The banks kept adding metal in a month when the price was already off its high, and the price kept falling after that. A central bank buys a bar for its own reserve. It does not buy your share. The five names below are types of mines and claims. They are not a list to purchase because a reserve manager in Warsaw or Beijing added tonnes.
This is not a recommendation to buy gold, a gold fund, or any mining share. It is not personal advice. It is not a gold price forecast you should treat as a target. Prices, costs, and guidance change. Read the filing. Check the tape.
The one idea is short. Official buying is a bid for bars. It is not a bid for shares. The ounces can keep arriving in vaults while the gold price decline continues. A stock worth watching for a long clock is a mine, or a claim on a mine, that can still make money after the price falls. It is not a ticker that should rise because a central bank bought gold.
Two numbers for one month
Start with the number that was reported, not the number that was estimated.
On October 6, 2026, the World Gold Council said central banks reported net purchases of 39 tonnes in August. Marissa Salim, in the Council’s note, said large buyers stayed active. China led. Uzbekistan and Poland followed. Reported buying for 2026, through August, was 170 tonnes. July, in an earlier Council note, had been 23 tonnes, and the year through July had been about 130 tonnes. August did not reverse the year. It added to it.
The Council’s August ledger, as later tables drew it from that note, looks like this. The People’s Bank of China added 20 tonnes. That was its 22nd month in a row of reported buying. China has reported 80 tonnes this year. Official holdings are about 2,387 tonnes, which the Council put at 9 percent of China’s reserves. Poland added 8 tonnes and still leads the year, at 98 tonnes. Polish holdings are 648 tonnes. The Council said a 700-tonne target is getting closer. Uzbekistan added 8 tonnes. Kazakhstan added 7. The Czech National Bank added 2 tonnes, its 42nd month in a row. Turkey added 3 tonnes in August after three months of net sales, and is still down about 82 tonnes for the year. Jordan sold 3 tonnes. Russia sold 6 tonnes in the month and is down about 56 tonnes for the year. Buyers and sellers sat in the same month. The net was still a plus.
Then a second number arrived, and it is not the same fact.
On October 19, Goldman Sachs was reported to have estimated central bank purchases of 79 tonnes in August. Newsquawk carried the line. So did later write-ups. The pre-2022 monthly average in that note was 17 tonnes. China was the largest identifiable buyer in the estimate, at 25 tonnes, against the 20 tonnes the People’s Bank reported. One write-up said Goldman’s July estimate had been 44 tonnes, and that a three-month seasonally adjusted pace was about 85 tonnes a month, or about 1,000 tonnes a year. Before 2022, the same write-up said, the pace was about 200 tonnes a year. Goldman was also reported to still call official demand a structural support for gold.
Hold the gap. Thirty-nine tonnes is what was reported to the data the Council compiles. Seventy-nine tonnes is a bank’s estimate of what may have been bought, including metal that does not show up on time. Estimates of official buying are built on lags. The International Monetary Fund numbers arrive late. A share of purchases has, for years, gone unreported. Revisions are normal. The estimate can be useful. It is not a tonne you can point to in a vault from the headline alone.
China makes the gap concrete. Twenty tonnes reported. Twenty-five tonnes estimated. Five tonnes is not a rounding error in a press release. It is the difference between a disclosure and a model. If you trade the 79, you are trading the model. If you cite the 39, you are citing the disclosure. Say which one you mean. A gold market outlook that adds them together is counting the same month twice.
Why the banks keep buying
People ask why central banks continue buying gold when the price is already far above where it was a few years ago, and why they did not stop when the price fell. The answers that fit the record are dull. Dull is useful.
They are not buying a trade. A reserve manager does not have your holding period. Poland has a published target, 700 tonnes, and it is at 648. A target is a reason to buy the next tonne even if the price had a bad week. China has reported buying for 22 months. Its gold is 9 percent of reserves. Goldman’s note, in secondary accounts, said emerging-market banks, and China in particular, still hold less gold than a long-run allocation would imply. That is a view, not a law. It does explain why a lower price is not, by itself, a reason for them to stop.
