← all systems · see it in The System
Tacit collusion
Tacit collusion is coordination among competitors (the encyclopedia's 'agreement about coordination of conduct' means a shared understanding, not an agreement in the legal sense, which tacit collusion lacks by definition) who do not explicitly exchange information but reach supra-normal (supracompetitive) prices or other coordinated conduct. Economics reads it as an equilibrium of repeated play: firms that meet again and again hold prices above the one-shot level because each expects a deviation to be detected and answered by retaliation (a price war, or a punishment aimed at the deviant), and it holds only if they weigh future profits enough. What sustains it: few sellers, entry barriers, frequent interaction and price changes, transparent prices, symmetric firms, stable or growing demand, contact in several markets; what breaks it: entry, demand swings and booms, innovation, asymmetries, mavericks, large buyers, opaque or infrequent dealing. Coordination itself runs through price leadership (dominant-firm or barometric), focal points and prices read as messages. Competition law sorts it into conscious parallelism (no communication) and concerted action (some indirect communication), separates it from an explicit agreement (a cartel), and in the US and EU treats parallel conduct alone as lawful, requiring plus factors or evidence of communication; merger control acts on market structures conducive to it. Pricing algorithms (Q-learning, LLM agents) have been shown to reach it in simulation, with field evidence from German gasoline. Walked here as that market conduct and the practice of reading it; the market form is sys:oligopoly, the explicit kind sys:cartel.
Properties published scienceour proposal
Sourced fields, and our own readings.
| asymmetry | low-cost or larger firms need larger shares; a dominant firm leads |
| born at | per episode; the earliest dated tacit case read is the credit-card focal points of the early 1980s (Knittel and Stango 2003); the idea from Chamberlin (1929, contents only), Stigler (1964) and Friedman (1971) as cited |
| born precision | years (earliest case) |
| boundary | Inside: the coordinated conduct of a few rival sellers (or bidders) who meet repeatedly: the collusive path of prices, shares or bids, the shared understanding of it (focal points, who leads, that increases are matched), the monitoring that detects deviations and the retaliation that answers them. Outside: the buyers who pay, the firms' own internal decisions (each firm acts independently), entrants, competition authorities and courts, and explicit agreements (a cartel is the explicit kind, a sibling). The legal line runs through the middle of the economic one: economics counts any equilibrium supported by punishments as collusive, law asks whether an agreement was reached and how. |
| cheater | the deviant who would undercut for a one-period gain: an off-path figure that equilibrium keeps from appearing; the sources read do not say who started Perth's 2009 BP-Caltex price war |
| drive | equilibrium |
| enforcement | sanction |
| flips when | demand peaks, entry, innovation, a maverick, a large buyer concentrating orders, or the discount factor falling below the threshold |
| joint output | prices and profits above the one-shot level that no firm can hold alone |
| knockout | If instead one firm slightly undercuts the other, it captures the entire market and thus the entire collusive profit, but the ensuing price war will eliminate any future profit (Ivaldi et al., base model) |
| nested level | A kind of collusion (the tacit kind, beside the cartel); a conduct regime inside an oligopoly (its L5, in the oligopoly walk); runs on its member firms' decisions and, more and more, on their pricing algorithms; read and policed by competition law. |
| obligate | facultative |
| partners | sys:firm |
| recurrence | only with repeated interaction; more frequent meetings make it easier |
| regime | starts by learning (Perth: three and a half years) or after talk; ends in price war or breakdown |
| saturation | upper bound: the joint-profit (monopoly) price; with patient firms a range of prices can be sustained and where it settles is not predicted (o-which); with asymmetric firms the joint-profit point needs side-transfers tacit colluders cannot make |
| scale | social |
| separability | each firm keeps its own ownership, costs and decisions |
Measures (22) published science
| critical discount factor, n identical price-setting firms | delta* = 1 - 1/n: 0.50 for two firms, 2/3 for three (the report prints '1/3 = 0.67') dimensionless |
| critical discount factor with interaction every T periods | delta* = (1/2)^(1/T) dimensionless |
| critical discount factor with demand swings of size epsilon | delta* = (1+epsilon)/(2+epsilon) dimensionless |
| critical discount factor with entry probability mu | delta* = 1/(2-mu) dimensionless |
| encyclopedia duopoly example threshold | delta > 6/13 (gain from cheating $30 against $35 a period lost) dimensionless |
