Experiment – Vernon Smith, Gerry Suchanek, and Arlington Williams The Level-k

The intersection of empirical finance, experimental economics, and behavioral game theory represents one of the most intellectually transformative chapters in modern social science. For decades following the formalization of the Efficient Market Hypothesis (EMH) by Eugene Fama and the mathematical maturation of Rational Expectations Equilibrium (REE) models, prevailing theoretical orthodoxy maintained that asset prices in competitive markets must uniquely reflect their fundamental intrinsic values. According to this classical view, market participants operating in decentralized, price-setting institutions process information rationally, backward induct over known finite horizons, and immediately extinguish any temporary arbitrage opportunities. Speculative bubbles—defined as persistent, systematic trades occurring at prices substantially above the mathematical expected value of an asset’s future cash flows—were considered theoretical impossibilities in frictionless markets governed by common knowledge of rationality.

This theoretical edifice was radically challenged by a landmark 1988 study published in Econometrica by Vernon L. Smith, Gerry L. Suchanek, and Arlington W. Williams (hereafter SSW). Designing a rigorous laboratory environment in which the fundamental value of an asset was fully transparent, objectively computed, and declining deterministically to zero over a finite number of trading periods, SSW observed massive, repeatable speculative bubbles followed by catastrophic market crashes. In market after market, participants traded assets at multiples of their intrinsic value, despite possessing complete information regarding the dividend distributions and terminal horizons. The persistent replication of this empirical anomaly invalidated the hypothesis that laboratory double auctions naturally enforce rational expectations pricing when intertemporal speculation is possible, opening up deep questions regarding the cognitive foundations of market clearing prices.

To resolve the profound disconnect between the neoclassical prediction of backward induction and the observed reality of laboratory asset dynamics, modern behavioral economists have increasingly turned to non-equilibrium models of iterative strategic thinking, specifically Level-k reasoning and cognitive hierarchy theory. Rather than assuming that all market participants possess infinite cognitive depth and operate under flawless common knowledge of rationality, Level-k frameworks decompose the population into discrete cognitive strata. In this analytical architecture, zero-level agents act according to non-strategic heuristics or random behavior, while higher-order agents optimize their trading strategies against beliefs that counterparties operate at lower cognitive tiers. By viewing the foundational experiments of Smith, Suchanek, and Williams through the analytical lens of Level-k game theory, researchers can construct rigorous microfoundations for the emergence, parabolic expansion, and eventual collapse of asset bubbles, reconciling human cognitive architecture with observed market dynamics.

1. Historical Foundations of Experimental Economics and Asset Market Design

1.1 The Pre-1988 Landscape of Market Efficiency Hypotheses

In the decades leading up to the late 1980s, financial economics was almost entirely dominated by the paradigm of market efficiency and the analytical framework of rational expectations. Supported by the mathematical formalizations of Paul Samuelson and the empirical syntheses of Eugene Fama, the standard neoclassical view posited that competitive spot asset markets acted as informationally efficient aggregators. In an ideal double auction, market prices were presumed to equal the discounted present value of all expected future cash flows. Under this analytical paradigm, speculative price runs could only occur as rational responses to shifts in underlying exogenous fundamentals or risk premia; endogenous, self-reinforcing price inflation was treated as an empirical impossibility or an artifact of unobservable informational frictions.

Parallel to this theoretical development, the pioneering laboratory work of Vernon L. Smith during the 1960s and 1970s had established that experimental markets possessed an extraordinary capacity to achieve competitive equilibrium. In simple flow-demand commodity environments—where buyers and sellers were assigned stationary, private redemption values and unit costs—Smith’s computerized and manual continuous double auctions demonstrated rapid price convergence to theoretical Walrasian competitive equilibria within a few trading rounds, even when market participants possessed zero global knowledge of the aggregate supply and demand curves. These early triumphs cemented a broad presumption within experimental economics: the double auction institution was robust enough to drive decentralized human behavior toward neoclassical optimality.

However, an important theoretical distinction separated these early commodity markets from dynamic asset markets. In flow-demand commodity settings, units had no durability; value was realized instantly upon consumption, eliminating intertemporal speculation. In contrast, financial assets derive their value from prospective future streams of dividends and terminal liquidation values. Neoclassical theorists presumed that even in asset markets, the application of backward induction across a finite horizon would discipline prices to track fundamental value at every single period. If terminal payoffs are known, backward induction dictates that no rational trader will purchase an asset above its terminal value in the final period; working backward through time, this logic theoretically forces all preceding trading prices to equal the fundamental intrinsic value. Yet, until the late 1980s, this hypothesis had never been subjected to controlled empirical stress-testing under conditions of explicit, common-knowledge intrinsic value.

1.2 The Collaborative Genesis of Smith, Suchanek, and Williams

Recognizing the urgent methodological necessity to test asset pricing theory free from confounding real-world noise, Vernon L. Smith joined forces with Gerry L. Suchanek and Arlington W. Williams. In field financial markets, evaluating whether an asset’s price deviates from its fundamental value is notoriously fraught due to the joint hypothesis problem: any empirical test of market efficiency is simultaneously a test of the specific asset pricing model used to estimate fundamental value. In the wild, intrinsic value is unobservable, subjective, and contingent on unknown future macroeconomic shocks, cash-flow variability, and unobservable discount rates. Consequently, observers of historical episodes such as the Dutch Tulip Mania or the South Sea Bubble could always argue that sky-high valuations reflected rational beliefs regarding future economic transformations.

The explicit research objective of Smith, Suchanek, and Williams was to construct an experimental architecture that eliminated this observational ambiguity. By stripping away macroeconomic noise, private information asymmetry, and ambiguous dividend structures, they sought to evaluate trading behavior in an environment where fundamental value was mathematically deterministic, universally communicated, and common knowledge among all participants. The asset was designed with an explicitly defined, finite lifespan spanning a set number of trading periods (typically 15 or 30). At the end of each period, the asset paid a stochastic dividend drawn from an objective, publicly disclosed probability distribution. Because the expected dividend payout per period was known with mathematical certainty, the fundamental value of the asset at any period was precisely the expected dividend multiplied by the number of remaining trading periods.

The publication of their findings in the seminal 1988 Econometrica paper, titled “Bubbles, Crashes, and Endogenous Expectations in Experimental Spot Asset Markets,” sent shockwaves through the economics profession. Rather than confirming the backward-induction predictions of rational expectations theory, the laboratory markets consistently generated colossal speculative bubbles. Prices routinely rose to three, four, or five times the underlying fundamental value, accompanied by massive trading volume, before experiencing precipitous, liquidity-starved crashes in the final periods of the market. The paper demonstrated that common knowledge of intrinsic value was entirely insufficient to prevent speculative mania, establishing a foundational empirical anomaly that demanded new behavioral microfoundations.

1.3 Intersection with Behavioral Game Theory and Bounded Rationality

The empirical results generated by Smith, Suchanek, and Williams exposed profound limitations in the classical assumption of unbounded human rationality. In seeking to explain why rational agents would participate in an escalating bubble within a deterministic finite-horizon market, economists were forced to look beyond standard Walrasian equilibrium concepts and engage directly with the principles of bounded rationality first articulated by Herbert A. Simon. Simon had long argued that human decision-makers do not possess the computational capacity, infinite memory, or hyper-rational foresight required to calculate optimal dynamic paths in complex strategic environments; instead, they rely on heuristics, satisficing rules, and simplified procedural models.

In standard game theory, solving a finite-horizon dynamic game requires agents to perform iterative backward induction: analyzing the terminal state $T$, determining optimal actions, rolling the timeline back to $T-1$, and continuing recursively to the initial period $t=0$. In the SSW asset market, this cognitive process should theoretically ensure that because the asset pays nothing after period $T$, its value in period $T$ is merely its single-period expected dividend. Knowing this, rational agents in period $T-1$ should refuse to pay more than twice the expected dividend, compressing all prices down to the deterministic fundamental trajectory. The systemic empirical failure of this backward induction indicated that human traders do not naturally compute solutions recursively from the end of time to the present.

This realization bridged experimental finance with emerging non-equilibrium frameworks in behavioral game theory. Deviations from backward induction are not merely stochastic errors; they follow structured cognitive patterns driven by strategic uncertainty. In a multi-agent environment, an individual trader’s optimal action depends fundamentally on their beliefs regarding the cognitive depth and strategic actions of other traders. Even if an individual trader understands the mathematics of fundamental value perfectly, it may be rational to purchase an overvalued asset if they anticipate that other participants will be willing to purchase it at an even higher price in subsequent periods. Thus, the experimental asset bubble exposed the critical need for cognitive depth models—specifically Level-k reasoning—to explain how strategic uncertainty drives endogenous market dynamics.

2. Theoretical Framework: Rational Expectations Versus Recursive Strategic Thinking

2.1 The Core Architecture of Level-k Reasoning

To rigorously analyze strategic environments where standard Nash equilibrium or rational expectations assumptions fail, behavioral game theorists developed structural models of iterated reasoning, pioneered by Dale Stahl and Paul Wilson (1994, 1995) and Rosemarie Nagel (1995). The fundamental insight of the Level-k framework is that individuals vary systematically in their depth of strategic thinking. The model organizes players into a cognitive hierarchy based on the number of recursive strategic iterations they perform when forming beliefs about counterparty behavior. Instead of assuming mutual best response—the hallmark of Nash equilibrium—Level-k assumes that each agent optimizes against a simplified, non-equilibrium mental model of other participants.