They are buying an asset that is not someone else’s promise. After Russian reserves were frozen in 2022, the pace of official buying stepped up. The pre-2022 average of 17 tonnes a month, in the Goldman line, is the old world. The new pace is the world in which a reserve can be locked. Gold in your own vault, or in a vault you trust, is a political choice as much as a price choice. That choice does not expire because spot gold had four soft sessions.
They are also moving metal, which is not the same as buying it. The Council has written about the Dutch central bank shifting about 86 tonnes out of New York and Ottawa toward London and its own books, with sales in one city and purchases in another. One later write-up of a Goldman note said gold held for foreign central banks at the New York Fed fell 40 tonnes in August, while Bank of England custody rose 46 tonnes. That custody pair is the write-up, not the Council’s 39-tonne release. A relocation is not demand. It does not lift the price by the number of tonnes on the truck. It does tell you that where the bar sits has become part of the reserve decision. Security, access, and the risk of a freeze are now in the same sentence as the ounce.
None of that is a gold price forecast. A buyer who does not care about this month’s yield can keep buying while the price falls. That is what “inelastic” means. It does not mean the price cannot fall. It means the official bid is a slow bid. It cushions a long decline more than it stops a short one. If you need the price to rise this month because China bought 20 tonnes in August, you are using a slow bid to pay for a fast trade. The clocks do not match.
The price fell anyway
The gold price decline is not a rumor. It is on the screen.
On October 21, 2026, spot quotes sat in a band around $4,100 to $4,140, depending on the hour and the screen. One service had about $4,114 during the day and called it a fourth soft session. Another had about $4,136 later in the North American afternoon. A daily series put the October 20 close near $4,125, and the October 14 close near $4,105. Desks have also cited an October 14 low near $4,073. Use the print you can see. Do not average them into a fifth number.
The high is the part people skip when they hear “central banks are buying.” One daily series put the all-time closing high at $5,405 on January 29, 2026, and put the October 20 close about 24 percent under that high. The same series had gold down about 5.5 percent for 2026 so far. A separate screen had the one-month change near minus 6 percent. This is not a crash to a forgotten price. It is a fall from a peak that was, in January, above $5,400. A fall of that size can happen while official buyers are still active. August is the proof. The banks were buying. The peak was already behind them.
The rival is the bond. Gold pays no interest. The 10-year Treasury yield has been around 5.1 percent to above 5.3 percent this month. A closing high of 5.310 percent was recorded on October 5, and the yield was 5.145 percent on October 21. The inflation-protected yield, which is the part a bar has to beat, printed 2.95 percent on October 5 and 2.77 percent on October 20. September consumer prices rose 3.7 percent on the year. Gasoline rose 9.1 percent in the month and accounted for over half of the monthly increase. Brent crude was back near $110 on October 21. A Financial Times screen showed $109.52. The October 20 settle was $108.79 after a $111.99 high. The Federal Reserve, chaired by Kevin Warsh since May, meets on October 27 and 28. A Reuters wrap of CME FedWatch on October 20 put the odds of a hike at that meeting at 17.2 percent. A hold is the heavy bet. One news report said most traders were also pricing a December hike. That December reading was not re-checked as a percentage here. Warsh has said he does not prejudge meetings. He has also described an investment surge as a near-term demand shock that puts upward pressure on inflation.
That is why the gold price forecast for the next two weeks is a yield story, not a central-bank story. The banks bought in August. The market in October is trading oil, inflation, and the chance of another hike. Official tonnes do not print a coupon. They do not vote at the Federal Open Market Committee. A long-term gold investment can live with that. A trade that needs the August tonnes to overwhelm the October yield cannot.
There is a forecast in the Goldman complex that should be labeled as a forecast, and the labels should not be mixed. An August 14 Goldman comment carried an end-2026 gold forecast of $4,900 an ounce. Later secondary write-ups of a note associated with analyst Lina Thomas have put a target at $5,400. Those are two figures. A target is a view. It is not a floor under today’s print, and it is not a reason to buy a miner. The 79-tonne estimate and a price target answer different questions. One is a guess about last month’s tonnes. One is a guess about a future price. Neither is a share.
