5 percent a year while US consumer prices rose 2.6 percent, so the old model would have forecast about 2.2 percent and been wrong by more than six points a year, every year.

Measured against the money supply the picture is completely different. Gold has been close to flat against broad money for fifty years while drifting upward against consumer prices, and of every measure tested the gold to money ratio was the best predictor of the next ten years of returns. So money is what gold is now anchored to.

Broad money here is the US, euro area, China, Japan and the UK added together and converted to dollars, which is most of the world’s money rather than all of it. Growth is measured over the trailing twenty years. The ratio is compared with its own median over thirty years.

That is why the three scenarios are far apart. The bull case assumes the ratio never reverts, which is the argument that central bank buying and moves away from the dollar have permanently repriced the metal, and leaves you collecting money growth. The bear case assumes the ratio returns fully to its median. The base case assumes it travels a third of the way, deliberately gentler than the fitted relationship implies, because that relationship has called gold expensive every year for a decade while gold rose.

None of this is an argument against holding gold. It is an argument against expecting a repeat of the last decade. What gold has done historically is hold its value when other things do not, and that is a different job from building wealth.

Cash

Cash pays whatever short-term interest rates happen to be, which is knowable today and unknowable in three years. The question is what they will average over your horizon.

There is a good answer to that, and it is free: the bond market has already priced it. A ten-year government bond yield is approximately what investors expect the overnight rate to average over ten years, plus a premium for tying money up for a decade. Take the premium off and you have an estimate of cash.

The half point term premium is itself an estimate. Published estimates move about and have occasionally been negative. It is a reasonable middling figure and nothing more precise than that.

Private equity and hedge funds

These two are different from everything else here, because neither has a market price to work from. Private equity is valued by the people who own it, a few times a year. Hedge fund index returns are self-reported by the funds themselves. So there is no yield to read and no price series to trust.

The only honest approach is to price them off what they are actually made of, which is listed companies and cash, and both are therefore derived from the forecasts elsewhere in this tool. Change the equity forecast and these change with it.

Private equity

A buyout fund is, in economic terms, listed companies with borrowed money on top. Funds add roughly fifty pence of debt for every pound of equity, which is why they move more than the stock market does. AQR settle on 1.2 times a small-cap index, about 1.3 against a broad one.

The starting index matters more than the gearing does. AQR measured the 1.2 against public small caps, so that is what it is applied to here. An earlier version pointed it at our US large cap forecast instead, which paired a small-cap beta with a large-cap return and matched neither of the two figures AQR report. Because our small cap forecast currently sits well above our large cap one, that mismatch was holding the private equity number about five points too low.

Hedge funds

Ibbotson, Chen and Zhu decomposed hedge fund returns from 1995 to 2009 into 3.0 percent of alpha, 4.7 percent from market exposure and 3.4 percent of costs. The market exposure implies a beta of roughly 0.4 to equities, which is what is used here.

The alpha assumed is half a point, far below the 3 percent measured for that period. Two reasons. It has compressed as the industry has grown, and the measured figure was flattered to begin with: the same authors found reported hedge fund returns overstated by about 5.68 percent a year once survivorship and backfill bias are removed. Funds that fail stop reporting, and funds that survive add their good early years to a database when they join it.

Both are also difficult or impossible for most private investors to buy at the terms these figures assume. Access to good funds is the whole game in private equity, and the dispersion between the best and worst managers is far wider than in listed markets, so an average is less meaningful than it would be for an index fund.

Bitcoin

Bitcoin is the one asset on this page we do not forecast. You enter your own number, and the tool shows reference points beside the box to help you pick it.

Every other forecast here starts from income. A property fund pays rent, a bond pays a coupon, a company pays a dividend out of earnings, and the yield you can see today does most of the work in the answer. Bitcoin pays nothing to its holders, so that arithmetic has nowhere to start. This is a property of the asset rather than a mark against it, and it is the same reason gold gets the thinnest treatment of anything above. The difference is that gold has several thousand years of price history to lean on and Bitcoin has about sixteen.

What the published models are worth

Three models get quoted often enough to be worth answering directly. Stock to flow compares the existing supply to yearly new issuance. It fits the past closely, but its explanatory power largely disappears once you control for the passage of time, because scheduled halvings and the calendar move almost in lockstep. Metcalfe style models tie value to the square of the number of users, and share the same problem of describing history better than they call the future. The power law, which fits price against time on a log scale, has the best fit of the three, and it is a fitted trend line rather than a tested forecasting model.

Independent tests at horizons of one to six months have not found a model that reliably beats the assumption that tomorrow’s price is today’s. That finding is the reason this tab asks you for a number instead of printing one.

The four reference points

The past rates are compound annual growth over one, three, five and seven years, taken from a weekly price series that we store and refresh every Monday. They are the record, not a projection, and the spread between them is the useful part: seven year and one year figures on the same screen usually disagree violently.

Those two constants are fitted here over the whole stored series, back to September 2014, rather than copied from a chart. The fit has an R squared of 0.92 on log scaled data, which sounds better than it is, because almost anything that rises over time will fit a rising line on a log scale.

The adoption line asks a simpler question: if roughly three times as many people hold Bitcoin in ten years and the price scales with that, what annual return does it imply. It is written as a multiple rather than as a move between two ownership percentages because estimates of how many people hold Bitcoin today disagree with each other by a wide margin, and only the ratio changes the answer.

The digital gold line asks what return is implied if Bitcoin reaches half the value of all the gold ever mined. That figure comes from the World Gold Council’s estimate of about 216,265 tonnes above ground, priced at the latest gold price we hold, so it moves as gold moves. An earlier version of this page used a fixed figure that gold had long since left behind, which understated this reference by several points a year.

None of the four is a forecast, and they disagree with each other on purpose. If they all pointed at the same number it would mean the assumptions behind them were doing no work.

Where every number comes from

What this cannot tell you

The bear and bull cases for infrastructure, commodities and gold are fixed steps up and down rather than anything derived. They give a sense of range and should not be read as scenarios anybody has modelled.

None of these forecasts includes fund charges or tax, and commodity and gold funds are often more expensive to own than share funds.

The REIT and infrastructure figures are for listed funds you can buy on an exchange. Directly owned property behaves differently, is valued far less often, and is much harder to sell.

Above all, three of these five assets have most of their return coming from a price nobody can forecast. Treat the cash and property numbers as reasonably grounded, and the commodity and gold numbers as little more than a statement that they should roughly keep pace with prices.