At the base of this cognitive structure sits the Level-0 ($L_0$) agent. The $L_0$ player represents a non-strategic, naive, or automated baseline. In classical game-theoretic treatments such as the beauty contest game, $L_0$ behavior is typically modeled as uniform random play across the strategy space. In asset market microstructures, an $L_0$ trader may be conceptualized as an agent who trades based on naive heuristics, anchoring, pure liquidity needs, or zero-intelligence order placement without any regard for fundamental valuation or strategic counterparty reactions. While $L_0$ agents are rarely assumed to exist in pure form in professional adult populations, their theoretical specification is crucial, as it serves as the cognitive anchor for all higher tiers of reasoning.

Building recursively upon this foundation, a Level-1 ($L_1$) agent assumes that all other market participants are $L_0$ actors. The $L_1$ agent calculates a best response to this non-strategic baseline, exploiting the predictable patterns or naive liquidity provision generated by $L_0$ traders. Progressing upward, a Level-2 ($L_2$) agent believes that the entire market consists of $L_1$ strategic actors; the $L_2$ agent consequently forms beliefs regarding how $L_1$ agents will react to $L_0$ behavior and optimizes accordingly. In general, an $L_k$ agent calculates their best response under the operational assumption that all counterparties belong to tier $L_{k-1}$. A closely related refinement, the Cognitive Hierarchy (CH) model developed by Colin Camerer, Teck-Hua Ho, and Juin-Kuan Chong (2004), relaxes this rigid assumption by modeling $L_k$ players as best-responding to a perceived probability distribution over all lower tiers ($L_0$ through $L_{k-1}$), typically parameterized as a Poisson distribution with mean cognitive depth $tau$.

2.2 Common Knowledge of Rationality (CKR) and Its Laboratory Breakdown

The neoclassical pricing of finite-horizon assets depends upon an extraordinarily stringent epistemic condition: the Common Knowledge of Rationality (CKR). Formally articulated in interactive epistemology by Robert Aumann (1976), an event or property is common knowledge among a group of agents if all agents know it, all agents know that all agents know it, and so on ad infinitum. In the context of the SSW asset market, backward induction requires not merely that Agent $i$ is individually rational (i.e., would never choose a strictly dominated action, such as holding an asset past period $T$ when redemption is zero). It requires that Agent $i$ knows that Agent $j$ is rational, knows that Agent $j$ knows that Agent $k$ is rational, through infinite recursive epistemic layers.

Mathematically, let $\Omega$ represent the state space of the market game, and let $K_i$ be the knowledge operator for trader $i$. Individual rationality implies that for each trader $i$, their choice maximizes utility given their information partition: $R_i \subset \Omega$. Mutual knowledge of rationality of order 1 is defined as:

$$E^1 = \big\cap_{i} R_i$$

Mutual knowledge of order $m$ is defined recursively as $E^m = \big\cap_{i} K_i(E^{m-1})$. Common Knowledge of Rationality corresponds to the infinite intersection:

$$CKR = \big\cap_{m=1}^{\infty} E^m$$

In the controlled laboratory setting of the SSW experiments, individual rationality was demonstrably present: subjects passed comprehensive quizzes proving they understood that dividends were stochastic and finite, and that the asset possessed zero residual value after period $T$. Yet, the market completely failed to enforce fundamental value pricing. The breakdown occurred precisely because mutual knowledge of rationality rarely extended beyond order 1 or order 2. If Trader $A$ believes that Trader $B$ might be willing to pay an inflated price in period $t+1$ (perhaps due to confusion, speculative excitement, or a higher-order belief about Trader $C$), then it becomes entirely rational for Trader $A$ to purchase the overvalued asset in period $t$. Thus, the breakdown of CKR provides a robust mathematical justification for speculative pricing within an experimental economy of individually rational agents.

2.3 Price Expectations and the Speculative Premium Formulation

The strategic decomposition of pricing decisions reveals that an asset’s market price at any period $t$ can be segmented into two fundamentally distinct components: the fundamental value expectation and the speculative premium. Let $D_s$ denote the stochastic dividend paid at the end of period $s in {1, 2, dots, T}$, with expected value $\mathbb{E}[D]$. In a risk-neutral setting with a zero risk-free interest rate, the fundamental value $FV_t$ of the asset at period $t$ (prior to the dividend realization of period $t$) is given by the deterministic sum of all remaining expected dividend disbursements:

$$FV_t = \sum_{s=t}^{T} \mathbb{E}[D_s] = (T – t + 1)\mathbb{E}[D]$$

Under the Rational Expectations Hypothesis, the equilibrium market price $P_t^*$ must equal $FV_t$ for all $t$. However, if an agent forms beliefs using a Level-k cognitive architecture, their subjective valuation of the asset is conditioned not merely on dividend expectations, but on the expected selling price in period $t+1$. The subjective valuation $V_{i,t}$ for trader $i$ of cognitive level $k$ is defined by the Bellman formulation:

$$V_{i,t}^{(k)} = \mathbb{E}[D] + \max \left{ \sum_{s=t+1}^{T} \mathbb{E}[D], \mathbb{E}_{i,t}^{(k)}[P_{t+1}] \right}$$

The second term inside the maximization operator captures the speculative exit option. If the agent’s subjective expectation of the counterparty clearing price in the next period exceeds the remaining dividend stream—that is, if $\mathbb{E}_{i,t}^{(k)}[P_{t+1}] > FV_{t+1}$—the agent will be willing to pay a speculative premium $S_t = P_t – FV_t > 0$. This premium represents a direct formalization of the classical “greater fool” dynamic. Within a Level-k framework, an $L_1$ agent pays an inflated price because they anticipate that $L_0$ agents will continue to buy indiscriminately; an $L_2$ agent pays an even higher price because they project that $L_1$ agents will attempt to front-run the $L_0$ traders. Speculation is transformed from an irrational psychological mania into a structured game of strategic market timing, where every agent seeks to capture capital gains and liquidate their position immediately before the downstream cognitive tiers exhaust their purchasing power.

3. The Canonical Experimental Protocol of Smith, Suchanek, and Williams (1988)

3.1 Asset Life, Trading Periods, and Dividend Structure

The experimental protocol designed by Smith, Suchanek, and Williams in their 1988 investigation established the gold standard methodology for studying laboratory asset pricing. The experiment was structured around a closed, finite-horizon dynamic economy consisting of either 15 or 30 consecutive trading periods. The temporal finiteness of the market was explicitly emphasized to all participants prior to the commencement of trading. Each trading period lasted for a fixed duration, typically between 180 and 240 seconds, during which subjects could actively submit orders and execute transactions. Crucially, the asset carried an absolute terminal expiration: upon the conclusion of period $T$, the asset expired completely worthless, paying no terminal liquidation bonus or redemption premium.

The cash flow generating mechanism of the asset was governed by an explicit, discrete dividend payout schedule. At the end of each trading period $t$, every unit of the asset held in a subject’s inventory generated an independent and identically distributed stochastic dividend $D_t$. In the canonical baseline design, the dividend distribution took one of four equally likely values: $D_t in {0.00, 0.08, 0.28, 0.60}$ experimental currency units, yielding an expected dividend of:

$$\mathbb{E}[D] = 0.25(0.00) + 0.25(0.08) + 0.25(0.28) + 0.25(0.60) = 0.24 \text{ dollars}$$

To guarantee complete transparency and establish common information, SSW implemented rigorous disclosure protocols. Every participant was provided with a comprehensive instruction booklet and an explicit expected-value table. This table calculated the precise fundamental intrinsic value of the asset for every period from $t=1$ to $t=T$. For example, in a 15-period market with $\mathbb{E}[D] = 0.24$, the table explicitly noted that the expected holding value of the asset was $15 \times 0.24 = 3.60$ in period 1, $14 \times 0.24 = 3.36$ in period 2, and steadily decayed at a constant linear rate of $0.24$ per period down to $0.24$ in period 15. The experimenters verbally reviewed this schedule, had subjects complete written comprehension tests, and displayed the expected-value path prominently. Thus, informational ambiguity regarding fundamental value was systematically engineered out of the environment.

3.2 Trading Mechanism: The Continuous Double Auction Institution

To execute trades, SSW implemented a computerized version of the Continuous Double Auction (CDA), an institution historically proven to maximize allocative efficiency in commodity experiments. The double auction operated asynchronously in real time: at any second during an active trading period, any market participant could enter a public bid to buy or a public ask to sell a single unit of the asset. The computerized system maintained a centralized order book governed by strict price-priority rules: the highest active bid (the market best bid) and the lowest active ask (the market best ask) were displayed prominently on all subjects’ computer monitors, forming an open bid-ask spread.

Execution occurred instantaneously whenever a buyer accepted an existing ask (a market buy order) or a seller accepted an existing bid (a market sell order). Alternatively, a transaction was generated whenever a newly submitted bid equaled or exceeded the prevailing best ask, or a newly submitted ask equaled or undercut the prevailing best bid. Upon execution, the transaction price was immediately broadcast to all terminal displays, accompanied by an updated visual graph plotting historical transaction prices against the remaining period timeline. The order book cleared the matched orders and updated the remaining queue.