A bar is not a share
This is the step where long-term investors get hurt. They hear that central bank gold buying continues. They see the gold price decline. They conclude that gold mining stocks are on sale, and that the sale is safe because the banks will keep the floor in place. The conclusion skips the object.
A central bank buys metal. The metal sits in a reserve. It does not pay a dividend. It does not have a pit wall, a mill, a permit, or a union. It does not issue shares. When the price falls, the bank’s new tonne is cheaper than last month’s tonne. The bank may like that. You do not own the bank’s tonne.
A miner sells metal that it has to dig up. The thing you own is the spread. The spread is the gold price minus the cost to produce an ounce, minus the capital the mine still needs, minus the country risk and the operating risk the model left out. When gold falls from $4,300 to $4,100, the spread narrows by $200 if costs do not move. The share can fall by more than the metal’s percent, because the market is pricing next year’s spread, not the bar in Warsaw. That is how falling gold prices affect mining stocks. The effect is arithmetic. It is not a vote of no confidence in central banks.
A royalty or a stream is a third object. It is a claim on someone else’s gold. It does not run the pit. It usually moves less than the wildest miner and more than the bar. It is still not the central bank’s reserve. A reserve manager is not your bid.
The August numbers show the split in time as well as in object. The reported 39 tonnes, and the estimated 79, describe July’s successor month. They were published in October, into a market that was already trading a different set of facts. By the time you read “central banks bought,” the price had its own reasons to fall. Using a two-month-old tonne to explain today’s share price is looking at the right buyer and the wrong clock.
There is also a split inside the tonne. Turkey’s August purchase did not erase a year of net sales. Russia was a seller. A net figure hides that. A headline that says “central banks are buying” is true as a net and incomplete as a list. Long-term gold investment that rests on “the banks” as if they were one desk will be surprised the first month the net shrinks because two large holders sold.
What “long-term” has to survive
A long clock is not a slogan you attach to a dip. It is a list of things that can go wrong while you wait for the official bid to matter.
The price can stay lower for longer than the reserve story. Official buying of 170 reported tonnes in eight months is large against the old 17-tonne months. It is small against the stock of gold already above ground, and it does not set the yield. If real yields stay near the highs of this cycle, the bar can drift while the banks add. Your share drifts with the bar, and then some.
Costs can rise while the price falls. That is two hits. A cost guide that was written at a higher gold price does not shrink because gold fell. Fuel, labor, and contractors do not take their cue from the People’s Bank. A mine that looked cheap on the spread at $4,500 can look ordinary at $4,100 if the cost line was optimistic.
Guidance can be cut for reasons that have nothing to do with central banks. A seismic event, a strip ratio, a mill, a snow season. The ounce in the reserve does not fix the ounce that did not come out of the pit. Buying the share because “the banks are buying” is buying the wrong dip when the miss is operational.
The balance sheet decides whether two or three weak quarters are an annoyance or a share sale. Cash and debt are not in the tonne. A company that must refinance while the gold price decline is on the screen will not be saved by Poland’s next 8 tonnes.
Jurisdiction is the part a vault does not share with a pit. The Dutch can move bars from New York to London. You cannot move a deposit from one country to another. A long-term holder of a single-asset miner is concentrated in a way no reserve manager would accept for the whole reserve. Concentration is the return. It is also the risk. Name it before you call the stock a long-term hold.
A gold investment strategy that survives this is boring. Decide whether you want the bar, the fund that holds bars, or the share. Do not buy the share as a cheap substitute for the bar and then blame the central bank when the share falls further. Size it so a move from about $4,100 toward $4,000, or through it, does not force you out. From $4,140 to $4,000 is roughly 3 percent on the metal. The share can do more. If that larger move would make you sell, the position was a trade, and a trade needs a level where the idea is wrong. “The banks are still buying” is not that level. The banks were still buying in August. The price fell after August.
