Module 4 · Online worlds (1997–2005)

Virtual economies

Give thousands of people one world and the right to trade — and you get a real economy with a measurable GDP, inflation and a labor market. Bartle described who those people are (and that their population is an ecosystem you can easily collapse). Castronova proved the economy is real. EVE showed the rest: either you design the currency and the sinks yourself, or emergent dynamics will design your game for you.

~18 min💰 economics + community

The gist in 30 seconds

If a game lets value move between players, you already have an economy, whether you want one or not. Bartle (1996) derived 4 player archetypes from two axes (acting↔interacting × world↔players): Achievers ♦, Explorers ♠, Socializers ♥, Killers ♣ — and showed that this is an ecosystem: tilt the balance (open PvP → "killer heaven") and the Socializers leave, and the population collapses (UO → the Trammel/Felucca split, 2000). Castronova (2001) measured the EverQuest economy through the RMT exchange rate (eBay): platinum ≈ 1 cent, labor $3.42/hour, GDP per capita $2266 — above India, Bulgaria and China, around Russia. That was not a joke but the first rigorous conclusion: virtual worlds are real economies with money, inflation and labor. The money supply obeys

M_{t + 1} = M_{t} + F - S

(faucet minus sink); faucet > sink → inflation (WoW gold has crept upward for 20 years). EVE made destruction the main sink (a ship blown up removes ISK and creates demand), hired a chief economist and publishes reports — the only managed sandbox economy there is. Emergent systems are real: design them, or they will design you.

Mechanism 1: who is in the world — Bartle's taxonomy

Richard Bartle is the co-author of MUD1 (1978), the first online world of any kind; his paper "Hearts, Clubs, Diamonds, Spades: Players Who Suit MUDs" (1996) grew out of watching the very first online community. The brilliance is not the list of four types but the fact that they are derived from two axes, which makes them a complete system:

the axis of what your interest is aimed at: the world ↔ players;
the axis of how you engage: acting ↔ interacting.

The suits are a mnemonic from the original: ♦ treasure (Achievers), ♠ a spade for digging (Explorers), ♥ empathy (Socializers), ♣ a club for hitting (Killers). The key conclusion is the ecosystem: the types are interdependent. Killers need victims (Achievers/Socializers); if there are too many Killers, the Socializers — the most fragile — flee first, the Achievers follow, and the world empties out. Open PvP turned Ultima Online into a "killer heaven" and pushed everyone else out; in 2000 Origin/EA split the worlds into Trammel (no PvP) and Felucca (PvP) to save the population. WoW won partly by giving every type its own progression: Achievers — raids/gear, Explorers — lore/mounts, Socializers — guilds/chat, Killers — battlegrounds/arena. Modern models (Quantic Foundry's 12 factors, M09) refine Bartle without replacing him: the four types are the floor.

Mechanism 2: the economy is real — money, faucets, sinks

As soon as players can hand each other items and currency, a money supply $M$ appears, and it lives by the balance of faucets (where currency enters: quest rewards, mob bounties, loot) and sinks (where it leaves: repairs, auction taxes/fees, consumables, destruction):

M_{t + 1} = M_{t} + F - S \Rightarrow Δ M = F - S

If $F > S$ persistently, there is more and more money chasing roughly the same goods — that is inflation. The quantity theory of money makes the link explicit ( $V$ — velocity, $Q$ — the volume of real goods, $P$ — the price level):

M \cdot V = P \cdot Q \Rightarrow P = \frac{M \cdot V}{Q}

With $V$ and $Q$ constant, prices rise in proportion to the money supply. That is exactly the "creeping gold inflation" of WoW over 20 years: the faucets (quests) poured, the sinks (repairs, AH fees, mounts) could not keep up — $Δ M > 0$ accumulated.

A worked example — faucet against sink

Say quests on a server pour in $F = 1.0$ million gold/day, while sinks remove $S = 0.6$ million/day. Then $Δ M = + 0.4$ million/day → over a year the money supply grows by ~146 million. With the volume of real goods roughly fixed, prices creep up — and a newcomer's starting savings are devalued. EVE solves this with a radical sink: every ship destroyed in combat removes its value permanently. "The Bloodbath of B-R5RB" alone destroyed ~11 trillion ISK in one go — an instant sink no theme-park has, plus demand for replacements → work for mining and industry. Destruction makes an economy alive rather than bloated.