To maintain strict control over the market’s solvency and financial boundaries, the software enforced hard liquidity and inventory constraints. Short-selling of asset shares was strictly prohibited: an agent could not enter an ask or sell shares if their current inventory balance was zero. Similarly, borrowing cash on margin was entirely disallowed: an agent could not submit a bid if the total cash required to honor that bid exceeded their available liquid cash balance. These boundary conditions ensured that all recorded bids, asks, and transaction prices represented fully backed, solvent commitments, precluding insolvency default risks from driving observed pricing deviations.

3.3 Endowment Structures, Liquidity Ratios, and Payoff Incentives

At the start of the experimental session, participants were initialized with heterogeneous portfolios consisting of tradeable asset shares and liquid working cash balances denominated in experimental currency units. A typical market comprised 9 to 12 subjects divided into three endowment tiers. For instance, Tier 1 subjects might receive an endowment of 3 shares and $10.00 in cash; Tier 2 might receive 2 shares and$20.00 in cash; and Tier 3 might receive 1 share and $30.00 in cash. While individual cash and asset ratios varied across subjects to incentivize trade based on liquidity dispersion, the aggregate market balance was carefully calibrated by the experimenters.

A critical parameter within this design was the aggregate Cash-to-Asset Ratio ($C/A$), defined as the total quantity of liquid cash in the market divided by the total fundamental value of all outstanding shares at period 1:

$$C/A = \frac{\sum_{i=1}^{N} \text{Cash}_i}{\sum_{i=1}^{N} \text{Shares}_i \times FV_1}$$

This ratio measured the systemic buying power present within the laboratory economy. As later theoretical and empirical treatments demonstrated, variations in this initial liquidity endowment exerted a direct, powerful impact on the ceiling of speculative price runs.

To ensure that experimental decisions were non-trivial and reflected genuine economic preferences, the incentives satisfied Vernon Smith’s classic precepts of salient monetary reward. Experimental currency units accumulated through trading profits, dividends, and cash balances were converted into real US dollars at an established, pre-announced exchange rate at the conclusion of the session. Earnings were substantial: subjects could earn two to three times the typical hourly student wage, ensuring active cognitive engagement. The subject pool was initially drawn primarily from undergraduate and graduate students in business, finance, and economics at the University of Arizona and Indiana University, populations possessing formal training in dynamic discounting and mathematical expectation.

4. Empirical Anomalies: The Anatomy of Bubbles and Crashes

4.1 Temporal Trajectory of Price Deviations

The empirical results documented by Smith, Suchanek, and Williams revealed a remarkably consistent, non-linear pricing trajectory that defied the predictions of rational expectations. Rather than tracking the downward-sloping linear fundamental value line, the canonical SSW market unfolded across four distinct, reproducible temporal phases: initial underpricing and price inertia, mid-period parabolic escalation, volume-intensive peak divergence, and a catastrophic terminal crash.

During the opening periods of the market ($t=1$ to $t=3$), transaction prices typically exhibited noticeable inertia. In many sessions, opening transactions cleared at prices below the initial fundamental value ($P_1 < FV_1$). This initial discount reflected subject risk aversion and hesitation regarding how counterparties would price the newly introduced asset. However, as trading progressed into periods 4 through 8, the price dynamics shifted dramatically. Transactions broke away from the fundamental trajectory, entering an explosive, parabolic upward climb. Prices systematically surpassed the fundamental ceiling, rising through the fundamental line and accelerating upward even as the remaining dividend stream continued to shrink deterministically with each elapsed period.

By periods 9 through 12, the bubble reached full maturity. Transaction prices were often observed trading at two to five times the actual fundamental value of the asset. Remarkably, this peak divergence was characterized by massive trading volume; shares changed hands rapidly, demonstrating that liquidity and transaction intensity peaked at the very height of market overvaluation. Finally, as the terminal period approached ($t=13$ to $t=15$), the speculative architecture buckled. Bids dried up instantaneously as buyers realized that no downstream periods remained to offload shares. Prices plunged precipitously toward zero in a matter of minutes, leaving late buyers holding worthless paper and generating severe capital redistribution across the market participants.

4.2 Quantitative Metrics of Market Inefficiency

To rigorously quantify the magnitude, duration, and structural characteristics of asset bubbles across different experimental treatments, experimental economists developed a standardized suite of empirical metrics. The primary measure of bubble size is the Price Amplitude ($A$), which captures the maximum percentage deviation of market prices relative to the underlying fundamental trajectory:

$$A = \max_{t} \left( \frac{\bar{P}_t – FV_t}{FV_1} \right) – \min_{t} \left( \frac{\bar{P}_t – FV_t}{FV_1} \right)$$

where $\bar{P}_t$ denotes the mean transaction price in period $t$, and $FV_1$ is the initial fundamental value at period 1. A related metric, the Normalized Absolute Deviation ($NAD$), aggregates pricing errors across the entire lifespan of the market:

$$NAD = \sum_{t=1}^{T} \frac{|\bar{P}_t – FV_t|}{FV_1}$$

The $NAD$ metric provides an overall index of allocative inefficiency, penalizing prolonged deviations regardless of direction.

In addition to amplitude and deviation, researchers track the Duration ($DU$) of the bubble, defined as the continuous count of trading periods in which the mean transaction price strictly exceeds the fundamental value: $DU = \sum_{t=1}^T \mathbb{I}(\bar{P}_t > FV_t)$. Furthermore, the Share Turnover Rate ($TO$), computed as the total number of shares traded across all periods divided by the total number of unique shares in existence ($TO = \sum_{t=1}^T Q_t / \sum_i \text{Shares}_i$), measures the speculative velocity of the market. In standard SSW environments, turnover routinely exceeded 300% to 500%, confirming that the market was characterized by intense asset churn rather than passive buy-and-hold investing. Finally, the tracking error metric Haessel-$R^2$ evaluates the goodness-of-fit between observed price paths and fundamental trajectories; in canonical bubble sessions, Haessel-$R^2$ values were frequently negative, mathematically demonstrating that fundamental value lacked explanatory power over high-frequency price paths in naive markets.

4.3 Endogenous Market Psychology and Panic Dynamics

The granular microstructure data recorded in SSW markets provides profound insights into the endogenous psychological transitions of market participants. During the mid-period price expansion, the order book reflects an environment of aggressive buying euphoric coordination. Buyers routinely submit market bids that leapfrog the existing order book, aggressively matching standing asks. Sellers, observing this sustained upward drift, progressively raise their asking prices, widening the spread and pulling transaction prices higher. Interviews and post-experiment debriefings indicate that during this phase, traders become fixated on capital gains; the actual dividend yield becomes completely secondary to the realized short-term profit generated by selling shares purchased moments earlier.

The transition from speculative euphoria to systemic collapse is typically triggered by subtle shifts in order-book liquidity. As the market approaches the terminal cliff (e.g., period 12 or 13 in a 15-period game), sophisticated traders quietly begin to liquidate their inventories, attempting to execute an orderly exit. However, the continuous double auction requires counterparty liquidity to clear transactions. When lower-tier buyers exhaust their available cash balances or abruptly realize that only two periods of dividends remain, bid orders vanish from the book. The bid-ask spread widens catastrophically.

The collapse is characterized by pure strategic panic. Sellers, suddenly faced with an empty bid queue, drastically cut their asks to salvage capital. This triggers a downward cascade: traders who had intended to hold for one more period observe the falling asks and dump their holdings simultaneously. The sudden evaporation of market depth generates a liquidity black hole, wherein the asset’s price crashes faster than it climbed. In the final two periods, transaction prices frequently fall below the remaining fundamental value, reflecting severe market revulsion and terminal distress liquidation.

5. Microfoundations: Modeling the SSW Findings via Level-k Game Theory

5.1 Formal Level-0 Behavioral Characterization in Asset Markets

To construct a rigorous Level-k behavioral foundation for the empirical anomalies documented by Smith, Suchanek, and Williams, one must first formally specify the baseline Level-0 ($L_0$) agent within a continuous double auction. In static normal-form games, $L_0$ is almost universally operationalized as uniform random choice across all permissible actions. However, within an asset market, unrestricted uniform bidding would imply submitting bids and asks across completely absurd price ranges, spanning zero to the trader’s total cash balance. While some models utilize zero-intelligence (ZI) traders following budget-constrained uniform random order placement, empirical observations suggest a more structured heuristic specification for asset market $L_0$ agents.

