Can you predict the gold price?

Short answer: not reliably. The textbook model that came closest just broke in 2022. Here is what the peer-reviewed evidence actually supports, what it does not, and the data to test any claim yourself.

The model that worked for fourteen years

For most of the post-2008 era, one relationship explained gold better than any other: real interest rates. When the real yield on government bonds falls, gold tends to rise. The logic is direct: gold pays no coupon, so the real yield you give up to hold it is its entire carrying cost. A 2025 paper in the Review of Financial Studies put a number on the mechanism — an extra percentage point of expected long-run inflation maps to roughly a 37% higher real gold price, and most of gold's value comes from its investment role rather than from jewelry or industry. The wider set of forces is covered in What actually drives the gold price.

That is a strong in-sample result. It fits the history well. Whether it forecasts the future is a separate question, and the gap between those two things is what this piece is about.

The chart that broke in 2022

Then, after Russia's invasion of Ukraine, the relationship came apart. Research from the European Central Bank documented the breakdown: central banks roughly doubled their share of global gold demand, from about 10% in the 2010s to more than 20%. They were not buying because real yields moved. They were buying for sanctions insurance and reserve diversification, a hedge against the dollar-based system itself. The mechanics of that bid are covered in How central-bank gold buying moves the market.

The numbers make the break impossible to miss. Across 2023–2025 real yields sat near 2%, a level the old model says should have weighed on gold. Gold rose about 65% anyway. The World Gold Council's 2025 demand data shows central banks bought 863 tonnes — the fourth-largest year on record — exchange-traded funds took in a record amount, and the average price was up 44% year on year, setting dozens of fresh all-time highs. A model trained only on pre-2022 data would have gotten all of it backwards.

What the evidence actually supports

Two honest conclusions survive the recent literature.

Drivers, not forecasts. Central-bank demand, geopolitical risk, and fund flows are measurable, and they clearly move the price. The catch is that you read them off the tape after they have already happened. Knowing what drove last month's rally tells you almost nothing about next month's.

Size is easier than direction. The one out-of-sample result that holds up is that a geopolitical-risk index helps forecast how much gold moves — work in Finance Research Letters by Gupta, Pierdzioch and Salisu shows this for realized volatility. It says nothing about which way. Sizing the swing is a tractable problem; calling the direction is the one nobody has cracked. (Whether gold reliably tracks inflation runs into the same timeframe trap, covered in Does gold actually hedge inflation?.)

How to spot a junk gold-forecasting claim

Search "gold price prediction model" and you will get a few hundred blog posts and at least one breathless video backtest, most of them claiming 90%-plus accuracy. Almost all fail one of four basic tests:

• No benchmark. If a model is not measured against the dumbest possible forecast — tomorrow's price equals today's, a "random walk" — the accuracy number means nothing.
• No out-of-sample test. Fit enough parameters and you can explain any history perfectly. The only honest test is performance on data the model never saw.
• No significance test. Beating the benchmark on error is not enough; the gap has to be big enough that luck cannot explain it.
• No transaction costs. An edge that disappears once you subtract the bid-ask spread was never an edge.

Even the serious literature counsels humility here. A 2024 study in Management Science by Da, Tang, Tao and Yang found that the flood of index-fund money into commodities, precious metals included, mostly pumps noise into prices rather than signal. More money chasing an asset does not make it easier to predict. If anything, it is the opposite.

Test it yourself

You do not have to take anyone's word for this, ours included. The minimal honest backtest is short:

1. Pull a clean, consistent daily price history from a single source.
2. Build the random-walk benchmark: tomorrow's forecast equals today's price.
3. Train your model on past data only, then forecast forward one step at a time.
4. Compare its error against the benchmark, then run a significance test on the gap.
5. Subtract realistic costs.

Clear that bar and you have something worth talking about. Most models do not clear it; they have memorized one stretch of history and mistaken that for a forecast. The goldprice.dev API gives you the clean, single-source daily OHLC history to run exactly this test, with the data source labeled on every response so you always know what you are testing — there is a worked example in Backtest a gold strategy with historical OHLC, and the live reference is at XAU/USD.

Where the science actually stands in 2026: gold has richer fundamentals than most currencies, yet its out-of-sample record for price direction is still weak, and the post-2022 shift has made historical models shakier than at any point since 2008. Anyone selling you a gold forecast should be able to show you their out-of-sample test. Most cannot.