| Perth price cycle after the focal points (characteristic clock) | 7 days (before April 2009: 7 to 35 days; station-level mean over the sample 8.8 days, SD 4.4) |
| Perth retail gasoline margin before and after March 2010 | 4.85 cents a litre (SD 4.19) before; 10.01 (SD 4.68) after cents per litre |
| Perth margin rise relative to other Australian cities (difference-in-difference) | 3.49 (SE 0.39) cents per litre |
| Perth transition to the focal-point equilibrium | three and a half years (April 2009 to August 2012) |
| BP share of Perth stations | 22 percent on average, 2005-2015 percent of stations |
| wait for a price lead to be matched under Perth's next-day price notification law | until the next day (before the law, Shell led over 85% of the time) days |
| length of the learned punishment (Q-learning duopoly) | 5-7 periods in most cases |
| Q-learning profit gain by number of firms | 85% (two firms), 64% (three), 56% (four) percent of the gap between Bertrand-Nash and monopoly profit |
| Q-learning periods to convergence | about 850,000 periods (a few seconds of CPU time) |
| lab no-talk selling price, by whether a talk phase came first (four firms) | 6.0 (SD 0.6) when no talk came first; 67.0 (SD 33.8) after a talk phase experimental price units |
| lab price rise from talking, by number of firms | +43.2 (n=2), +75.3 (n=4), +62.5 (n=6), +54.1 (n=8) experimental price units |
| experimental Cournot markets classed collusive | 2 of 6 duopolies; 0 with three, four or five firms markets |
| oligopoly experiments behind the meta-analysis | more than 100 papers, some 500 settings papers / settings |
| bids in the PCS C-block auction | roughly 30,000 bids in 183 rounds, all public |
| share of AI pricing jobs among pricing jobs | up more than ten-fold since 2010 ratio |
| the textbook spiral (The Making of a Fly) | one seller priced at 1.27 times the other, which matched it; the book reached $23 million a copy US dollars |
| MillerCoors joint venture: prices and markups above Nash-Bertrand | prices 6-8% higher; markups 17-18% higher percent |
Open contentions (7) open contention
Where sources or readings disagree. Both stay on the map.
- C-3462 Can pricing algorithms learn to collude on their own?
booked: The map keeps the competing readings; none is ruled.
the other reading: the Q-learning results need long training and particular settings (oligopoly walk's specialist lead); Fish et al.: questions have arisen about the real-world relevance of Q-learning; LLMs sidestep those barriers; Brown and MacKay 2023: algorithms that differ in repricing speed raise prices in a competitive equilibrium with no reward-punishment, so the German result fits a non-collusive mechanism too; specialist's unsourced lead: Abada and Lambin 2023 read high Q-learning prices partly as a product of imperfect exploration - C-3545 Do price wars come in booms or in slumps?
booked: The map keeps the competing readings; none is ruled.
the other reading: as in the registry; specialist's reading (unsourced): the two are not rivals but conditional on the information structure (unseen demand and imperfect monitoring vs seen demand and perfect monitoring) - C-3599 Can firms hold prices high without ever talking to each other?
booked: The map keeps the competing readings; none is ruled.
the other reading: Green, Marshall and Marx (as cited by Byrne and de Roos): firms are unlikely to reach a mutual understanding of a collusive strategy without direct communication in the initiation phase - C-3602 Does a growing market make silent price coordination easier or harder?
booked: The map keeps the competing readings; none is ruled.
the other reading: case law and Commission guidelines often read demand growth as hindering collusion (reported by Ivaldi et al.); entry in growing markets may reverse the effect - C-3633 How many competitors does it take to end silent price coordination?
booked: The map keeps the competing readings; none is ruled.
the other reading: theory gives a smooth rise of the critical discount factor with n (1-1/n), no sharp cutoff - C-3670 Where does the law draw the line between parallel pricing and collusion?
booked: The map keeps the competing readings; none is ruled.
the other reading: Posner's view (quoted by Harrington): a tacit meeting of the minds can be an agreement; Ezrachi and Stucke: algorithmic tacit collusion should raise concerns and call for intervention; Kaplow (via Byrne and de Roos): the legal definition of collusion is highly ambiguous; US law also reaches unilateral invitations (FTC order against U-Haul, 2010) and accepted invitations without previous agreement (Interstate Circuit, as quoted by Kovacic et al.); the economists' 'tacit' examples at the signalling edge (code bids, advance announcements in Ethyl) overlap the lawyers' 'agreement'; EU: a concerted practice under Article 101 is itself prohibited once shown (Ezrachi and Stucke); the specialist's Airtours reading of collective dominance is not yet read - C-3672 Do equal-sized firms coordinate prices more easily, or does one big leader make it easier?