In experimental asset markets, $L_0$ agents can be characterized as naive trend-followers or anchored heuristic traders who exhibit zero strategic anticipation of future counterparty behavior. Let $B_{i,t}$ and $A_{i,t}$ denote the bid and ask submitted by an $L_0$ trader at time $t$. An $L_0$ agent’s reservation price $R_{i,t}^{(0)}$ is determined strictly by a backward-looking anchoring mechanism, combining the most recently observed transaction price $P_{t-1}$ with a stochastic perturbation $\epsilon_{i,t}$ and an immediate cash-spending impulse:

$$R_{i,t}^{(0)} = P_{t-1} + \alpha \Delta P_{t-1} + \epsilon_{i,t}, \quad \epsilon_{i,t} \sim \mathcal{N}(0, \sigma_0^2)$$

where $\alpha ge 0$ captures naive momentum extrapolation, and $\Delta P_{t-1} = P_{t-1} – P_{t-2}$. If an $L_0$ agent holds excess cash, their propensity to submit a bid slightly above the current best bid increases, driven simply by the psychological desire to participate in market activity. Crucially, the $L_0$ trader performs zero backward induction: they do not compute the sum of remaining dividends, nor do they account for the finite terminal boundary $T$. When these non-strategic orders hit the computerized double auction book, the interaction of random order flow with the asymmetrical boundary conditions of the market (cash constraints on bids, inventory constraints on asks) introduces an immediate upward drift. This price noise generates an initial momentum vector that departs from fundamental value.

5.2 Level-1 Optimization and Exploitation of Perceived Trends

A Level-1 ($L_1$) strategic agent possesses cognitive depth sufficient to observe market trends and calculate optimal responses against the perceived pool of $L_0$ traders. The $L_1$ agent does not believe that other participants are engaged in recursive game-theoretic calculations; rather, they view the market as an exogenous, predictable price-generating process populated by naive, trend-chasing $L_0$ participants. Under these subjective beliefs, the $L_1$ agent identifies a clear rent-extraction opportunity.

The $L_1$ trader computes their expected capital gain by modeling one-step-ahead price continuation. Given an observed upward drift in prices during early periods, the $L_1$ agent estimates:

$$\mathbb{E}_{i,t}^{(1)}[P_{t+1}] = P_t + \hat{\mu}_t$$

where $\hat{\mu}_t > 0$ represents the perceived positive price drift generated by naive $L_0$ bidding. The $L_1$ agent’s optimal strategy is to purchase the asset in period $t$ at price $P_t$, even if $P_t > FV_t$, provided that the anticipated selling price in the subsequent period plus the expected dividend exceeds the purchase price:

$$P_t < \mathbb{E}_{i,t}^{(1)}[P_{t+1}] + \mathbb{E}[D] = P_t + \hat{\mu}_t + \mathbb{E}[D]$$

Because the condition $\hat{\mu}_t + \mathbb{E}[D] > 0$ holds mechanically during the expansionary phase, the $L_1$ agent willingly buys overvalued shares. Far from acting irrationally, the $L_1$ trader is optimizing against their model of the world. Their planned exit strategy is explicitly one-step-ahead: they intend to ride the upward bubble curve and dump their inventory onto $L_0$ traders at period $t+1$. However, because all $L_1$ agents simultaneously submit aggressive bids to secure shares for near-term flipping, their aggregate competitive demand amplifies the very price drift they sought to exploit. The influx of $L_1$ capital accelerates the expansion of the bubble, driving market prices further into territory detached from underlying dividends.

5.3 Higher-Level Strategic Calculations (Level-2 to Level-k)

The entry of Level-2 ($L_2$) agents introduces strategic competition between sophisticated tiers. An $L_2$ agent understands that the market contains $L_1$ traders who are actively attempting to front-run the naive $L_0$ population. The $L_2$ agent’s mental model incorporates this dynamic: they realize that $L_1$ agents are planning to exit the market in period $t+1$. To maximize their own expected return, the $L_2$ agent must execute their liquidation strategy precisely one step ahead of the $L_1$ wave. The recursive logic cascades through higher cognitive tiers: an $L_k$ agent calculates that they must exit exactly one period before the $L_{k-1}$ cohort plans to liquidate.

This dynamic can be formalized as an optimal stopping problem across discrete cognitive strata. Let $\tau_k^*$ denote the planned market exit period for an agent of level $k$. If an $L_1$ agent projects that the bubble will peak at period $T-1$ (just before the terminal expiration $T$), their exit rule is $\tau_1^* = T-1$. An $L_2$ agent, anticipating this liquidation, sets $\tau_2^* = T-2$. More generally, the planned exit period unravels according to the recursive relationship:

$$\tau_k^* = \max{1, T – k}$$

The stability of the experimental market depends directly upon the cognitive distribution of the participating population. If the population consists overwhelmingly of $L_0$ and $L_1$ agents, prices will inflate unchecked until hitting the hard terminal boundary at period $T$. If, however, the market possesses a sufficient proportion of higher-order strategic agents ($L_2, L_3, dots, L_\infty$), the anticipated exit period unravels backward in time. When agents possessing cognitive depth $k to \infty$ enter the market, they realize that no downstream buyers will exist to absorb the overvalued shares at the terminal margin; the unraveling collapses all speculative premia, forcing initial prices to converge to the fundamental baseline $FV_t$. The persistent laboratory bubbles observed by Smith, Suchanek, and Williams demonstrate conclusively that naive laboratory subject pools possess a low mean cognitive depth (typically estimated between $tau = 1.2$ and $tau = 1.8$), preventing the backward unraveling from extinguishing the bubble.

6. Cognitive Asymmetries and Strategic Uncertainty in Double Auctions

6.1 Strategic Uncertainty Versus Fundamental Uncertainty

A crucial conceptual achievement of experimental economics is the rigorous decoupling of fundamental uncertainty from strategic uncertainty. Fundamental uncertainty pertains strictly to the exogenous state of nature: the stochastic roll of the dice that determines whether an asset pays a dividend of $0.00$ or $0.60$ at the end of a given period. In the SSW experimental paradigm, fundamental uncertainty is completely transparent and well-behaved; the probability distribution is stationary, discrete, and universally known, rendering the calculation of mathematical expected values trivial for any individual endowed with basic numeracy.

In sharp contrast, strategic uncertainty is endogenous; it arises from an agent’s uncertainty regarding the behavioral actions, cognitive depth, and beliefs of other human participants within the double auction. A trader may know with mathematical certainty that $FV_t = 2.40$, yet face intense strategic uncertainty regarding whether counterparty Trader $j$ will submit a bid of $1.50$, $2.40$, or $4.80$. Because an individual’s financial payoff in a double auction depends critically upon the future market clearing prices generated by counterparties, strategic uncertainty dominates decision-making.

This strategic uncertainty prevents the realization of Common Knowledge of Rationality, even when every single trader in the room is individually rational. If an agent assigns a non-zero subjective probability to the hypothesis that some counterparties are trading under non-fundamental heuristics, standard subjective expected utility theory dictates that the agent should adjust their bidding behavior away from fundamental value. Furthermore, this strategic uncertainty compounds exponentially across consecutive temporal periods. In a 15-period game, evaluating an action in period 1 requires projecting the distribution of counterparty strategic types across 14 subsequent iterations of order-book matching, creating an epistemic environment where rational backward induction completely dissolves.

6.2 Iterated Reasoning Constraints and Cognitive Load

The failure of human subjects to perform higher-order strategic thinking in laboratory asset markets is deeply rooted in biological and cognitive architecture. Iterative recursive reasoning imposes severe demands on human working memory and executive processing capacity. In a Level-k calculation, computing the optimal action requires an agent to hold multiple nested counterfactual mental models simultaneously: “I think that you think that she thinks…” Cognitive psychological research demonstrates that the human prefrontal cortex experiences severe performance degradation when processing recursive mental representations beyond two or three iterative steps.

This biological constraint is amplified by the intense time pressure and sensory stimulation inherent in the continuous double auction institution. Unlike static normal-form games played with pen and paper over generous time limits, a computerized double auction is a high-speed, asynchronous, real-time environment. Traded prices, bids, asks, and order cancellations flash across the computer terminal continuously. Participants must monitor the order book, calculate remaining cash balances, verify inventory constraints, track elapsed time, and execute keyboard entries within fractions of a second.

Empirical studies measuring response times and cognitive load in financial auction environments confirm this mechanism. When cognitive load is elevated—either by speeding up the trading clock or introducing complex multi-asset portfolios—the strategic depth of participants degrades significantly, shifting the population distribution downward toward $L_0$ and $L_1$ behaviors. Conversely, neuroeconomic and eye-tracking studies demonstrate that individuals who successfully execute higher-order strategic thinking exhibit extended fixation times on historical price queues and fundamental value schedules, expending substantial mental effort to suppress the immediate emotional impulse to chase market momentum.

6.3 The Role of Overconfidence and Epistemic Fallacies

The cognitive dynamics within experimental asset markets are further distorted by pervasive behavioral biases, most notably overconfidence and the “better-than-average” epistemic fallacy. In post-experiment surveys across dozens of SSW replications, researchers have documented a systematic asymmetry: an overwhelming majority of participants report believing that their own cognitive capabilities, market timing skills, and strategic sophistication are strictly superior to the median participant in the room. Traders openly acknowledge that the market is overvalued, but express supreme confidence that they will successfully liquidate their inventories moments before the eventual crash occurs.

This widespread cognitive bias directly drives speculative overbidding within a Level-k framework. An agent who classifies themselves as an $L_2$ or $L_3$ strategist consistently underestimates the number of equally or more sophisticated agents operating in the market. Consequently, they delay their planned exit period, mistakenly assuming that they possess exclusive front-running capability. Furthermore, this epistemic fallacy is reinforced by transient confirmation bias during the bubble’s upward trajectory. As the bubble expands through periods 5 to 10, traders who bought overvalued shares observe rising paper wealth and realized capital gains. This short-term positive feedback loop validates their speculative behavior, blinding them to the mathematical inevitability of the terminal boundary.