Five types, not five tips
The five names are Canadian gold stocks, and they are the Canadian gold mining stocks that show up whenever a reader asks for gold stocks to watch. They are here as types. A type teaches the spread. A tip pretends the spread does not exist. Nothing in this section is a recommendation to buy or sell any of them. Share prices move. The operating figures below are from the companies’ own releases. Check the filing. The next reports for several of these companies fall in late October and early November. A new number replaces an old one.
Agnico Eagle: the wide book
Agnico Eagle is the large, multi-mine case. Its own second-quarter release put payable gold production at 855,816 ounces and all-in sustaining costs at $1,459 an ounce. The same release kept full-year 2026 production near the low end of a range of 3.3 million to 3.5 million ounces. A peer slide at the Mining Forum Americas, as written up afterward, ranked a 2026 cost-guidance midpoint near $1,475 an ounce. That $1,475 figure is a conference slide, not a line in the quarterly release. Use the release.
Why it belongs on a long-term watch list, and not on a shopping list, is the shape. More than one mine means one bad pit does not decide the company. A cost line near $1,500, against a gold price still above $4,100, is a wide spread. The spread is the point. It is also the risk in the other direction. A wide spread is what the market already knows. You are not being paid a secret for noticing that Agnico is a major. You are being paid, if you are paid, for whether the cost line stays near that level while the gold price decline works through the next two quarters.
Central bank gold purchases do not operate Agnico’s mills. If the banks add another 20 tonnes and the 10-year yield makes a new high, Agnico’s share can still fall. The long-term question is whether the spread stays wide enough that the fall is a price, not a threat to the business. That is a filing question. It is not an August-tonne question.
Barrick: the book that still has to prove the streak
Barrick Mining Corporation is the other large book, with a different recent argument. Its second-quarter release put gold production at 796,000 ounces and all-in sustaining costs at $1,866 an ounce. The same release left 2026 production guidance at 2.90 million to 3.25 million ounces, and all-in sustaining costs at $1,760 to $1,950 an ounce. The midpoint of that cost range is $1,855. Mark Hill, president and chief executive on the second-quarter call, said production and cost guidance were unchanged.
A guide that is unchanged is a start. It is not a reputation. A gold price decline tests whether the guide was built with room, or built to the old price. Barrick’s own cost guide for 2026 assumes a gold price of $4,500 an ounce. The screen is no longer there. The spread at a realized price above $4,000 is still wide if the $1,866 line holds. Wide is not the same as safe. Barrick’s assets sit in more than one country. Country risk does not average out just because the company is big. A long-term holder is underwriting those countries, not underwriting China’s reserve policy.
The watch item is the next print, not the bank bid. If the ounce holds and costs do not jump, the spread story holds. If the print arrives with a softer ounce and a higher cost, the share is reacting to Barrick, not to the 39 tonnes. Do not comfort yourself with Poland’s target while you wait for the release.
Kinross: a cut is a fact
Kinross is the case that keeps the list honest. Its second-quarter release put attributable production at 492,326 gold-equivalent ounces. Attributable all-in sustaining cost was $1,821 per equivalent ounce sold. The cost is per ounce sold. The 492,326 is ounces produced. They are not the same unit. That quarterly line is not the story. The story is the guide.
At the second-quarter report, the 2026 guide was 2.0 million gold-equivalent ounces, plus or minus 5 percent, at an all-in sustaining cost of $1,730 an ounce, plus or minus 5 percent. On September 23 the company changed that. It set 2026 and 2027 attributable production at about 1.84 million to 1.86 million gold-equivalent ounces a year. It said the new band was 2 to 3 percent below the low end of the old range. It raised 2026 all-in sustaining cost to about $1,850 to $1,900 per equivalent ounce sold. The company tied the change to two smaller mines. At La Coipa, in Chile, winter weather cut mining rates and mill throughput, and some sulphide ore carried higher copper grades and lower recoveries. At Round Mountain, in Nevada, mining performance was weaker. The notice puts 2027 in the same 1.84 million to 1.86 million band as 2026. It does not schedule a return to 2.0 million ounces.
