OSRS Investing With a Big Bank: 8 Ideas We Tested

The question

Our Trader puts at most 3M GP into one item, or 5M for plans of 50M or more. Players with big banks ask a fair question: shouldn’t more GP earn more? A big bank could buy expensive items, hold through price moves or buy in bulk.

Our earlier study found that a bigger bank barely helped flips that last a few hours [1]. So this time we looked at everything else we could think of. We tested items worth up to 1.6B each, holds of up to a year, update news, trained models and ten years of prices.

Here “safe” means a rule had to make money on average, with its 95% range above zero, on most days, with no very bad days. Each study wrote down its own exact test.

Every idea we tested

We tested 8 ideas, and a ninth is still running. To keep ourselves honest, we built each idea on older months of prices. If it made money there, we wrote the rule down and tested it once on newer months we had kept aside, so it had never seen them. If it lost money even on the months we built it on, we stopped, because there was nothing left to test.

Big items: lots of room, no margin

There is plenty of room at the top of the GE. If you took a tenth of normal trading, items over 10M could take about 170B GP a day. The Twisted bow alone could take about 22B.

The tax is also kind to big items. It stops at 5M per item, so a Twisted bow at about 1.59B pays only 0.31% tax, not 2%.

But other traders got there first. On all 25 regularly traded items over 100M, the typical hourly gap between buy and sell prices was within 0.66 points below to 0.11 points above the tax. Even a perfect flip, buying at an hour’s low and selling at the high 1 or 4 hours later, made between −0.72% and +0.04% after tax.

Next we tried patient offers every day: buy at yesterday’s typical low price and sell a few days later at its typical high price. They lost money at every price level with room. The only items that made money were cheap ones that pay no tax.

Cheap tax-free items made about +1.3–1.5% a round, and 81–83% of days made money. That shows the test can find profit where a margin exists. The catch is room. GE buy limits cap each item, so all 233 of them together could take only about 84M GP a day. Ancient essence alone was about a third of that.

Bonds are the one big item that pays no tax. But selling a bond on the GE makes it untradeable, and making it tradeable again costs 10% of its price [2].

What this means for you: On items over 100M, the gap between buy and sell prices is about the size of the tax. Flipping them is close to a coin toss that costs a little each time.

Game updates: the move comes before you can act

Game updates do move prices. Items an update post named made an unusual move in the following week 17% of the time, against 11% for other items. Items that were only similar to them didn’t move more than usual.

But most moves can’t be caught in time. Few started before the update came out, and early moves tended to reverse.

Other items did the same, so a big first move is mostly noise that settles back. Either way, buying an early rise lost money.

We built a buying rule on 11 updates and then ran it once on the next 10. It lost 15.4% a trade (95% range −23.7% to −7.5%). All 18 versions we tried lost money on average. A trip in and out cost about 4% for these items, and only about 1 trade in 5 moved more than that.

The Masori body shows why a single story misleads. Its price rose almost 30% in the three days after The Fractured Archive came out on 20 October 2026 [3]. But of 171 items of the same combat styles, 149 fell over those three days. The middle item in the rest of the market fell 5.9%.

What this means for you: Don’t chase a price that jumps on update day. On average, it gave the gain back within 3 days.

Update news: reading the post first didn’t help

If prices move too fast, maybe reading the post first is the edge. We tested that with 20 updates. A separate AI helper read each post with no prices and labeled each item: more wanted, less wanted, new or unclear. We saved its labels before loading any price.

On the 14 updates we kept aside, the labels called the price move right 47% of the time over 7 days. Buying the items labeled “more wanted” lost 15.1% of the 1.10B GP bought (95% range −19.1% to −6.9%). Only 1 of 9 updates made money.

By the time a player can act, the post’s meaning is already in the price. Or the market does something the post doesn’t predict, such as buffed items falling.

Trained models and setting your own prices

Next we trained three kinds of models to predict how items worth 10M or more would move over 1, 3 and 7 days. They used past moves, volume, the buy–sell gap, each item’s past results, the whole market, the day of the week and the item’s yearly range.

The model’s picks lost less than buying everything, so it learned something. But it still lost 2.2%, and only 14% of days made money. None of 114 settings passed on the build months.

The signal is small. From the price a player can actually see, the tenth of items the models liked best rose only about 0.1–0.6%. A trip in and out costs about 1–4%.

