People sometimes email us to suggest we show a smoothed version of the premium series alongside the raw one. A five-day moving average, or a Hodrick-Prescott filter, or even just a weekly close.
The argument is that the raw daily series looks noisy; a smoothed version would be more readable.
We have talked about it internally more than once. Each time, we have decided against it, and it is worth explaining why.
A smoothed premium series is not a more accurate representation of the market. It is a specific modeling choice with specific consequences.
A five-day moving average implicitly assumes that today's premium is best described as an average of the past five days. A weekly close implicitly assumes that Friday's reading is the "true" weekly value and the other four days were noise.
Both assumptions are defensible in some frameworks and indefensible in others. There is no neutral smoothing.
Once we embed a smoothing choice in the display, we are effectively telling readers "this is how you should look at the data." Which we do not want to do. We want readers to see what the exchanges actually settled at each day, decide how much of the movement they consider signal, and reach their own conclusion.
The methodology page describes the raw data source and the transformations we apply (unit normalization, FX conversion). Smoothing is a further transformation, and we would prefer readers to apply it themselves if they want it, rather than having us apply it invisibly.
There is also a practical problem. Different downstream users want different smoothing.
A physical bullion buyer thinking about a purchase window this month wants to see the last two weeks in detail. A macro researcher looking at multi-year regime shifts wants a monthly average.
A retail investor comparing "the SGE premium" to a Twitter claim wants a single number. We cannot provide all three simultaneously without either cluttering the interface or picking a single lens that suits none of them well.
The raw series works for all three purposes because each user can then apply their own smoothing (or none) as needed.
The one thing we do smooth is the display of small values near zero. A premium of 0.03% and 0.05% are not meaningfully different for any real purpose, so we round the display to two decimal places.
This is a display convention, not a data transformation; the underlying number stored and returned by our API is at higher precision.
If you have a strong preference for a specific smoothing in your own analysis, our JSON endpoint at /api/premium/history returns the daily series in a form easy to plot with any tool. Apply whatever transformation your analysis requires.
What you see on our dashboard is the exchanges' actual daily settlements, projected against COMEX, without opinion embedded in the shape of the line.
— The Editorial Desk
Related Posts
A Chronology of Major Gold Market Events, 2020–2026
A reference chronology of the events that actually moved gold prices, premium regimes, or physical delivery mechanisms from the start of the COVID period through mid-2026.
April 2025 Tariff Week: A Case Study in How a Policy Rumor Reshaped Premiums for Ten Days
A week-by-week account of what happened to COMEX–LBMA and MCX–LBMA premiums during the April 2025 US tariff scare, and what the data can and cannot tell us about who was actually moving metal.
Central Bank Gold Holdings by Country: Who Owns What, and What the Trends Show
A tour through the top central bank gold holders as of mid-2026, the buyers and sellers of the past decade, and what these numbers do and do not tell you about future prices.