Twenty years of front-month COMEX copper futures, split into intraday, overnight, and full daily returns. Friday is copper's strongest day, Thursday its weakest, and Monday quietly gaps up overnight before fading during the session.
Day-of-week effects are catnip for backtesters. The idea is simple and seductive: if one weekday has quietly outperformed for years, you would want to know. Copper is a good place to ask. It trades nearly around the clock, it is sensitive to macro headlines that often land on a schedule, and it has a long, clean history. So we took front-month COMEX copper futures (ticker HG), pulled 20 years of daily opens and closes, and asked the plain question: is there a best day to be long copper, and if so, where does that edge actually come from?
The short answer is that the spreads between weekdays are large on paper, but they are built from per-day edges so small that they sit right on top of a coin flip. That gap, between an eye-catching cumulative return and a barely-there daily win rate, is the whole story, and it is worth seeing in full before anyone treats it as a signal.
We used 20 years of front-month COMEX copper futures (HG), from June 2006 to June 2026, which works out to 5,033 trading days of daily opens and closes. For each weekday, Monday through Friday, we ran three separate strategies. Each one starts with a hypothetical $1,000 and compounds the relevant return on every instance of that weekday over the full 20 years. The three strategies slice the trading day into its two natural halves and then put them back together:
Splitting the day this way is the useful part. A weekday can look strong on a full close-to-close basis for two very different reasons: because copper drifts up while the market is open, or because it gaps up before the open. Those are not the same thing, and as we will see, copper treats Monday and Wednesday as near-opposites on exactly this point.
Here is every weekday, every strategy, as the value that a starting $1,000 grew to over the 20 years, with the cumulative return alongside.
| Weekday | Intraday | Overnight | Close-to-close |
|---|---|---|---|
| Monday | $760 (-24.0%) | $1,763 (+76.3%) | $1,340 (+34.0%) |
| Tuesday | $918 (-8.2%) | $781 (-21.9%) | $717 (-28.3%) |
| Wednesday | $1,787 (+78.7%) | $704 (-29.6%) | $1,259 (+25.9%) |
| Thursday | $435 (-56.5%) | $947 (-5.3%) | $411 (-58.9%) |
| Friday | $2,613 (+161.3%) | $1,344 (+34.4%) | $3,511 (+251.1%) |
And for reference, trading every session, every day combined: intraday returned +41.5% ($1,415), overnight +23.5% ($1,235), and close-to-close +74.7% ($1,747). Note that the all-days close-to-close figure of +74.7% is essentially buy-and-hold copper over the period, and that the intraday return (+41.5%) compounded with the overnight return (+23.5%) reproduces it. The day is just the sum of its two halves.
The most consistent pattern in the table is the two extremes, and they are the same on a full-day basis as they are during the session. On a close-to-close basis, Friday is the best weekday by a wide margin (+251.1%) and Thursday is the worst (-58.9%). The intraday session tells the same story even more starkly: Friday's daytime session alone returned +161.3% while Thursday's lost -56.5%. By that measure Friday is copper's best session and Thursday its worst, and the full-day numbers simply inherit that ranking.
If you stopped here, you would have a tidy headline: buy copper Friday, avoid it Thursday. Hold that thought. The size of these spreads is real in the arithmetic, but the per-day edge underneath them is tiny, and we will come back to why a +251% versus -59% gap can still be mostly noise.
Monday is the most interesting day in the whole study, because the two halves of it point in opposite directions. The weekend gap into Monday is the single best overnight bucket of any weekday (+76.3%). But Monday's own daytime session is one of the worst (-24.0%). In plain terms: copper tends to gap up over the weekend, then give a good chunk of it back once Monday's session opens. The full close-to-close Monday return (+34.0%) is what survives after the strong gap and the weak session partly cancel.