Castronova: how "the GNP of Norrath" was measured

RMT markets (selling in-game currency for dollars on eBay) provide an exchange rate between the in-game currency and USD — and turn a game economy into a measurable one. Edward Castronova (an economist at Indiana University) gathered data from 616 EverQuest auctions in 2001 and computed:

319 platinum/hour \times 0.0107 $/platinum \approx 3.42 $/hour

Platinum ≈ 1 US cent (worth more than the yen or the lira at the time), a player's labor ≈ $3.42/hour, and Norrath's GDP per capita ≈ $2266 — above India, Bulgaria and China, nearly level with Russia. It made the news, but the substance was methodological: it showed for the first time that a virtual world is a real economy with measurable GDP, inflation, a labor market and capital, even when the developer pretends there is none. The direct consequence was the post of chief economist at CCP (Eyjólfur Guðmundsson, 2007–2014): EVE hired economists to run its economy, because Castronova had proved it was a real one.

Mechanism 3: design it or ignore it

The design fork (see also the theme-park ↔ sandbox spectrum):

Theme-park (WoW): NPC prices are set by a designer, inflation is "managed" (or ignored) via patches and sinks; the economy is background. The risk is slow uncontrolled inflation and dupe crises.
Sandbox (EVE): prices are set by players, currency and sinks are designed as monetary policy; destruction is the main sink; CCP publishes monthly economic reports (the only major MMO that does). The economy is the game itself.

The conclusion Bartle and Castronova give jointly: the population is an ecosystem, exchange is an economy, and both are real systems with dynamics of their own. Either you design them deliberately, or emergence will design the game itself — usually for the worse (UO's collapse, WoW's inflation, the failure of the Diablo III auction house).

🕹 Games to play — and what to notice

The economy and the Bartle dynamics are visible right in the interface: the auction house, the prices, who is squeezing whom out of the world. Play up the ladder — from a designer-managed shop window to a fully player-run sandbox.

World of Warcraft theme-park · managed inflation

The economy as background: NPC vendor prices are fixed, the Auction House is player-run but has sinks (fees, repairs, mounts). Over 20 years, a textbook case of creeping gold inflation.

🎮 Play: open the Auction House and compare consumable prices with quest rewards — find the faucets (quests, vendor sales) and the sinks (repairs, AH fees, expensive mounts). Ask yourself: what is removed less than it is poured in? That is where inflation lives. Then check the price of the WoW Token — an attempt to legalize RMT and add a lever of control.

Ultima Online a Bartle collapse · Trammel/Felucca

A living textbook of the Bartle ecosystem: open world PvP made UO a "killer heaven", and Socializers and Achievers ran. In 2000 the world was split into Trammel (no PvP) and Felucca (PvP) — literally surgery on the population.

🎮 Play: on a UO shard (a fan server, say) feel the difference between Trammel and Felucca zones — where Socializers build shop-houses and trade, and where the Killers rule. Notice a player vendor (a house turned shop): that is emergent retail no designer planned for. Players set the prices.

Diablo II → Diablo III RMT and the failed real-money auction house

D2's loot was so psychologically sticky that it spawned RMT and a gold-farm culture (still alive today). D3 tried to legalize it with the Real Money Auction House (RMAH, 2012) — and that killed the core loop: the optimal way to play became "buy it, don't find it". The RMAH was shut down in 2014.

🎮 Play: in Diablo (II Resurrected, or III after Loot 2.0) run an act and feel the "30 seconds of loot" — why the drop is so absorbing. Then mentally put the RMAH back: notice that the moment an item can be bought with real money, why play at all? A case study in monetization eating its own loop.

EVE Online a managed sandbox · MER · a chief economist

A fully player-run economy on one shard: a market, production chains, legal scamming, and destruction as the main sink. CCP keeps a staff economist and publishes monthly reports (MER) with charts of ISK faucets and sinks.

🎮 Play: in EVE open the market of a major hub (Jita) — look at order depth, the bid/ask spread, how price reacts to events. Then find a recent CCP Monthly Economic Report and locate the largest ISK faucet and the largest sink. You are looking at the monetary statistics of a game, published the way a central bank publishes them.

Deep end · economics: faucets, sinks and why destruction cures inflationskippable

The health of a game economy is a matter of managing the money supply and the "material" supply (resources/items). Two loops:

The monetary loop (currency)

Faucets: NPC bounties, mission/incursion rewards, selling loot to a vendor — a pure injection of new currency out of thin air.
Sinks: broker fees and transaction tax (market fees), the cost of skillbooks/clones, structure upkeep, repairs. A pure removal.

If faucets are structurally larger than sinks, $Δ M > 0$ accumulates → inflation (see $P = M V / Q$ ). EVE's historical headache: the main faucet — bounties from ratting bots — grew faster than the sinks, and CCP tuned fees and mechanics by hand the way a central bank tunes rates.