A smaller guide for two years is a smaller business than the old one. A higher cost line against a lower gold price is the double hit. How falling gold prices affect mining stocks is easiest to see here. The banks did not cause Round Mountain. The banks will not fix La Coipa. A long-term investor who buys Kinross because central bank gold buying continues is mixing a reserve policy with a mine plan. The mine plan is the thing that changed.
The watch item is whether 1.84 million to 1.86 million ounces stays the band. The company has already put 2027 there. If that band holds and the cost guide stops rising, the spread can heal even if gold does not revisit a bank’s target. If the band is cut again, the old 2.0 million-ounce figure, plus or minus 5 percent, was a hope. Hopes are not tonnes.
Alamos: one event, a smaller year
Alamos Gold is the mid-size case, and it is a reminder that rock does not read a reserve report. Its July 29 release put second-quarter production at 130,600 ounces and all-in sustaining costs at $1,728 an ounce. The same release cut 2026 production from 570,000 to 650,000 ounces to 510,000 to 560,000 ounces. On the earnings call the company called that a 12 percent cut. It raised the 2026 all-in sustaining cost guide from $1,500 to $1,600 an ounce to $1,775 to $1,875. The midpoint of those ranges rises by about 18 percent. Two things drove the cut. A seismic event in June at Young-Davidson limited access to higher-grade stopes. The company expects mining rates there to average about 5,000 tonnes a day in the second half of 2026. Slower recoveries at La Yaqui Grande delayed ounces already stacked on the leach pad. Chief executive John McCluskey said lower production from Young-Davidson was the primary driver. The release also says production is expected to reach about one million ounces a year by 2030. It does not set a climb back through 7,000 tonnes a day, and it does not print a 2028 cost of $1,200 to $1,300.
The later number is a company path. A path is not a reserve, and it is not a result. A seismic event and a slow leach pad already changed 2026. Another event can change the path to 2030. The spread today is still wide if gold holds above $4,000 and costs stay inside the new guide. The new guide is higher than the old one. That is the fact a long-term buyer has to underwrite. You are not buying the guide that was cut. You are buying the guide that replaced it, plus the time between now and 2030.
Official sector demand does not reopen the 9410 level at Young-Davidson, and it does not speed a leach cycle. If you want Alamos on a watch list, watch the tonnes per day and the cost line, not the next Chinese disclosure. A central bank cannot make a seismic event unhappen.
Lundin Gold: one mine, a wide spread, one country
Lundin Gold is the cleanest illustration of concentration. Its second-quarter report put production at 118,994 ounces and all-in sustaining costs at $1,176 per ounce sold. The cost is per ounce sold. The 118,994 is ounces produced. Ounces sold in the quarter were 110,385. That cost line is the lowest of these five. Against a gold price above $4,100, the spread is very wide. Fruta del Norte, in Ecuador, is the mine. In substance, the company is that mine.
A wide spread on one mine is a wonderful quarter and a single point of failure. The Dutch central bank can split bars across cities. Lundin cannot split Fruta del Norte across countries. Politics, a mill, a community agreement, a power line: any one of them is a larger share of the company than the same event would be inside Agnico or Barrick. The return for accepting that is the cost line. The cost of accepting that is the lack of a second mine.
This is the name on the list that most tempts a false link to central banks. Low cost, high gold price, official buyers still active: it sounds like a story with no villain. The villain, if it comes, will not be the gold price alone. It will be the single asset. A long-term watch means you can say, in one sentence, what happens to the equity if Ecuador is the problem and the gold price is not. If you cannot say it, you do not own a long-term position. You own a hope that the one mine stays a good mine.
What these five do not tell you
They do not tell you that large is always safer. Barrick’s cost line is higher than Agnico’s on the quarter just cited. Size did not put them on the same line.
They do not tell you that a cut is fatal. Kinross and Alamos cut guides and still, at a gold price above $4,000, have spreads that would have looked extreme in any year before this cycle. A cut changes the slope. It does not, by itself, erase the business. It does disqualify the old slide from your model.
They do not tell you that the lowest cost is the best stock. Lundin’s $1,176 is a real advantage and a real concentration. Advantage and concentration are the same mine.