Wednesday is the mirror image. Its daytime session is strong (+78.7%, the best intraday bucket of any weekday), but the gap into Wednesday is the worst overnight bucket of the week (-29.6%). So Monday makes its money in the dark and loses it in the light; Wednesday does the reverse. This is exactly why splitting the day matters. On a full-day basis Monday (+34.0%) and Wednesday (+25.9%) look like fairly ordinary, similar days. Under the hood they are built from completely different pieces.
Monday and Wednesday have similar full-day returns but opposite internal structure. Monday is a strong weekend gap that fades during the session; Wednesday is a weak gap that recovers during the session. A close-to-close number alone would have hidden both.
Step back from the weekday slicing and look at the all-days decomposition, because it answers a question traders argue about constantly: does the long-run return live in the daytime session or in the overnight gap? For copper over these 20 years, the daytime session did the heavier lifting. The intraday strategy returned +41.5%, the overnight strategy +23.5%, and compounding the two reproduces the +74.7% full close-to-close return.
That is worth flagging because it runs against the most famous version of this decomposition. In U.S. equities, study after study has found that nearly all of the long-run return is earned overnight, while the daytime session contributes little or even nothing. Copper did not behave that way here. Its daytime session contributed more than its overnight gap. You should not over-read a single asset over a single 20-year window, but it is a useful reminder that the "all the return is overnight" result is an equity-market regularity, not a law of physics that every market obeys.
Two sessions dominate the tails, and they are not coincidences of seasonality, they are policy events. On a close-to-close basis, the best single day in the entire 20 years was +13.3% on July 8, 2025, when a 50% U.S. copper tariff was announced and copper posted a record one-day jump. The worst single day was -22.3% on July 31, 2025, when refined copper was exempted from that tariff, the trade unwound, and copper recorded a one-day crash. The same July 31, 2025 session was also the worst overnight move of the entire sample at -17.5%.
The yearly extremes are macro stories too. The best calendar year on a close-to-close basis was 2009 at +138.5%, the post-crisis recovery, and the worst was 2008 at -54.0%, the financial crisis. None of these record moves landed because of which weekday it was. They landed because of what was in the headlines, which is precisely the kind of event a day-of-week strategy cannot anticipate and should not get credit for.
| Extreme | When | Move | What happened |
|---|---|---|---|
| Best day | Jul 8, 2025 | +13.3% | 50% U.S. copper tariff announced |
| Worst day | Jul 31, 2025 | -22.3% | Refined copper exempted, trade unwinds |
| Worst overnight | Jul 31, 2025 | -17.5% | Same exemption, gap down into the day |
| Best year | 2009 | +138.5% | Post-crisis recovery |
| Worst year | 2008 | -54.0% | Financial crisis |
A 20-year weekday average hides a lot. If Friday's edge were a real, structural feature of how copper trades, it should show up in most months, not just a handful. So we cut every weekday return into the twelve calendar months. That turns five buckets into sixty, each holding only about 81 sessions, so individual cells should be read loosely. The table below is the close-to-close cumulative return for each month and weekday, with the full month on the right for context.
| Month | Mon | Tue | Wed | Thu | Fri | Month |
|---|---|---|---|---|---|---|
| Jan | 0% | +21% | -11% | -6% | +7% | +8% |
| Feb | +6% | +37% | +7% | +12% | +9% | +89% |
| Mar | -14% | +4% | +11% | +18% | +8% | +26% |
| Apr | -4% | 0% | +18% | +10% | +7% | +33% |
| May | +9% | -13% | -35% | -26% | +56% | -28% |
| Jun | +16% | -15% | +24% | -15% | -6% | -2% |
| Jul | -1% | +10% | +22% | -13% | +7% | +23% |
| Aug | 0% | -14% | -19% | -3% | +11% | -25% |
| Sep | -13% | -8% | +29% | -7% | -8% | -11% |
| Oct | +31% | -19% | -6% | -26% | +1% | -26% |
| Nov | +8% | -8% | -13% | -4% | +22% | +2% |
| Dec | -1% | -14% | +20% | -14% | +32% | +17% |
The tidy headline does not survive the cut. Friday's giant +251% full-period edge turns out to live in just a few months: May (+56%), December (+32%), November (+22%), and August (+11%). In June (-6%) and September (-8%) the supposedly best day of the week is actually negative. The villain is just as inconsistent. Thursday's weakness is concentrated in May (-26%) and October (-26%), but in February (+12%), March (+18%), and April (+10%) Thursday is one of copper's better days. "Buy Friday, avoid Thursday" is really "buy Friday in a few specific months, and watch it reverse in others."