The psychological shock that occurs during the crash phase is the direct empirical consequence of this widespread epistemic error. When the bubble reaches its peak, dozens of traders simultaneously initiate their liquidation orders, under the erroneous assumption that a deep pool of lower-tier buyers remains available to absorb their positions. The sudden discovery that counterparty liquidity has evaporated shatters the illusion of individual strategic superiority, triggering panic selling and exposing the fundamental structural flaw of the speculative hierarchy.

7. The Impact of Experience and Market Repetition on Cognitive Depth

7.1 Single-Session Market Convergence and Learning Dynamics

One of the most remarkable empirical discoveries documented by Vernon Smith, Gerry Suchanek, and Arlington Williams in their 1988 paper was the transformative impact of market repetition and subject experience. To evaluate whether the observed speculative bubbles were permanent features of human market interaction or merely transient non-equilibrium learning artifacts, SSW designed an experimental protocol that brought identical cohorts of human subjects back to the laboratory for repeated, identical market sessions.

In a standard three-session design, a cohort of naive subjects trades a complete 15-period market in Round 1, generating the canonical massive bubble-and-crash profile. Several days or weeks later, the exact same cohort is reassembled to trade the identical market structure in Round 2, with identical dividend probabilities, endowments, and rules. In Round 2, the bubble amplitude ($A$) and normalized absolute deviation ($NAD$) are typically dampened by 50% to 70%; peak prices occur earlier, and the market crash is far less severe. When the identical cohort returns for Round 3, the speculative bubble almost completely disappears. Transaction prices in Round 3 track the fundamental value line ($FV_t$) with remarkable fidelity from period 1 through period 15, achieving the precise rational expectations equilibrium predicted by neoclassical economic theory.

Within the analytical architecture of Level-k game theory, this experiential convergence represents an endogenous upward migration of the aggregate cognitive depth parameter $tau$. In Round 1, subjects enter with uncalibrated, heterogeneous beliefs, resulting in a low effective cognitive level dominated by $L_0$ noise and $L_1$ momentum chasing. However, surviving the catastrophic Round 1 crash provides an undeniable, salient learning signal. Traders directly experience the financial penalty of holding overvalued inventory across the terminal boundary. In Round 2, subjects incorporate this structural knowledge, elevating their strategic depth to anticipate the crash earlier. By Round 3, the shared public experience transforms the epistemic environment: subjects not only understand backward induction individually, but they also possess common knowledge that their counterparties have experienced the crash. Common knowledge of rationality is thus endogenously generated through historical market trauma, driving the Level-k hierarchy to collapse into the neoclassical $L_\infty$ equilibrium.

7.2 Transfer of Experience and Subject Pool Heterogeneity

While identical-cohort repetition reliably extinguishes asset bubbles, experimental researchers quickly discovered that this acquired market rationality is remarkably fragile and context-dependent. When experienced subjects who have successfully achieved rational convergence in Round 3 are placed into a modified market environment—such as altering the dividend probability distribution, changing the length of the finite horizon, or injecting a substantial quantity of newly minted liquidity—speculative bubbles frequently re-emerge.

This fragility was systematically documented in studies examining the transfer of experience across institutional environments. The cognitive learning acquired by human traders is largely procedural rather than abstractly deductive; subjects do not necessarily learn the mathematical principle of backward induction across all dynamic games, but rather memorize the historical trajectory and timing of the specific market they previously traversed. When structural parameters change, strategic uncertainty returns, resetting the subjective Level-k beliefs of participants and allowing speculative premia to re-inflate.

Furthermore, extensive research exploring subject pool heterogeneity has yielded surprising results regarding the role of professional training. Naive undergraduate students, graduate students, and general corporate employees produce broadly comparable bubble profiles in Round 1 markets. However, experiments conducted with professional financial market traders, investment bankers, and asset managers—documented by researchers such as Utz Weitzel, Christoph Huber, and Michael Kirchler—revealed that professionals often generate larger, more volatile speculative bubbles than naive student cohorts. debriefings and order-flow analysis indicate that professional traders possess highly developed $L_1$ and $L_2$ instincts: they exhibit extreme confidence in their ability to out-trade and front-run their peers, aggressively buying overvalued assets with the deliberate intention of riding the bubble to its absolute peak. The professional trading culture’s institutional emphasis on momentum and short-term liquidity timing actually exacerbates Level-k strategic friction, preventing spontaneous price convergence in the absence of explicit institutional safeguards.

7.3 Level-k Evolution Through Dynamic Iterated Play

To mathematically characterize the evolution of strategic thinking across repeated market sessions, behavioral econometricians utilize dynamic reinforcement learning and Experience-Weighted Attraction (EWA) models. Developed by Colin Camerer and Teck-Hua Ho (1999), the EWA framework synthesizes standard choice reinforcement with belief-based fictitious play, tracking how agents update their subjective probability distributions over counterparty cognitive types as empirical outcomes unfold.

Let $N_i^{(k)}(t)$ denote the attraction of trader $i$ to adopting strategic rule $k$ at trading round $t$. In a multi-session market, the updating of cognitive attractions is governed by the recursive equation:

$$N_i^{(k)}(t) = \frac{\phi N_i^{(k)}(t-1) + [\delta + (1-\delta)\mathbb{I}(s_i(t) = s^{(k)})]\pi(s^{(k)}, s_{-i}(t))}{\phi(1-\kappa) + 1}$$

where $phi$ represents an experience discount parameter, $\delta$ is an imagination weight assigned to foregone counterfactual payoffs, and $\pi(\cdot)$ is the realized financial payoff. In Round 1, lower-level rules ($L_1, L_2$) yield high intermediate payoffs during the bubble’s ascent, reinforcing speculative bidding. However, during the terminal crash, agents holding high-tier rules ($L_{ge 3}$) or those who liquidated early avoid catastrophic terminal losses, earning vastly superior net session payoffs.

As these terminal payoffs are processed between sessions, the structural parameters of the population’s cognitive distribution shift. Econometric estimations of the Poisson Cognitive Hierarchy parameter $tau$ demonstrate an upward trajectory: $\hat{\tau}$ typically rises from approximately $1.3$ in naive Round 1 sessions to over $4.5$ in experienced Round 3 sessions. At $tau ge 4.5$, the proportion of agents performing four or more recursive steps of strategic thinking becomes sufficiently large to suppress any initial $L_0$ upward price drift, anchoring market clearing prices to the fundamental dividend trajectory from the opening bell.

8. Liquidity, Speculation, and Institutional Design Variations

8.1 The Liquidity Hypothesis: Cash-to-Asset Ratio Manipulations

While Level-k reasoning provides the cognitive architecture for speculative behavior, cognitive capacity alone cannot generate transactions without purchasing power. The physical expansion of an asset bubble requires liquidity. The profound relationship between market buying power and speculative amplitude was systematically unraveled through the pioneering work of Gunduz Caginalp, David Porter, and Vernon Smith (1998, 2001) in their formulation and experimental validation of the Liquidity Hypothesis.

In their experimental designs, Caginalp, Porter, and Smith systematically manipulated the initial aggregate Cash-to-Asset ratio ($C/A$) while holding the fundamental expected dividend schedule completely constant. Their empirical findings established an unambiguous, statistically robust positive correlation: the peak amplitude, duration, and normalized absolute deviation of speculative bubbles are direct increasing functions of aggregate initial market liquidity. When experimental markets were initialized with high cash balances relative to share endowments ($C/A gg 1$), massive bubbles inevitably formed, driving prices to astronomical multiples of intrinsic value.

The Level-k framework provides a clean microeconomic explanation for this empirical phenomenon. High liquidity acts as an institutional subsidy for lower-order strategic thinking. When cash is abundant, $L_0$ and $L_1$ traders face minimal budget constraints; they can repeatedly submit higher bids, absorb selling asks, and maintain upward momentum without exhausting their balances. This abundance of liquidity validates the expectations of $L_1$ and $L_2$ front-runners, extending the duration of the profitable expansionary phase. Conversely, when the experimenters engineered low-liquidity environments ($C/A ll 1$), speculative bubbles were strangled at inception. Even if naive traders harbored non-fundamental speculative desires, the physical absence of liquid cash balances established a hard institutional boundary condition, forcing transaction prices to remain anchored near fundamental value regardless of the participants’ cognitive limitations.

8.2 Short-Selling and Margin Trading Mechanisms

In standard financial theory, the absence of short-selling is frequently cited as the primary market imperfection responsible for asset overvaluation. According to Edward Miller’s classic hypothesis (1977), when short-selling is prohibited, pessimistic or rational investors are sidelined, leaving market prices to be set exclusively by the most optimistic participants. To evaluate whether this institutional friction was the root cause of laboratory bubbles, experimental economists introduced explicit short-selling mechanisms into the canonical SSW framework.

In comprehensive experimental treatments conducted by Ernan Haruvy and Charles Noussair (2006), subjects were permitted to short-sell shares under varying institutional configurations, including unconstrained shorting, margin-account borrowing, and borrowing subject to strict capacity limits. The empirical results contradicted neoclassical predictions: the introduction of short-selling mechanisms failed to eliminate speculative bubbles. While shorting slightly dampened peak bubble amplitude in some sessions, large and sustained price runs persisted.