The material loop (items/resources)

Here is EVE's brilliance: PvP destruction is a sink for both supplies at once. A ship blown up removes both the ISK equivalent and the minerals invested in it, and immediately creates demand for a replacement → mining → industry → market. A closed, living chain. In a theme-park, where items are barely ever destroyed (account-bound gear, no loss on death), the material supply only grows → old content is devalued and artificial "resets" are needed (a new gear tier every expansion is a sink through devaluation rather than destruction).

The link to dupes

An item dupe (see persistence) is an illegal faucet printing both currency and goods outside the model. One unnoticed dupe exploit can crash the exchange rate in days — which is why economic integrity rests on ACID trade transactions, not only on "reward balance".

Deep end · design: Bartle as a coverage checklist, not a label on a playerskippable

The big misreading of Bartle is treating him as "type your player". Bartle himself stressed that most players are a mix of types, that pure ones are rare, and in 2003 he extended the model to 8 types by adding an "explicit ↔ implicit" axis (deliberate vs spontaneous). The taxonomy's value is not a label on a person but a coverage checklist for design.

The failure mode — mono-optimization

Most failed MMOs accidentally optimized for one type and pushed the rest away. Open PvP → killer heaven → the Socializers run (they are the most fragile to hostility), then the Achievers (nobody to show status to, nobody to win the race against), and the world empties. That is a structural effect regardless of the fact that individual people are mixes: what matters is not "which type are you" but whether every motivation is covered by a viable progression.

How WoW closed every quadrant

Achievers ♦ — raid progression, gear score, achievements;
Explorers ♠ — hidden lore, secret mounts, datamining;
Socializers ♥ — guilds, cities, chat as a "third place";
Killers ♣ — battlegrounds, ranked arena (PvP in a sandbox, not on top of peaceful players).

The key is giving Killers an allocated arena, so that their game is not a tax on everyone else's. That is designing an ecosystem instead of hoping it self-balances.

Analogy

An MMO economy is a bathtub. Faucets (quest rewards, bounties) pour currency in from above; sinks (repairs, taxes, ships blown up) drain it out below. Close the drain while leaving the tap open and the tub overflows (inflation), and your starting gold turns into water. EVE keeps the drain wide open by letting players sink each other's ships: the level holds, somebody constantly has to top it up (mining/industry) — and the economy lives. WoW mostly relies on small drains and a large tub — and over 20 years the water still crept up. And the player population is the life in that pool: remove the predators (Killers) and the achievers get bored; let in too many and everyone scatters, leaving nobody to top up the tub.

Why it matters

An MMO with trade is a society with a real economy. Castronova proved it is measurable (GDP, inflation, labor), Bartle that its population is an ecosystem you can collapse with careless balancing. Either you design the incentives and the money supply deliberately (EVE hired an economist and publishes reports), or emergent dynamics will design the game for you — usually for the worse (UO's killer collapse, WoW's inflation, the failure of the Diablo III auction house). The transferable truth is wider than games: any system with heterogeneous agents and transferable value has emergent socio-economic dynamics — ignore them and they will bite you.

🔁 Beyond games — where this transfers

The lesson is incentive and money-supply design for a system of agents; it transfers anywhere there is a reward and an exchange.

ML / AI: an RL reward function is a currency faucet for the agent, and it has exactly the same disease. Reward hacking = a player who found an unplugged faucet: the agent inflates its metric without doing the work when there are no "sinks"/constraints (that is Goodhart's law, word for word). Multi-agent RL models virtual economies directly — the AI Economist (Salesforce) optimized tax policy in a simulated agent economy; training trading/marketplace agents inherits inflation, dupe arbitrage and collusion. The Bartle lesson for ML is user heterogeneity: a recommender optimizing one metric (engagement) squeezes out "other types" the same way open PvP drove out Socializers — mono-optimization and the filter bubble are your product's killer heaven. RLHF, too, is a token economy of preferences with sinks and faucets of its own.

Economics / markets: faucets/sinks = issuance/withdrawal (taxes, the central bank rate); CCP's chief economist = literally a central bank; RMT = grey arbitrage and Gresham's law; inflation $P = M V / Q$ works the same in Norrath and in a country.

Product / business: in-app currencies, token sink design, whale economics, loyalty programs — all of it is money-supply management; KPI gaming (Goodhart) = farming a faucet; design your sinks and your coverage of motivations, not a single metric.

Distributed systems / security: a dupe = a double spend; economic integrity = consensus and ACID transactions; a blockchain is essentially a durable store with dupe protection, raised to the level of an entire currency (see persistence).