May is the most polarized month on the board. It holds both the single best cell in the whole grid, Friday at +56%, and the single worst, Wednesday at -35%, even though the month as a whole is one of copper's weakest (-28%). And the "best weekday" title rotates almost at random: Tuesday owns January and February, Wednesday owns April, June, July, and September, Friday owns May, August, November, and December, and Monday owns October. No single day is reliably on top.
This is the real test, and the pattern struggles to pass it. A genuine, tradeable edge should be reasonably stable across months. Copper's weekday effect is not. Split into sixty buckets of roughly 81 days each, the strong aggregate "Friday" signal scatters into a few lucky months and reverses in others, which is exactly what noise dressed up as a pattern tends to look like.
One thing in the table is more familiar, and it runs down the rightmost column rather than across the weekdays: copper's monthly seasonality. February has been the strongest month by a wide margin (+89%), the late-winter into early-spring stretch from February through April is consistently green, and the May, August, and October cluster is consistently red. That is the old "sell in May" tendency showing up in an industrial metal. It is a calendar-month effect, not a day-of-week one, and it is a separate question from the one we set out to answer.
That rightmost column kept pointing at something larger than the weekday noise, so it is worth asking directly: forget the day of the week, what if you simply bought copper at the start of every month and sold it at the end? We took the first trading day's open and the last trading day's close of each of the 241 months in the sample, and compounded $1,000 through all of them.
| Month | $1,000 grew to | Avg / month | Up years |
|---|---|---|---|
| January | $1,045 (+4.5%) | +0.47% | 60% |
| February | $2,081 (+108.1%) | +3.87% | 70% |
| March | $1,225 (+22.5%) | +1.29% | 45% |
| April | $1,312 (+31.2%) | +1.60% | 50% |
| May | $724 (-27.6%) | -1.45% | 45% |
| June | $887 (-11.3%) | -0.38% | 57% |
| July | $1,242 (+24.2%) | +1.34% | 60% |
| August | $746 (-25.4%) | -1.36% | 40% |
| September | $891 (-10.9%) | -0.21% | 50% |
| October | $757 (-24.3%) | -0.82% | 55% |
| November | $961 (-3.9%) | +0.08% | 55% |
| December | $1,118 (+11.8%) | +0.80% | 60% |
Run every month back to back and $1,000 becomes $1,506, a +50.6% gain over the 20 years. That is actually a little less than simply buying and holding copper (the +74.7% close-to-close figure from earlier), because sitting out the turn of each month means you miss the gaps between one month's last close and the next month's open. The interesting part is not the total, it is the shape.
And the shape is more convincing than the weekday version. February stands alone: a +108.1% cumulative return, an average gain of +3.87% in the month, and a finish in the green in 70% of years. That is a genuinely lopsided record for a single month, and it lines up with a plausible story, the restocking and demand pickup around Chinese New Year and the start of the building season. The other side of the calendar is the familiar "sell in May" cluster: May (-27.6%), August (-25.4%), and October (-24.3%) are the three worst months, and August in particular finished green in only 40% of years. Spring is copper's friend; late summer and early autumn are not.
As always, the tails were macro, not seasonal. The best single month in the sample was March 2009 (+22.6%), the turn of the post-crisis recovery, and the worst was October 2008 (-36.3%), the depths of the crash. And even February, the strongest case on the board, rests on just 20 observations. A 70% hit rate is a real tilt worth knowing, but it is a tilt, not a machine, and a single bad February would move it a lot.