The persistence of bubbles in the presence of short-selling is directly explained by the strategic risks faced by higher-level cognitive agents ($L_k$) in a double auction. An agent who correctly calculates fundamental value ($L_\infty$) and initiates a short position at period 5 (when the asset is already trading above fundamental value) faces severe short-squeeze risk. If the market is dominated by $L_0$ trend-chasers and $L_1$ front-runners, prices will continue to climb aggressively through periods 6, 7, and 8. The rational short-seller faces catastrophic margin calls or paper losses before the bubble eventually crashes at period 13. Because short-sellers face theoretically unbounded downside risk while waiting for the terminal convergence, the strategic uncertainty surrounding when the crash will occur makes shorting highly hazardous. Thus, Level-k friction prevents rational agents from aggressively shorting the market, neutralizing the price-correcting power of the short-sale institution.

8.3 Futures Markets and Derivative Instruments

Another profound institutional variation explored in experimental literature is the introduction of concurrent derivative institutions, specifically forward and futures markets. In an influential study by David Porter and Vernon Smith (1995), experimental markets were structured to allow simultaneous trading in both the standard spot asset market and a suite of parallel futures contracts expiring at various intermediate periods (e.g., periods 5, 10, and 15).

Theoretically, the existence of a complete set of futures contracts dramatically simplifies backward induction. A futures contract expiring at period $T$ trades with direct, transparent reference to the terminal redemption value of zero. By establishing explicit market prices for future delivery dates, the futures institution provides public, decentralized price discovery regarding the future trajectory of the asset, theoretically anchoring expectations and extinguishing the speculative premium in the spot market.

The empirical data revealed that parallel futures markets significantly dampened, but did not completely eliminate, spot asset bubbles. The introduction of futures trading altered the strategic landscape by providing an institutional coordination device for higher-order agents. Level-2 and Level-3 traders could observe futures prices trading closer to fundamental values, utilizing this information to discipline their spot market bids. However, the spot bubble continued to exhibit residual speculative life whenever sufficient liquidity existed to support short-term spot momentum trading. The experiment demonstrated that while derivative architectures improve intertemporal coordination and accelerate strategic backward induction, cognitive bounded rationality continues to introduce pricing friction across interconnected institutional layers.

8.4 Information Displays and Fundamental Transparency

A natural hypothesis advanced by skeptics of the early SSW findings was that laboratory bubbles were driven by simple informational obscurity: subjects perhaps found it difficult to track the declining fundamental value line while actively monitoring the fast-paced continuous double auction. To subject this explanation to a definitive empirical test, experimenters developed software interfaces that provided persistent, dynamic, real-time fundamental transparency displays.

In these modern protocols, the computerized trading interface displays a dedicated on-screen graphic that updates every second. The software explicitly plots the exact expected fundamental value line directly adjacent to the order book, accompanied by prominent digital readouts displaying: “Current Period Expected Fundamental Value: $X.XX.” Furthermore, before entering an order, subjects are presented with confirmation dialogues warning them if their proposed bid exceeds the remaining expected dividend stream.

Remarkably, experimental results—such as those documented by Michael Kirchler, Jürgen Huber, and Thomas Stöckl (2012)—demonstrate that real-time fundamental transparency displays alone are insufficient to prevent speculative bubbles among naive subject cohorts. Even when subjects are continuously reminded of the true fundamental value on their screens, transaction prices routinely soar into major bubbles. Post-session debriefings reveal the underlying cognitive mechanism: participants openly acknowledge seeing the fundamental value display, but choose to ignore it because they believe that other market participants will ignore it. This empirical finding provides decisive proof of epistemic decoupling: asset bubbles in experimental markets are not the product of individual computational ignorance, but rather the structural manifestation of strategic uncertainty and recursive Level-k coordination games.

9. Methodological Extensions and Alternative Behavioral Paradigms

9.1 The Cognitive Hierarchy Model Versus Classical Level-k

While the standard Level-k model provides an intuitive framework for analyzing iterated strategic thinking, it possesses structural rigidities that have led behavioral economists to develop refined alternatives, most notably the Cognitive Hierarchy (CH) model of Colin Camerer, Teck-Hua Ho, and Juin-Kuan Chong (2004). The critical difference between the two formulations lies in how higher-tier agents conceptualize the cognitive composition of the market.

In classical Level-k theory, an $L_k$ agent possesses an extremely dogmatic mental model: they assume that 100% of other market participants belong specifically to tier $L_{k-1}$. For example, an $L_2$ agent assumes everyone else is an $L_1$ agent, completely ignoring the existence of $L_0$ random traders or other $L_2$ peers. This assumption becomes increasingly problematic at higher tiers, as it implies that sophisticated agents hold completely unrealistic beliefs regarding the homogeneity of counterparty populations. In contrast, the Cognitive Hierarchy model assumes that an $L_k$ agent recognizes that the market contains a heterogeneous mixture of all lower cognitive strata ($L_0, L_1, dots, L_{k-1}$).

In the CH framework, the true distribution of cognitive types within the population is parameterized by a single-parameter Poisson distribution $f(k) = \frac{e^{-\tau}\tau^k}{k!}$, where $tau$ represents the mean cognitive depth of the population. An agent of level $k$ does not know the true distribution, but correctly forms normalized conditional beliefs over all tiers strictly below them:

$$g_k(h) = \frac{f(h)}{\sum_{j=0}^{k-1} f(j)}, \quad \forall h < k$$

The CH model provides a substantially superior empirical fit when estimating structural parameters from experimental asset market transaction data. Because CH agents optimize against a blend of lower-tier behaviors, the model smoothly accounts for the long tail of market participants who fail to anticipate aggregate selling pressure, accurately reproducing the gradual rounding of bubble peaks and the asynchronous timing of subject liquidation orders.

9.2 Quantal Response Equilibrium and Noisy Best Response

An alternative behavioral approach to modeling deviations from neoclassical equilibrium in dynamic markets is the framework of Quantal Response Equilibrium (QRE), developed by Richard McKelvey and Thomas Palfrey (1995, 1998). Unlike deterministic game-theoretic models that assume agents always execute an exact mathematical best response, QRE incorporates structural execution error and bounded perceptual acuity into human decision-making, utilizing a statistical framework of noisy best response.

In a dynamic market game, an agent calculating the expected payoffs of various actions does not select the payoff-maximizing choice with probability 1. Instead, the probability of choosing action $a_i$ from action space $\mathcal{A}$ is governed by a smooth, stochastic choice function, typically the logit (softmax) distribution:

$$Pr(a_i) = \frac{\exp(\lambda \mathbb{E}[U(a_i)])}{\sum_{a_j in \mathcal{A}} \exp(\lambda \mathbb{E}[U(a_j)])}$$

where $\lambda in [0, \infty)$ is a precision parameter representing the rationality or cognitive sensitivity of the market. When $lambda to 0$, trading choices degenerate into uniform random noise ($L_0$ behavior); as $\lambda to \infty$, the system converges to the standard Nash or backward-induction equilibrium.

When applied to experimental asset markets through Agent-Based Quantal Response frameworks, QRE demonstrates how stochastic error propagation across multiple trading periods can mimic the emergence of speculative bubbles. Even if human agents possess relatively high precision ($lambda$), small perceptual errors in period 1 propagate forward through the dynamic game tree. As transaction prices drift slightly upward due to stochastic order execution, downstream agents condition their expected utility calculations on these distorted historical outcomes. Furthermore, economists have successfully synthesized these frameworks into Noisy Level-k and Quantal Cognitive Hierarchy models, which combine hierarchical strategic depth with logit execution error, providing an exceptionally robust econometric framework for capturing both strategic miscoordination and high-frequency order-book noise.

9.3 Agent-Based Modeling of Experimental Asset Markets

The complex, non-linear interactions between heterogeneous cognitive tiers and the microstructure of continuous double auctions have inspired extensive computational investigations utilizing Agent-Based Models (ABMs). Pioneered by researchers at the Santa Fe Institute and expanded by computational financial economists, agent-based architectures allow researchers to simulate the exact experimental protocol of Smith, Suchanek, and Williams in silico.

In a canonical asset market ABM, thousands of software agents are instantiated with heterogeneous parameters governing their cognitive depth ($k$), cash endowments, and behavioral heuristics. Artificial $L_0$ agents execute budget-constrained random walks or simple trend extrapolation; artificial $L_1$ agents run real-time regression models on recent price ticks to optimize short-term order placement; and artificial $L_2$ agents continuously evaluate the aggregate liquidity balance of the simulated order book to time their market exits. These agents interact asynchronously through a simulated continuous double auction matching engine identical to the laboratory software used with human subjects.

These computational simulations successfully replicate the macroeconomic stylized facts documented in human laboratory experiments: fat-tailed return distributions, volatility clustering, persistent price bubbles, and sudden, liquidity-starved crashes. Furthermore, ABM sensitivity analyses have illuminated the exact structural tipping points of market stability. Simulations reveal that market stability does not require 100% of agents to be hyper-rational; rather, there exists a critical percolation threshold. If the proportion of strategic agents possessing cognitive depth $k ge 2$ exceeds approximately 30% to 40% of the aggregate market population, their combined inventory management and liquidity withdrawal are sufficient to prevent the initial $L_0$ price noise from escalating into a full-scale parabolic bubble.