booked: The map keeps the competing readings; none is ruled.
the other reading: symmetry as the main merger concern (Ivaldi et al.); asymmetry as an enabler through leadership (Byrne and de Roos); MillerCoors (Miller and Weinberg 2017): coordinated price rises after a joint venture between two large brewers; the specialist reads it as a case on the symmetry side (not the paper's words)
Cross-scale patterns seen here (22) our proposal
Shapes that recur when the same kind of problem is solved at different scales, and how each shows in Tacit collusion. Each links to every other system that shows it.
- borrowed-then-held (structural)
Coordination first supplied by talk (a lab phase in which firms could communicate) is then held tacitly after talk is switched off: no-talk prices 67.0 after a talk phase against 6.0 without one. Both phases shown. - same role, different builds (structural)
The job of keeping rivals from undercutting is done by unrelated builds: grim-trigger reversion in theory, stick-and-carrot punishments learned by Q-learning, price-war worry in LLM reasoning, fighting ships, a retaliatory bid on the rival's own licence; in animals, territory holders lowering defenc - same toolkit, independent builds (structural)
Q-learning algorithms and, separately, LLM pricing agents, given only prices and profits, both arrive at reward-punishment schemes whose punishment decays over several periods, the shape repeated-game theory names for people. - trade one property for another (structural)
Price is traded for legal safety: firms that coordinate tacitly avoid fines and jail but reach a lower steady-state price than talk would give; in the lab, industries whose gain from talking is below the expected fine stay tacit, the rest form cartels. A second form: longer price wars buy discipline - who opens the door (structural)
The door a price signal enters by is opened by the receivers ahead of time: each firm's monitoring tools and the shared rule that increases will be matched are in place before the lead is posted; regulators close doors (click-box bidding, anonymous bids) to stop coded signals. - barrier-gates-complexity (partial)
Behind entry barriers the few can build and keep a coordinated regime (leaders, focal points, punishment); the without case is shown in the model: as entry becomes certain the critical discount factor goes to 1 and collusion is almost impossible. - causal ceiling (partial)
Two limits set by the information structure, not by effort: with only own sales visible perfect collusion is impossible; without express communication no firm will lead above the steady-state price, so prices stay below the best sustainable level. - conservation law plus closure (partial)
The incentive constraint (a balance of today's gain against tomorrow's punishment) leaves the outcome open: many equilibria satisfy it and theory 'does not predict which'; a second rule is needed to close it, equilibrium selection by focal points, leadership and experiments. - coordination-before-mapping (partial)
Pricing agents coordinate without any model of the market: LLM agents are not given the demand function, Q-learners learn by trial and error; their value tables or prompts are the only held state. - holding / entrainment (partial)
Once Thursday jumps became focal, every firm's station-level cycle converged to seven days: the coordinated regime sets the clock of each station's price changes (the leader's Wednesday move led it). - lightning (partial)
Two repricing rules (match the rival; price 27% above the rival) ran away with no held regulation until a person noticed and reset the price, after the book reached $23 million a copy. - nested ladder (partial)
US law hedges against false inferences from parallel conduct at nested stages: pleading (plausibly suggesting agreement), summary judgment (tending to rule out independent action), directed verdict (more than parallel conduct). Each stage's bar includes the one below. - same shape, opposite sign (partial)
One incentive condition (today's gain from deviating against tomorrow's punishment) runs both ways with the sign of demand's path: growth makes tomorrow weigh more and eases collusion; a peak before a decline makes today weigh more and breaks it. - symmetry breaking (name the cut) (partial)
Among symmetric firms, who should lead and at what price is not common knowledge; the cut is one firm taking the lead (BP, largest network) and the broken direction (leader and followers, Wednesday then Thursday) sets the structure that follows. - templating (partial)
A simple pricing structure reused unchanged week after week by every station (Thursday jump, 2-cent cuts), and the same device across cases: BP's standardisation recalls the Sugar Institute cartel's rules; record companies' few-category price grids. - The carry (partial)