Patient offers also fill on the wrong days. A buy offer below the market fills when the price is falling. On the build months, offers that never filled would have made +1.6% on paper, and the ones that filled made −1.2%.

A follow-up set its own buy and sell prices on big items, but only when the gap looked wider than the tax. All 412 settings lost money, even on the months they were built on. Even if every trade had filled at its price, which no real offer can, it lost 0.4–1.9%.

What this means for you: If a tool says it can predict big-item prices, ask how it did on months it never saw, after tax and with real fills.

Ten years of calendar events and new items

Some events come once a year, so 2026 alone can’t test them. We used ten years of daily guide prices from Weirdgloop [4] to test 19 calendar rules: Leagues and Deadman seasons, holiday events, gear releases, the month or day, and three-year lows and highs. We built and tested them on 2017 to 2022.

A trip in and out costs about 5% here: 2% tax and about 3% between buy and sell prices. None of the 19 rules made money after that. Most lost 1–11% each time.

Items at a three-year low did rise about 5–9% before costs, once we left out items whose price never moves. But nearly all of that came from one burst in 2021 and 2022, when godswords and Armadyl gear shot up. In other years they went nowhere.

New expensive items lost the most. A year after release, the typical new item over 10M was worth 23% less. After two years it was worth 37% less.

Every rule that bought new items after release lost money. In the worst cases, a rule lost 47% to 78%.

What this means for you: Big prices don’t move on a schedule you can trade, and there is no safe bottom to buy new expensive items.

What did work: the game’s fixed-rate swaps

The game has fixed swaps. You buy the inputs, change them in a fixed way and sell the result. Tanning hides, sawing planks, combining potion doses, high alchemy and selling to NPC shops all work like this. The profit is set when you buy, as long as prices hold until you sell.

We built 15,136 swaps from the OSRS Wiki [5] and saved them before loading any price. We kept the ones that made money from January to June. Then we checked them once on July to October, for one account with 8 GE slots and 2 hours of play a day.

Every registered test passed, and the months kept aside did a little better than the months we built on. The swaps we could check on market prices made about 1.18M GP a day, or about 600K for each hour of play. Swaps paid in coins by the game made about 0.89M a day, and none of them lost GP. Of 169 swaps that buy and sell on the GE, 164 made money.

The biggest earners were tanning blue dragonhide (about 466K a day), sawing mahogany planks (241K) and combining super potions. For coins, it was selling rune javelins and rune warhammers to NPC shops.

But a big bank adds nothing here. The account bought only 7–21M GP of items a day, with 54–84M in use at the most. A 50M bank made as much as 1B. The limits are play time, GE buy limits and how much the GE trades.

This passed a test on months we kept aside. It is not yet proven on live prices. A live paper test runs the same frozen rules on prices from 27 October to 9 November 2026. It is scored once on 14 November, and we’ll update this page either way.

What this means for you: If you want more GP from a big bank, the bank isn’t the limit. Fixed-rate swaps paid in our test, but they pay for the hours you play, not the GP you hold.

Still running: low offers to catch panic sales

One last idea is being tested now. You leave buy offers well below the market on expensive items, so they only fill when someone sells in a panic or by mistake. Then you sell once the price comes back.

The five-minute prices this test needs are still downloading. The plan will be saved before the months kept aside are opened, like the others. We’ll add the result here when it’s done.

What this means for you

• A bigger bank won’t raise your profit by itself. In every test, the limit was the margin, the GE buy limit or your play time.
• On items over 100M, the buy–sell gap is about the size of the tax. Check the margin after tax before you flip one.
• Don’t chase update-day jumps. Early rises gave the gain back within 3 days on average.
• Be wary of new expensive items as a way to make money. They lost about a quarter of their value in a year, on average.
• If you want more GP an hour, fixed-rate swaps paid in our test while you play. Margins change, so check today’s prices first. The OSRS Wiki’s money-making guide lists many of them with live profits.
• For flips, the free Market ranks trades by expected profit after tax and shows how often trades like each one made money.