10. Critical Debates and Methodological Controversies

10.1 Confusion Versus Speculation: The Ball and Holt Debate

Following the publication of the 1988 SSW paper, a heated methodological debate emerged within experimental economics regarding the true psychological drivers of laboratory asset bubbles. In an influential critique, Sheryl Ball and Charles Holt (1998) argued that the observed price bubbles were not primarily driven by sophisticated strategic speculation or “greater fool” dynamics, but were instead mundane artifacts of subject confusion.

Ball and Holt contended that novice human subjects, placed into an unfamiliar computerized double auction with unfamiliar financial accounting terminology, simply failed to grasp the mechanics of declining fundamental value. They suggested that subjects confused the asset’s expected dividend in a single period with its total remaining holding value, or failed to realize that the asset expired completely worthless at period $T$. To support this critique, Ball and Holt demonstrated that when subjects were provided with extensive algorithmic accounting training, or when market instructions were simplified to remove complex probabilistic dividend schedules, the magnitude of the speculative bubbles was significantly reduced.

This critique provoked extensive counter-investigations that definitively demonstrated that while confusion can exacerbate price variance, intentional speculation remains the primary engine of large-scale asset bubbles. In an elegant experimental design, Vernon Smith and his collaborators tested subjects who had undergone rigorous accounting training and demonstrated flawless mathematical comprehension of fundamental decay. When these trained, completely unconfused subjects were placed into markets alongside naive traders, they did not enforce fundamental value; instead, they aggressively bought overvalued shares during the early periods, explicitly seeking to extract speculative capital gains from their less sophisticated peers. These results confirmed that intentional Level-k strategic exploitation, rather than mere accounting confusion, is the driving microfoundation of persistent asset bubbles.

10.2 Ecological Validity and the Laboratory-to-Field Transfer

A perennial methodological challenge confronting experimental economics centers on the issue of ecological validity: to what extent can the behavior of 10 or 12 undergraduate students trading tokens for 2 hours in a university computer laboratory inform our understanding of multi-trillion-dollar global financial markets? Mainstream financial economists initially dismissed the SSW findings as idiosyncratic laboratory artifacts, arguing that real-world financial markets possess institutional features that naturally prevent such anomalies.

Traditional finance critics pointed to several key structural disparities between the laboratory and field markets. In real-world economies, assets exist within a vast, interconnected macroeconomic ecosystem characterized by infinite or indefinite horizons, continuous alternative investment opportunities (such as risk-free government bonds), deep corporate capital pools, highly sophisticated institutional asset managers, and extensive legal disclosures. Furthermore, in field markets, professional arbitrageurs manage billions of dollars, theoretically providing the capital depth required to crush any speculative price bubble at its inception.

In response, Vernon Smith and contemporary behavioral economists mounted a vigorous defense of the laboratory method. Smith emphasized that the laboratory double auction was never intended to be a photorealistic replica of Wall Street; rather, it was designed as a rigorous empirical testbed for theoretical boundary conditions. Neoclassical finance theory claimed that market efficiency and backward induction were universal mathematical truths that held in any competitive market operating under common knowledge of fundamental value. By demonstrating that efficiency fails completely in the simplest imaginable setting—a closed economy with identical assets, deterministic fundamentals, zero transaction costs, and complete transparency—the laboratory experiments decisively proved that the theoretical conditions required to guarantee rational pricing do not spontaneously hold among human decision-makers. Furthermore, the empirical trajectory of experimental bubbles—characterized by early inertia, parabolic momentum runs, turnover spikes, and sudden liquidity freezes—mirrors historical macroeconomic episodes, such as the 1720 South Sea Bubble and the late-1990s Dot-com bubble, with astonishing structural fidelity.

10.3 Elicitation of Subject Beliefs and Identification of Level-k Types

As the Level-k paradigm gained prominence as the leading theoretical explanation for experimental asset dynamics, experimentalists encountered a profound econometric challenge: the identification problem. In a standard continuous double auction, researchers observe only market actions—bids, asks, and transaction prices. Inferring an agent’s internal cognitive depth ($k$) strictly from their market transactions is econometrically fraught, as an identical observed bid could theoretically be generated by an $L_0$ agent making a random error, an $L_1$ agent pursuing a naive trend, an $L_2$ agent executing an intentional speculative front-run, or a risk-loving agent with idiosyncratic utility preferences.

To overcome this identification hurdle, modern experimental designs incorporate sophisticated incentivized belief elicitation protocols. Prior to each trading period, the market clock pauses, and subjects are required to submit private numerical forecasts predicting the average transaction price for the upcoming period and future periods up to the terminal horizon $T$. To ensure that subjects reveal their true subjective expectations, their forecasts are incentivized using strictly proper scoring rules, such as the Quadratic Scoring Rule (QSR):

$$S(f_t, P_t) = \alpha – \beta (f_t – P_t)^2$$

where $f_t$ is the elicited price forecast, $P_t$ is the realized transaction price, and $\alpha, \beta > 0$ are calibration constants.

The integration of incentivized belief elicitation has yielded critical econometric breakthroughs. In seminal studies by Ernan Haruvy, Yaron Lahav, and Charles Noussair (2007), analysis of elicited forecasts confirmed that during the early and middle periods of an asset bubble, subjects’ subjective expectations of future prices are aggressively adaptive and non-fundamental. Crucially, the data demonstrated that subjects do not mistakenly believe that the fundamental value is rising; rather, they accurately forecast that market prices will trade far above fundamental value. This empirical separation of fundamental value beliefs from market price expectations provides direct, irrefutable verification of Level-k cognitive modeling: market participants knowingly buy overvalued assets because their higher-order expectations anticipate profitable downstream liquidation opportunities.

11. Policy Implications and Institutional Design in Modern Financial Systems

11.1 Circuit Breakers, Price Limits, and Trading Halts

The microeconomic insights generated by the SSW paradigm and Level-k reasoning carry profound implications for the regulatory architecture of modern financial exchanges. In global financial markets, regulatory authorities frequently implement institutional circuit breakers, price collars, and mandatory trading halts to curb extreme market volatility and prevent catastrophic speculative runs. In the United States, for example, the Securities and Exchange Commission (SEC) enforces market-wide circuit breakers alongside specific Limit-Up/Limit-Down (LULD) bands for individual securities.

Experimental asset markets have served as a critical testing laboratory for evaluating the efficacy of these regulatory interventions. When experimenters implement hard price collars (e.g., prohibiting any transaction from clearing at a price more than 10% above or below the preceding period’s closing price), the results reveal complex behavioral distortions. While price collars mechanically limit the instantaneous velocity of price runs, they frequently generate a pernicious behavioral anomaly known as the “magnet effect.” As prices approach the upper price collar, traders anticipating that trading will soon be halted accelerate their buying activity, prematurely pulling market prices directly into the upper boundary.

However, temporary trading halts—wherein the market is completely paused for an extended duration without order execution—exhibit distinct cognitive benefits when analyzed through a Level-k framework. In experimental treatments evaluating volatility halts, pausing the trading clock disrupts high-frequency heuristic momentum chasing ($L_0$ and $L_1$ behavior). The quiet interval lowers cognitive load, dissipates immediate emotional arousal, and provides participants with the time required to consult fundamental value displays and re-engage higher-order cognitive processing ($L_k ge 2$). Consequently, markets resuming after structured trading halts frequently exhibit sharp downward corrections toward fundamental value, demonstrating that regulatory institutional design can actively reshape the cognitive depth of market participants.

11.2 Central Bank Liquidity Injections and Asset Inflation

The profound empirical link established by Vernon Smith and his colleagues between aggregate Cash-to-Asset ratios ($C/A$) and speculative bubble amplitude provides an invaluable analytical perspective on real-world macroeconomic monetary policy. Following the 2008 Global Financial Crisis and the 2020 COVID-19 pandemic, the world’s major central banks—including the Federal Reserve, the European Central Bank, and the Bank of Japan—embarked on unprecedented monetary expansions, injecting trillions of dollars of high-powered liquidity into the banking system through Quantitative Easing (QE) and near-zero interest rate policies.

While standard macroeconomic models framed these interventions strictly in terms of lowering borrowing costs and stimulating aggregate demand, the experimental literature illustrates the direct, unintended side-effects of systemic liquidity expansion on asset market microstructures. In the laboratory, whenever the cash-to-asset endowment ratio is dramatically expanded, the aggregate market purchasing power acts as an open-ended subsidy for speculative overbidding. Abundant market liquidity suppresses the natural disciplinary constraint that forces prices to converge to intrinsic cash-flow values, enabling lower-order Level-k momentum strategies to thrive unchecked across real estate, equity, and corporate debt markets.

Furthermore, experimental simulations of monetary tightening provide stark warnings regarding the terminal phase of liquidity-driven expansions. In experimental markets where liquidity is abruptly contracted midway through an active trading session, the resulting market correction is rarely smooth or orderly. Because the underlying market structure has adapted to continuous liquidity absorption, the sudden withdrawal of buying power triggers an instantaneous collapse in order-book depth. Bids vanish, spreads blow out, and market prices plunge into severe liquidity-starved crashes, directly mirroring the acute financial fragility observed in real-world debt and equity markets whenever central banks attempt to unwind balance sheets and initiate quantitative tightening.

