OSRS Flipping After the 2% Tax: What It Really Earns

Period. 280 rounds, one every 2 hours from 15 September 12:00 to 8 October 18:00 UTC (24 days). Each round keeps its offers for 4 hours. Rounds overlap, so they can’t be added up into a day’s income.

Offers. For every item with a GE buy limit, buy at the middle (median) instant-sell price of the last hour and sell at the middle instant-buy price. An offer only counts if it beats the tax, the margin is under 15%, and both sides traded in at least 6 of the last 48 five-minute windows.

Picking with hindsight. Knowing how every offer turned out, we found the best 8 positions that fit in the bank. A position can be made smaller to fit. We solved this exactly, checked the method against brute force on 300 test cases, and confirmed it never did worse than a simple rule in 84,096 checks.

A correction. Our first version took the biggest profits first and shrank whatever didn’t fit. That shortcut isn’t the best choice once the bank runs out, so it understated the ceiling: 231K instead of 397K with a 10M bank, for the same offers.

One item supplies a big share. Ancient essence sells for about 18 GP, pays no tax because it sells for under 50 GP, and has a buy limit of 300,000. It supplied 7–30% of the ceiling, the most for small banks. Without it, the ceiling was 322K–488K with 10M and 0.71–1.33M with 200M. Our fill model is probably most generous for items like this, where many players compete at the same price.

The market paid more later on. With 10M and offers sized to recent trading, the ceiling was 302K in the first half (15–26 September) and 489K in the second (27 September to 8 October). On 9 October, a day we checked afterwards, it was 539K. Each part told the same story: well under 1M every 4 hours with 10M.

More fills, higher ceiling. We assume an offer gets at most a quarter (25%) of the trading at its price. With full GE limits, getting half would raise the ceiling 1.27–1.68 times, and getting all of it 1.57–2.57 times. With offers sized to recent trading, it barely matters (1.09–1.27 times, even at 100%).

Typical rounds. With full GE limits, the middle round made 541K with 10M and 1.29M with 200M. One round in 10 made 336K or less with 10M.

Method. We replayed Trader’s own choices for every round, using only prices it could have seen at the time, and scored them with the same fill model as the ceiling. Build period: 280 rounds, 15 September to 8 October 2026.

Test month. Before running it, we wrote down what would count as a pass. Then we ran it once on 17 August to 14 September 2026, a month not used to build it, and it passed. The figures below are copied exactly from our test log.

After the test. A code review then made three small fixes, such as taking turns between equally good picks. On the build period they changed profit by less than 1K per round. We did not run the test month again. The build-period figures above are for the code Trader runs today.

Build period in detail. With 10M, 94% of rounds made money, the middle round made 136K, one round in 10 made 27K or less, and the worst round lost 451K. With 50M and 200M, 94% of rounds made money and the worst lost 420K.

What the track record added. Before track records, Trader made 103K, 118K and 117K with 10M, 50M and 200M. The track record added 34.7K (95% range 24.5K to 44.6K), 29.8K (19.3K to 39.7K) and 30.2K (20.3K to 39.3K) per round. Both halves of the build period show a gain.

Why a bigger bank didn’t help. Every strategy we tested made about the same from 50M to 200M. They found only enough good trades to use about 7–35M, even with 1B available. With perfect hindsight there was more to earn, but nothing we tried could pick those trades in advance.

A version that looked better on paper. Using the same track-record idea on every item’s simple offers made 201K per round with 10M and 236K with 50M or more. But 60% and 51% of that profit came from Ancient essence, an 18 GP item that pays no tax. Without it, the results were 114K (95% range 96K to 132K) and 126K (104K to 149K). It was built and measured on the same days and never tested on the test month, so we don’t count it.

Our first version. It quoted 120–155K from an offline re-selection of Trader’s suggestions. That re-selection gives the same results here (123K–156K), but the figures above replay the code Trader actually runs.

Break-even math. Profit per item = sell price − tax − buy price. The tax is 2% of the sell price, rounded down, and never more than 5M. So the sell price must be 2.04% above the buy price just to break even, against 1.01% under the old 1% tax. Items under 50 GP pay no tax, and once a sale reaches 250M the tax stays at 5M.