They do not include a royalty. Franco-Nevada and Wheaton Precious Metals are the other type a long-term reader should know exists. They are claims, not operators. They are not in the five because the lesson this month is the operator’s spread, and a claim hides that lesson. If your aim is a long clock with less pit risk, a royalty is a different watch list. It is still not a central bank reserve. It still falls when the gold price decline is steep, usually by less than the wildest miner.
They do not include a developer with no cash flow. A developer does not have a spread yet. It has a model, a capital need, and a permit clock. Official buying can lift the metal the model uses. It does not write the check for the mill. In a falling price, the developer’s equity is a bet that the model survives the new price. That is a reasonable bet only after you have read the model. It is not a reasonable bet because China bought 20 tonnes.
How to watch without turning it into a trade
Pick the object first. If you want what the central banks are buying, you want a bar or a fund that holds bars. The 39 tonnes and the 79-tonne estimate are arguments about that object. They are slow arguments. They do not require you to own Kinross.
If you want the spread, pick one type and write the number that would make you wrong. For Agnico, the number is the cost line, and whether production stays near the low end of 3.3 million to 3.5 million ounces. For Barrick, it is whether the unchanged guide survives a gold price below the $4,500 the cost guide assumed. For Kinross, it is whether 1.84 million to 1.86 million ounces, already the company’s band for 2026 and 2027, holds. For Alamos, it is tonnes per day in the second half, the leach pad, and whether the path to about one million ounces by 2030 is still the path. For Lundin, it is the single mine. One sentence each. If the sentence is “the banks are buying,” you have not picked a type. You have picked a headline.
Separate the two August figures in anything you write down. Reported net purchases, 39 tonnes. Estimated purchases, 79 tonnes. China’s reported 20 against an estimated 25. Year to date reported, 170 tonnes. Poland 98 of those, with 648 tonnes in hand and a 700-tonne target. If a later revision moves the 79, your sentence should move. If you built a portfolio on the 79 as if it were a customs form, you built it on a model.
Do not use a price target as a substitute for the sentence. An August Goldman comment put end-2026 gold at $4,900. Later write-ups have put $5,400 on a different note. Either figure is someone’s gold price forecast. It can sit next to a market that is trading $4,100 and an October meeting where a hike was priced at about 17 percent. Both can be printed on the same day. The long-term holder who needs a bank’s target to justify the purchase has a trade with a long label. The long-term holder who can live with $4,000 because the spread still works has a position. Know which one you are.
Revisit the bid that is actually yours. Central bank gold purchases will be updated again, late, with another gap between the reported net and somebody’s estimate. The gap is the story of the data. It is not a signal to add to a miner the morning the gap is published. The morning the gap is published, look at the yield. If the yield is the thing that moved the metal, the miner moved because of the yield. The tonne explains the floor under a long chart. It does not explain the red day.
The mistake the headline invites
The headline is true and easy to misuse. Gold prices fell. Central bank buying continued. Therefore, five stocks. The “therefore” is the mistake.
The banks continued because their reasons are slow. Targets, reserve shares, and the memory of frozen assets do not turn off when Brent is near $110 and the 10-year yield is above 5 percent. The price fell because a bar pays nothing and the coupon rose. Both facts fit in one month. August’s buyers and October’s sellers are not in a debate. They are on different clocks.
A long-term investor can use the slow clock. The slow clock says the official sector has been a net buyer, that the reported year is 170 tonnes through August, and that the estimated month was larger than the reported month. That is a reason the metal has a buyer who is not you. It is not a reason your share is cheap. Cheap is the spread after the fall, measured against the new cost guide, the balance sheet, and the number of mines. Agnico, Barrick, Kinross, Alamos, and Lundin Gold fail or pass that test one filing at a time. They do not pass it as a group because a reserve manager added weight to a vault.
If you need one line, keep this. A central bank buys a bar. You do not own that bar when you own the share. The price can fall while the bars keep moving. Watch the mine that still works after the fall. Do not watch the tonne and pretend it was a bid for you.