This is the section that matters most, so here is the honest accounting of why these numbers, despite the dramatic spreads, are not a trading signal.
The win rates are essentially a coin flip. On a close-to-close basis the daily win rates run Monday 52%, Tuesday 48%, Wednesday 50%, Thursday 48%, Friday 53%. The intraday and overnight win rates sit in roughly the same 44% to 51% band. A weekday that wins 53% of the time is not a different animal from one that wins 48% of the time. The big cumulative spreads come from those tiny per-day edges compounding over thousands of trading days, which makes the gaps look far more impressive than the underlying signal is.
With only five weekdays, a "winner" can appear by chance. When you carve a return series into five buckets and rank them, one of them is going to be the best and one the worst even if there is no real effect at all. The fact that Friday came out on top does not, by itself, tell you Friday has a durable edge. With this many comparisons and win rates this close to 50%, a meaningful part of what you see here is likely noise dressed up as a pattern.
There are no costs in these figures. Every number assumes perfect fills at the exact open and the exact close, with zero transaction costs, zero slippage, and zero commissions. Trading a position in and out every single day, as the intraday and overnight strategies do, would rack up costs that eat directly into edges this thin. The frictionless version flatters the strategy.
The "$1,000" is an illustration, not a position size. The $1,000 figures are a clean way to show compounded returns, nothing more. A single COMEX copper contract represents roughly $160,000 of notional value, so trading one contract is a heavily leveraged position, not a $1,000 bet. Do not read the dollar figures as literal account math.
The extreme days were one-off policy events. The records in both directions came from the July 2025 tariff announcement and its reversal, and the worst year came from a once-in-a-generation financial crisis. A handful of unrepeatable headlines can pull weekday averages around, and no day-of-week rule could have positioned for them in advance.
So treat all of this as a structural look at how copper's trading day is put together, where the gap up sits, where the session drift sits, and how they net out by weekday, and not as a recipe. The most defensible takeaway is the decomposition itself: copper tends to gap up over the weekend and fade on Monday, the reverse on Wednesday, and its long-run drift has come more from the daytime session than the overnight gap. The weekday "winner" is far less reliable than it looks. This is research and education, not investment advice.
Five weekdays, three strategies, twelve months, 20 years of front-month COMEX copper futures, every session counted.
On a close-to-close basis Friday is the strongest weekday (+251%) and Thursday the weakest (-59%). The intraday session ranks them the same way.
The weekend gap into Monday is the best overnight bucket (+76%), but Monday's own session is one of the worst (-24%). Wednesday is the mirror image.
Intraday +41.5% compounded with overnight +23.5% gives the +74.7% full return. Unlike equities, copper's daytime session, not the overnight gap, carried more of the drift.
Close-to-close win rates run 48% to 53%. The big spreads are tiny edges compounding, so with only five weekdays much of this is likely noise.
Slice the weekday returns by month and the edge scatters and even reverses. The steadier pattern is plain calendar seasonality: buying copper each February returned +108% over 20 years and finished green in 70% of them, while the May, August, and October stretch lagged.
What is real here, what is likely noise, and how to tell them apart.
Monday and Wednesday have similar full-day returns but opposite internal structure: one gaps up and fades, the other gaps down and recovers. A close-to-close number alone hides which is which. Always look at the overnight and intraday halves separately.
A +251% versus -59% spread looks decisive until you see the win rates sitting at 53% versus 48%. Cumulative returns compound tiny edges into dramatic-looking gaps. The daily hit rate tells you how much signal is really there, and here it is barely above a coin flip.
Copper's record up and down days both came from a single 2025 tariff episode, and its worst year from the 2008 crisis. A few unrepeatable headlines can shape a weekday average. Strip them out, or at least know they are there, before calling anything a pattern.
We test popular market ideas the honest way: every instance counted, every result measured against a plain baseline. See what else held up, and what did not.
Browse the Research
Comments
Loading comments…