Tax-free items. Bonds pay no GE tax but carry a fee to make them tradeable [1]; like our site’s calculator, we charged 10% on every bond sale. The game also doesn’t tax a short list of low-level items, such as energy potions, Mind runes, some foods, teleport tablets and some tools [2][3]. This study taxed those items anyway. Many sell for under 50 GP and pay nothing either way. For the rest, our profits are slightly too low.

Same offers, three taxes. With a 200M bank and perfect hindsight, offers sized to recent trading made 4.66M (95% range 4.53M to 4.79M) every 4 hours with no tax, 2.67M (2.55M to 2.80M) at 1% and 0.80M (0.69M to 0.93M) at 2%. With full GE limits: 4.85M, 3.00M and 1.43M.

The older taxes probably look too good. We held today’s prices fixed. Under a lower tax, players would likely have competed margins down, so the no-tax and 1% figures probably run high. The direction held in both halves of the period and on 9 October: with 200M and offers sized to recent trading, a 2% tax left 24%, 35% and 35% of the 1% figure. With full GE limits it left about half or less (44%, 50% and 46%).

Dates. The dates of the tax changes come from Jagex’s update posts [1][2], not from our data, which only covers 2026.

Margin details. Counting every 5-minute window the same, 88%, 72% and 57% beat no tax, 1% and 2%. The typical unweighted margin was 2.5%, because many cheap items move in 1 GP steps. Stricter filters (at least 5 or 25 units each side, or 1M GP each side) gave 31–43% beating 2%, against 57–68% beating 1%. Both halves and 9 October agree: 41%, 45% and 44% of GP traded beat 2%. These are 5-minute averages of each side, not the exact prices at one moment.

Bigger positions. We replayed Trader from before its current position limits with a cap per item from 1M to 20M (280 rounds). With 50M: a 1M cap made 69K per round, 2M made 106K (95% range 89K to 124K), 3M 113K, 5M 108K, 10M 101K and 20M 90K. With both 50M and 200M, the 20M cap made 16K less than the 2M cap (range 4K to 30K less). Grouped by planned size: under 100K returned +12.1%, 100K–1M +2.2%, 1–5M +0.5%, 5–20M −0.2% (range −0.4% to −0.1%) and over 20M −0.4% (−0.7% to −0.2%). Of 148 positions over 20M, only 16% sold out within 4 hours, and 41% were still holding unsold stock. Our first version said 60K–140K; that range mixed three separate stretches of time.

Buying low and selling high on everything. Offers at the last 4 hours’ middle prices with full GE limits, held 4 hours, lost 0.78% to 2.03% across eight price groups, every range below zero (overall −1.47%, range −1.53% to −1.41%). Held 24 hours, losses reached 2.68%. When a buy filled, the price was lower 4 hours later 49% of the time (58% for items over 1M), against 30% when it didn’t fill.

Expensive items. Ten simple strategies (4- and 24-hour holds, several price rules and sizes) lost 0.69% to 1.61% on items over 1M, every range below zero. Valuing unsold stock at market price instead of selling it, they lost 0.03% to 0.45%. Trader’s own suggestions: the older Trader lost 0.32% (range −0.51% to −0.18%); today’s Trader made +0.09% (−0.01% to +0.18%), about break-even.

Set arbitrage. On hourly quotes, buying the pieces and selling the set beat the tax in 54% (Torva) to 99% of hours: Justiciar in 89% of hours (typically 434K), Ancestral 88% (1.49M) and Inquisitor’s 77% (2.06M). With real fills over 4 hours, the expensive sets lost 0.48% to 1.19%, every range below zero. Cheap sets per round: Obsidian 35K (range 11K to 63K), dwarf multicannon 27K (13K to 43K), Barrows sets 12K–21K except Dharok’s −4K (−21K to 14K), Eclipse Moon 26K (−8K to 59K), Blue Moon 12K (−11K to 34K) and Blood Moon −77K (−124K to −30K). Sunfire fanatic also lost money: −23K (−44K to −3K). Held 8 hours, the best cheap sets reached about 59K–65K.