The wait: a leader's increase cannot settle until rivals match it, and the leader loses demand meanwhile, so the wait decides who will lead; when Perth's law made every match wait a day, leadership spread from Shell to BP, Caltex and Shell. Punishment waits too: the less often firms meet (every T pe - The Payment to Entropy (partial)
Signal form: coordination without words pays tolls. Noise: price wars set off by bad luck that cannot be told from cheating; a lower steady-state price than talk would reach; and each added player lowers the share of the joint-monopoly gain that arrives (Q-learning: 85% with two, 64% with three, 56% - the whole sets the local reference (partial)
What counts as a deviation for any one firm is fixed by the shared path (the focal price and day); each firm reads its own sales and prices against that reference. - Umwelt: each system carves its own slice of one shared medium (partial)
Each player carves its own slice of one market: a firm in a Green-Porter market sees only its own sales; LLM agents see only prices; courts see parallel conduct and plus factors; regulators narrow what bidders can see (anonymous bids). The set of channels here is the conduct's Umwelt. - cephalization (surface)
The lead tends to sit with the firm that meets new information first: the barometric leader as the best barometer, and in Perth possibly BP, owner of the only local refinery and perhaps best informed about rivals' costs (the source says 'may'). - runaway from an unstable balance (surface)
The textbook spiral: two pricing rules whose ratio exceeded one fed on each other with no balance point shown. - smooth to sharp in finite time (surface)
In the base model sustainable collusion jumps from none to full (the monopoly price) as the discount factor crosses 1/2; the source calls this knife-edge specific to the simple example.
Valence: what it trades, and with whom (9) published science
Each exchange with a neighbour, read from both sides: who gains, who pays, and what carries it.
- Human (Homo sapiens) via sys:sensory-environment · Tacit collusion mixed
- Law (rules and law) via sys:human-society · Tacit collusion mixed
- Large language model via sys:written-record · Tacit collusion gain
- Oligopoly via sys:market · Tacit collusion gain
- Firm (the economic unit) via sys:market · Tacit collusion gain
- Consumer via sys:market · Tacit collusion gain
- Market competition via sys:market · Tacit collusion gain
- Cartel via sys:market · Tacit collusion mixed
- Pricing algorithm via sys:market · Tacit collusion gain
Partners and parts published science
Built from / builds (5)
- Firm (the economic unit)
- Pricing algorithm
- Price leadership
- Focal points
- Retaliation (price wars, targeted punishment)
Sits inside / holds (3)
expressed by (17)
- Templating / Standardization
- Niche Construction (alter environment)
- Distributed Autonomy/Decentralized Execution (Subsidiarity)
- Internal Coordination/Regulation
- Sensing/Scanning
- Uncertainty Reduction/Learning
- Active Experimentation / Probing (Optimal Experimental Design)
- Surprise / Change-Point Detection
- Social Learning / Public Information Use
- Evade Detection
- Precision Attack
- Resource Denial/Attrition
- Reactive Response
- Self-Monitoring
- Validation of Identity/Intention
- Resilience & Repair
- Deterrence
floor filled by (12)
- Oligopoly
- Law (rules and law)
- Market and trade
- Auction
- Monopoly
- Firm (the economic unit)
- Consumer
- Entry barriers
- Pricing algorithm
- Price leadership
- Focal points
- Retaliation (price wars, targeted punishment)
represents (2)
- Game theory
- the repeated-game account of tacit collusion held by economics (trigger strategies, the critical discount factor)
Kinds (6)
- Market competition
- Collusion
- Conscious parallelism
- Concerted action (the economists' middle category)
- Algorithmic tacit collusion
- Tacit collusion in repeated auctions
Notes published science
- In Perth, one petrol company taught its rivals to collude without a word: BP's price jumps and experiments turned Thursdays into the market's price-rise day and 2-cent cuts into the rule, and over three and a half years margins roughly doubled, from 4.85 to 10.01 cents a litre (src:byrne-deroos-2019).
- Talking leaves a mark after it stops: in lab markets where firms could first talk and then could not, prices stayed at 67 on average, against 6 where they never talked; the authors warn this means fines based on post-cartel prices undercount the harm (src:fonseca-normann-2012).
- In a US spectrum auction, a retaliatory bid worked like a rap on the knuckles: after Cavalier Wireless bid on T-Mobile's Hawaii licence, T-Mobile bid once, in the same round, on Cavalier's Columbia, Missouri licence, and Cavalier never bid on Hawaii again (src:bajari-yeo-2008).