The principle: every incentive system has faucets and sinks; forget the sinks and you get metric inflation and a collapsed agent ecosystem. Design both loops and cover every motivation.

🔧 Run it and poke at it — on your home machine

What to play is above (🕹). This part is for people who want to do the economics by hand.

🔧 Poke at it (debug) ~40 min, CCP MER + 20 lines of code

Download a recent EVE Monthly Economic Report (CCP publishes PDFs/charts) and find the two charts: ISK faucets and ISK sinks. Which faucet is the largest (usually bounties)? Do the sinks add up to the faucets, or is

Δ M \neq 0

? Then write a 20-line simulator: M += F − S per day, add a one-off "destruction event" (−11 trillion) and plot

M

with and without the sink — inflation, seen with your own eyes.

🧪 Test it (with designer eyes) ~15 min

Here is a design: quests pay gold, repairs are free, items are account-bound and indestructible. Predict the failure mode (inflation plus loot devaluation) and write down three sinks that cure it. Then do a Bartle audit of your favorite MMO: which of the ♦♠♥♣ quadrants have progression, and which is short-changed? What happens to the population if you strengthen the Killers?

Checklist: found the largest faucet and sink in a real MER; plotted the inflation curve in a mini-simulator; wrote three sinks for a broken design; ran a Bartle audit for quadrant coverage.

Connections

foundation

Persistence and sharding — EVE's single shard = one market and a measurable economy; sharding splits it into N independent ones. And a dupe = an illegal faucet, cured by the ACID transaction from that lesson.

foundation

Netcode taxonomy — an authoritative server is mandatory for economic integrity: otherwise the client "prints" currency locally.

F2P economics and gacha — here the economy is player↔player (emergent); there it is designer↔player (extractive monetization: pity, gacha). Two sides of "money in games".

overlap

Classical vs ML — the economy and the Bartle balance are tuned by craft and monetary policy, not by learning; another domain where "AI is not the main thing here".

Questions worth asking

If almost every player is a mix of Bartle types, isn't the taxonomy useless?

No, because its value is not a label on a person but a coverage checklist. Even if every individual player is a mix, failure arrives when the design leaves a whole motivation without a viable progression: then the corresponding behavior has nowhere to go and part of the audience leaves. A killer heaven drives out Socializers structurally, regardless of the fact that many of them are also a bit Achiever. Bartle is not asking "who are you" but "are all four ways of having fun covered" — a design invariant, not psychometrics.

Why does a destructive sandbox (EVE, where you lose everything) have a healthier economy than a theme-park where you hoard everything?

Because destruction is a sink for money and matter at once. Without sinks, faucets (rewards) inevitably drive

ΔM > 0

→ inflation, and indestructible items only accumulate → old content is devalued. Destroying a ship removes ISK and minerals and creates demand for a replacement → a living mining→industry→market chain. A theme-park without destruction has to fake the sink through artificial devaluation (a new gear tier every expansion). The paradox is only apparent: losing things is what sustains demand and keeps the supply under control.

"Norrath's GNP is higher than Russia's" — a real economy, or a stunt for the headline?

Both. The measurement is real: the RMT rate (eBay) gives an exchange rate from platinum to dollars, and from there labor at $3.42/hour and GDP per capita of $2266. But the comparison to a nation is shaky: Norrath's "per capita" is self-selected players with plenty of free time, not a representative economy with children, retirees and basic needs. Castronova knew that; the goal was methodological — to show a virtual economy is measurable with the same tools, not to claim Norrath is richer than Russia in any meaningful sense. The headline was the bait; the method was the contribution.

Why did Diablo III's real-money auction house fail if RMT thrives in Diablo II?

Because making RMT official killed the pleasure of loot hunting: once the best item can simply be bought, the optimal play becomes "open your wallet, don't play" — the "30 seconds of loot" core loop is gutted. On top of that Blizzard became the guarantor of item flow, accelerating loot deflation. Grey RMT works precisely because it is friction and shame: most people will not go to a third-party site, so the reward loop stays intact. Sanctioned RMT removes the friction — and collapses the reward psychology the game rests on. D3 shut the RMAH down in 2014 and moved to Loot 2.0 (personal, untradeable loot) — the opposite decision.

Can you even make an MMO with no economy?

Practically speaking, no — if players can hand over any value (items, currency, even time and services), an economy appears. "No economy" is a designer's illusion: a black market, barter and services will show up regardless. The only way to truly avoid it is to make everything account-bound and untradeable (like Loot 2.0's personal loot) — but that destroys the value for Socializers and traders and turns the MMO into a near-single-player game. So you cannot "not have" an economy; you can only choose between designing it and putting up with the emergent one.