Dip buying. Across every item, with no limit on GP, over 22.4 days: buying 10% drops made 0.42M a day (range −1.05M to +1.38M, 248 drops a day) and 15% drops 1.20M a day (+0.39M to +2.67M). Requiring at least 10 units sold into the drop: 0.64M (−0.31M to +1.36M) and −0.03M (−1.43M to +0.81M). The profit came from items under 100K; 10% dips in 100K–1M items lost 0.72M a day. Within 24 hours, prices got back to 98% of the day before for 95% of drops under 10K (5,303 drops), 90% at 10K–100K (132), 72% at 100K–1M (82) and 52% at 1M–10M (21). We fixed a small error in our first script, which could sell units before buying them; it changed results by under 700 GP a day.

Where these come from. Every item Trader could suggest in each round, at Trader’s own prices and quantities, scored with the same fill model, whether or not Trader picked it. An item’s record is the return of all its earlier suggestions whose results were known, at least 6 hours old.

How strong the link is. The rank correlation between an item’s record and its next return was 0.60 (95% range 0.57 to 0.62). When we shuffled records between items in the same round 500 times, the link all but vanished every time. The top group beat the bottom group on 23 of 23 days. From 0–0.5% upward, each group beat the one below even at the low end of its range; the bottom two groups overlap.

Beyond the margin shown up front. The planned margin is an even stronger guide on its own (rank correlation 0.72 with the next return). With the margin held fixed, the record still adds something (0.15, range 0.13 to 0.16). Split into five groups by planned margin, items with a record above 1% beat the rest in every group, by 0.4 to 7.3 percentage points.

Halves and price. First half, from lowest group to highest: +0.34%, +0.33%, +0.56%, +1.10%, +2.17%, +7.50%. Second half: +0.23%, +0.29%, +0.56%, +1.36%, +2.76%, +8.81%. Items under 10K: +0.43% for the lowest group against +9.2% for the highest; 10K–1M: +0.31% against +6.46%. Items over 1M almost never had a record above 2%. On 9 October, the same idea applied to every item showed the same pattern.

Depth. A record rested on a median of 21 earlier suggestions (one in 10 had 4 or fewer, one in 10 had 62 or more).

Memory. Picking only items with a record above 1%, all of history beat the last 24 hours by 20K per round with 10M (95% range 9K to 33K) and 26K with 50M (15K to 37K). The 24-hour version kept at most 46% of the gain. With 24 hours of history, 61% of suggestions had a record; with all of it, 91%.

What Trader uses. Trader’s records count recent results more (older results fade by half every 7 days). That version gives almost the same table: +0.29% for the lowest group up to +8.37% for the highest.

Data. The OSRS Wiki real-time prices API, in 5-minute steps [4]. For each item and 5 minutes: the average price and number traded for instant buys, and the same for instant sells. 6,768 five-minute windows from 15 September 12:00 to 8 October 23:55 UTC. 4,479 items traded, 4,033 of them with a GE buy limit. 24 windows had no data for any item, during a feed gap on 29 and 30 September.

Rounds. A decision every 2 hours from 15 September 12:00 to 8 October 18:00 UTC: 280 rounds over 24 days. Offers stay open 4 hours. Anything unsold is sold at the next instant-sell price within 2 hours, minus tax.

Fills. A buy offer fills in a later 5-minute window only if instant sells averaged at or below its price, and then for at most a share of what traded. Sell offers work the same way against instant buys. It is the same fixed model we use to test every change to Trader.

Fill share. 25% unless we say otherwise. We also tried 10%, 50% and 100%.

Tax. 2% of the sell price, rounded down, up to 5M per item, and nothing under 50 GP. Bonds were charged a 10% fee instead, as on our site. Other tax-free items were taxed.

95% ranges. We re-drew the 24 days at random, with repeats, and worked out every number again each time: 5,000 times for most numbers, and 1,000 or 3,000 for a few slower checks. The middle 95% of the results is the range. We re-draw whole days because rounds on the same day overlap and share the same market. The track-record table also shows ranges from re-drawing items instead of days.

A month kept aside. Trader’s track-record rule was tested once on 17 August to 14 September 2026, after we wrote down what would count as a pass. This article never looks at that month again. Our first version started its build period on 14 September at 12:00 by mistake, which overlapped the last six decisions of the test month. This version starts on 15 September at 12:00.

Checks. We ran everything twice from the raw files, and all 56,725 numbers matched. Our data and simulation match the first version’s own files, apart from tiny rounding differences on items priced over 16.7M GP.