Layers
The ways it takes in its world, where they meet, and how it acts back, each from L1 (where it meets the world) up.
tacit collusion as economists and competition lawyers describe it: a few sellers' repeated market to coordinated prices (reader's view)
- L1: A market where the same few sellers meet buyers again and again, with posted prices (or repeated auctions with public bids)
- L2: The conditions that make it possible: few sellers, entry barriers, frequent interaction and price changes, symmetric costs and capacities, stable demand, contact in several markets
- L3: Monitoring: rivals' prices, bids and sales made observable (posted prices, trade data, pricing software and surveillance tools); a deviation identified
- L4: Reading the market: inferring from market data whether the others are on the collusive path, or whether a fall in sales is bad luck or cheating
- L5: Discipline: the incentive constraint of repeated play: the short-run gain from undercutting against the retaliation it would bring, weighed by patience (the discount factor)
- L6: Mutual understanding without agreement: focal points, who leads and that increases are matched, prices read as messages
- L7: Coordinated conduct acted on the market: supracompetitive prices held, leads followed, shares and licences divided, deviations punished
the boardroom (shared floors)
- L4: Reading the market: inference from prices, sales, leads, words and history about whether the others keep to the collusive path
- L5: The incentive constraint: short-run gain from deviating against retaliation, weighed by the discount factor; raised or lowered by entry, demand, the law and multimarket contact
- L6: Mutual understanding without agreement: focal points (Thursday jumps, 2-cent cuts, a nonbinding ceiling), who leads, that increases are matched
- L7: The coordinated conduct: supracompetitive prices held, shares divided, deviations punished
way in: Competition law and enforcement
- L1: The rule: the illegality lies in an agreement (Sherman Act section 1)
- L2: The evidence standard: parallel conduct alone is not enough; proof must tend to exclude independent action
- L3: Merger review of structures conducive to coordination
way in: Contact in other markets
- L1: The same rivals present in several markets
- L2: More frequent interaction across markets
- L3: Asymmetries in single markets evened out overall
way in: Demand swings, growth and innovation
- L1: Demand rising, falling or cycling; innovation arriving
- L2: Today's demand read against tomorrow's
- L3: A peak raises the gain from deviating now
way in: Own sales and demand
- L1: Buyers' purchases, which each firm sees only as its own sales when rivals' prices are hidden
- L2: Own sales and market share held against the collusive path
- L3: A fall in sales read as bad luck or a rival's cheating
way in: Potential entrants
- L1: Outside firms able to enter (hit-and-run entry)
- L2: Entry barriers that keep them out
- L3: The prospect of entry shrinking what retaliation can take away
way in: Pricing algorithms' inputs
- L1: All prices set in the period and own demand, fed to the algorithm
- L2: The algorithm's state: last period's prices (Q-learning) or the prompt and history (LLM agents)
- L3: A learned reward-punishment response to the rival's price
way in: Public words and coded signals
- L1: Unilateral public announcements (a CEO speaking of 'price leadership' on an earnings call)
- L2: Coded signals carried inside market acts (licence numbers in the last digits of bids)
- L3: Some information exchanged without an explicit agreement (concerted action)
way in: Rivals' prices and bids
- L1: The rivals' posted prices and bids, the only medium between them where nothing else is said
- L2: Observability: published prices, public bids, monitoring and pricing software that track rivals
- L3: A deviation identified as such
way in: The history of play
- L1: Past episodes of play, including any earlier phase of talk
- L2: An established leader's record of influencing rivals
- L3: The past carried forward as higher prices (hysteresis)
way in: The leader's price moves
- L1: The leader's price jumps and experiments across its network (BP's Wednesday jumps in Perth)
- L2: Rivals see the lead and follow on the focal day
- L3: The moves read as messages
way out: Leads, experiments and coded bids sent to rivals
- L1: A price jump or coded bid placed where rivals see it
- L6: Test rivals and signal intentions
way out: Parallel conduct left in the record
- L1: Long high-frequency price panels that reveal transitions
- L7: Parallel prices and bids, with or without plus factors
way out: Prices held above the competitive level
- L1: Buyers pay the margin (Perth: 4.85 to 10.01 cents a litre)
- L6: Prices at the focal point
- L7: Hold the collusive price
way out: Retaliation: price wars and targeted punishment
- L1: Price war or a retaliatory bid lands on the deviant's market
- L5: Kept just long enough to discipline
- L7: Punish the deviation for a finite phase, then return
way out: Shares and territories divided
- L1: Licences allocated among bidders
- L7: Give the low-cost firm the larger share