Limits

• Fills are estimates. We see hourly or daily averages, not each trade. We let an offer take at most a tenth of the trading at its price. Real offers wait in line behind others, so real fills are likely worse. That would only make the losing ideas lose more.
• We didn’t model our own effect on prices. Big offers on thin items would move the price against you, which would also make results worse.
• Some sets are small. The update studies had 10 and 14 updates kept aside, and some calendar rules had only 4 events. “No” means we found no reliable edge, not proof that none exists.
• The swap results are for one account. If many players use the same swaps, they share the same GE buy limits and trading, and each earns less.
• Shop requirements, such as quests and diaries, aren’t in the swap table. Banking and travel to some shops may take longer than we allowed.
• Months overlap between studies. Each study kept its own months aside for its own question, but the program as a whole used most of 2026.
• These are results from past prices, not a promise of future profit.

How we tested

We tried hard not to fool ourselves. With this many ideas, a few will look good by luck, so each study followed the same rules.

• Plan first. Five of the six studies wrote down their data, tests and what counts as a pass before the final check, and three did it before looking at any price. We saved a fingerprint (a hash) of each plan, so it can’t be changed quietly.
• Months kept aside. Each idea was built on one stretch of months and checked once on months it had never seen. If an idea failed on the build months, the kept months stayed closed.
• One look. Each final check ran once, and we logged the time before running it.
• A skeptic. In three studies, a separate reviewer tried to break the method before the final check. They found bugs, a small look-ahead and tests that couldn’t fail, and we fixed them first.
• Every try counts. The study files list every rule and setting we tried, failures included.

Download the data

The tables behind this article, as CSV files. They hold our results, not raw prices. The prices themselves are free from the OSRS Wiki and Weirdgloop.

• Every idea testedEach idea, its data, whether its plan was saved first, how many settings we scored and the result.
• Big-item gaps and the taxThe 25 most-traded items over 100M: typical gap, tax and what a perfect flip made.
• Patient offers by item priceEvery daily and weekly patient-offer setting, with returns and room.
• Game update rulesAll 18 buying rules after updates, on the build and kept-aside months.
• Update news by updateWhat buying the “more wanted” items made after each update kept aside.
• Trained model settingsThe registered model and every other setting we described on months kept aside.
• Calendar and new-item rulesEvery calendar and new-item rule over ten years, before and after costs.
• New items after releaseHow the typical new item over 10M moved from a week to two years.
• Fixed-rate swapsProfit a day and per play hour for each swap portfolio, bank and play time.
• Swaps by kindHow many swaps of each kind were tried and made money in the months kept aside.

Sources

1. Gielinor Gains, “OSRS Flipping After the 2% Tax: What It Really Earns”, 10 October 2026. What flipping earns by bank size.
2. OSRS Wiki, “Old school bond”. Any trade makes a bond untradeable, and making it tradeable again costs 10% of its price.
3. OSRS Wiki, “Real-time Prices”. The prices and trading volume we used for 2026, which come from RuneLite.
4. Weirdgloop, price history API. Daily official guide prices and volume for OSRS items, which we used for the ten-year tests.
5. OSRS Wiki, item, shop and skill pages, and its money-making guide. The rules, fees and shop data behind every swap.

Words we use

Bank

The GP you have to trade with.

K, M and B

Thousand, million and billion. 600K GP is 600,000 GP, and 1.6B GP is 1,600,000,000 GP.

GE tax

The share of each sale the Grand Exchange keeps: 2%, rounded down, up to 5M GP per item. Sales under 50 GP and a short list of low-level items pay nothing.

Buy–sell gap (spread)

How much higher the instant-buy price is than the instant-sell price. A flip has to beat the tax with this gap.

GE buy limit

The most of one item the Grand Exchange lets you buy every 4 hours.

Fill

When your offer actually buys or sells. An offer that never fills earns nothing.

Fixed-rate swap

Turning items into other items at a rate the game sets, such as tanning hides, sawing planks or high alchemy.

Return

Profit after tax as a share of the GP bought. Spending 1M and getting back 1.02M is a +2% return.

Build months and months kept aside

We built and tuned each idea on one stretch of months. We kept other months aside and used them once, to check the idea on prices it had never seen.

Pre-registration

Writing down the plan and what counts as a pass before looking at the results, so the test can’t be bent to fit them.

Median (middle, typical)

The middle value: half the results are above it and half are below.

95% range

How far a number could move by chance. We worked out each number again thousands of times on random mixes of the days (or updates) we measured, and 95 of every 100 results landed inside the range.