11.3 Cryptocurrency Markets and Retail Investor Bubbles

Perhaps the most striking contemporary real-world manifestation of the Smith, Suchanek, and Williams experimental architecture is observed in modern decentralized finance and cryptocurrency markets. Assets such as Bitcoin, Ethereum, and non-fungible tokens (NFTs), alongside algorithmic meme-coins, present a market environment that matches the theoretical boundary conditions of experimental asset economics to a startling degree.

Unlike traditional equities or corporate bonds, many digital assets possess zero deterministic cash-flow yields, no corporate balance sheets, and no terminal liquidation redemption guarantees. In the terminology of asset pricing, their fundamental intrinsic value ($FV_t$) is devoid of exogenous cash-flow anchors. In the complete absence of a cash-flow anchor, the market price of a cryptocurrency is determined exclusively by endogenous strategic expectations—a pure, unadulterated Level-k coordination game. Market participants openly acknowledge that a digital token has no fundamental dividend yield; the investment thesis is predicated entirely upon the anticipated future behavior of downstream buyers.

In this digital ecosystem, social media platforms (such as Twitter, Reddit, and Telegram) act as real-time, global coordination mechanisms that synchronize Level-1 and Level-2 trading strategies. Retail investors coordinate buying sprees to engineer viral momentum runs, explicitly relying on the “greater fool” formulation modeled in behavioral game theory. When mapped against the quantitative metrics of experimental finance, the historical price charts of major cryptocurrency speculative episodes exhibit Normalized Absolute Deviations ($NAD$), turnover velocities, and parabolic amplitude peaks that mirror the canonical laboratory charts recorded by Vernon Smith, Gerry Suchanek, and Arlington Williams in 1988 with uncanny structural precision.

12. Synthesis and Future Directions in Behavioral Market Design

12.1 The Legacy of the 1988 Econometrica Milestone

The publication of “Bubbles, Crashes, and Endogenous Expectations in Experimental Spot Asset Markets” by Vernon L. Smith, Gerry L. Suchanek, and Arlington W. Williams in 1988 stands as a permanent watershed moment in the history of economic science. By successfully bringing dynamic asset pricing into the controlled laboratory environment, SSW dismantled the longstanding theoretical dogma that competitive double auctions operating under full information automatically guarantee rational expectations pricing. Their work proved conclusively that market efficiency is not an innate property of decentralized human trade, but rather a contingent outcome that depends delicately upon institutional design, experience, and the cognitive architecture of participants.

The enduring impact of this research was formally recognized by the Royal Swedish Academy of Sciences in 2002, when Vernon L. Smith was awarded the Nobel Memorial Prize in Economic Sciences “for having established laboratory experiments as a tool in empirical economic analysis, especially in the study of alternative market mechanisms.” The experimental asset market protocol pioneered by SSW has become the canonical workhorse model for hundreds of subsequent empirical investigations across experimental finance, neuroeconomics, and behavioral game theory, forever cementing its status as one of the most productive laboratory paradigms ever developed in the social sciences.

Beyond its methodological contributions, the SSW experiment catalyzed a profound epistemological shift within mainstream economic theory. It forced mathematical economists to confront the reality of bounded rationality, paving the way for the integration of non-equilibrium game theory, recursive cognitive hierarchies, and agent-based computational modeling into contemporary financial economics. The experiment demonstrated that human strategic friction is an intrinsic, irreducible feature of financial markets that must be formally integrated into theoretical models rather than dismissed as random empirical noise.

12.2 Algorithmic and High-Frequency Trading in Cognitive Hierarchies

As financial markets enter an era dominated by artificial intelligence, automated machine learning pipelines, and High-Frequency Trading (HFT) algorithms, the interaction between human cognitive hierarchies and automated execution code has emerged as an urgent research frontier. Modern institutional financial markets are no longer purely human double auctions; they are complex socio-technical hybrid ecosystems where human retail traders interact continuously with deterministic algorithmic architectures.

Within the analytical framework of Level-k game theory, algorithmic execution engines are explicitly engineered to detect and exploit the bounded cognitive depth of human participants. Quantitative market-making and statistical arbitrage algorithms analyze real-time order-book flow to identify the predictable momentum signatures generated by human $L_0$ trend-followers and $L_1$ speculative front-runners. By submitting, modifying, and canceling thousands of limit orders per second, high-frequency algorithms extract rents from the strategic lag of human cognitive processing, effectively occupying the role of an automated, hyper-precise $L_{k+1}$ strategic tier.

Experimental economists have actively responded to this structural transformation by constructing hybrid laboratory platforms where human subjects trade directly against automated algorithmic agents parameterized with varying objective functions. Early results from these hybrid experiments indicate that while passive, fundamental-anchoring algorithms can help dampen speculative bubbles, aggressive momentum-seeking algorithms can interact with human Level-k cognitive bounded rationality to produce hyper-accelerated bubble-crash cycles, including localized “flash crashes.” As artificial intelligence systems begin deploying deep reinforcement learning models capable of simulating arbitrary cognitive depths, understanding the game-theoretic interactions within hybrid human-algorithm order books represents one of the most critical challenges facing contemporary market design.

12.3 Unresolved Questions and Emerging Research Frontiers

Despite more than three decades of intensive empirical investigation since the seminal SSW paper, several fundamental theoretical questions remain unresolved, defining an exciting frontier for the next generation of behavioral economists and market architects. Primary among these is the challenge of developing a unified, micro-founded macro-financial model that endogenously derives aggregate asset market dynamics from empirically calibrated distributions of Level-k cognitive parameters.

Current empirical models typically treat the cognitive parameter distribution (e.g., the Poisson mean $tau$ in Cognitive Hierarchy theory) as an exogenous, static variable fitted retrospectively to laboratory data. A profound open question centers on how cognitive depth evolves endogenously as a function of changing market regimes. How do macroeconomic volatility shocks, interest rate transitions, or social media communication cascades alter the active distribution of cognitive types in real time? Developing dynamic structural models where an individual’s effective cognitive depth expands or contracts in response to perceived information complexity and financial stakes remains an essential theoretical objective.

Furthermore, experimental researchers are actively designing new institutional architectures capable of decoupling strategic uncertainty from fundamental asset valuation without relying on the costly, destructive mechanism of repeated market crashes. Emerging research protocols are exploring innovative continuous double auction mechanisms that incorporate algorithmic liquidity buffers, decentralized prediction markets for real-time fundamental anchoring, and novel cryptographic settlement protocols that structurally limit speculative momentum. Ultimately, the synthesis of Vernon Smith, Gerry Suchanek, and Arlington Williams’ foundational experimental design with the analytical rigor of Level-k behavioral game theory will continue to provide the indispensable empirical and theoretical foundation for designing the resilient, efficient financial institutions of the twenty-first century.

Conclusion

The enduring brilliance of the experimental paradigm established by Vernon L. Smith, Gerry L. Suchanek, and Arlington W. Williams in 1988 lies in its profound simplicity and devastating empirical power. By stripping financial asset trading down to its absolute bare essentials—an isolated room, a transparent and deterministic declining dividend schedule, a finite time horizon, and a computerized double auction—SSW exposed a foundational truth of human economic interaction: common knowledge of fundamental value does not imply common knowledge of rationality. Even when every market participant possesses the mathematical competence to calculate intrinsic value, the presence of strategic uncertainty inevitably unleashes speculative forces.

The Level-k behavioral game-theoretic framework provides the vital microeconomic bridge required to reconcile this empirical reality with rigorous economic theory. By recognizing that human decision-makers operate across discrete, bounded tiers of recursive strategic thinking, economists can systematically decompose the lifecycle of asset bubbles into rational optimizations against perceived lower-order counterparty behaviors. Speculation is revealed not as an inexplicable collective madness, but as a structured, non-equilibrium coordination game where agents knowingly trade overvalued assets in an attempt to out-time their peers before the terminal boundary is reached.

As modern financial markets grow increasingly complex, volatile, and technologically interconnected, the lessons of the SSW laboratory asset market remain more relevant than ever. Whether analyzing high-frequency algorithmic front-running on electronic equity exchanges, systemic liquidity distortions fueled by central bank balance sheet expansions, or the wild, anchorless speculative frenzies of decentralized cryptocurrency networks, the fundamental dynamic remains unchanged. Efficient market design requires far more than transparent information disclosure; it demands institutional architectures that account directly for the biological and cognitive boundaries of human strategic reasoning. The intellectual legacy of Smith, Suchanek, and Williams will permanently endure as the foundation upon which this deeper, behaviorally realistic financial economics continues to be built.

References

• Aumann, R. J. (1976). Agreeing to disagree. The Annals of Statistics, 4(6), 1236–1239. https://www.jstor.org/stable/2958591
• Ball, S. B., & Holt, C. A. (1998). Classroom games: Speculation and bubbles in an asset market. Journal of Economic Perspectives, 12(1), 207–218. https://www.aeaweb.org/articles?id=10.1257/jep.12.1.207
• Caginalp, G., Porter, D., & Smith, V. L. (1998). Initial cash/asset ratio and asset prices: An experimental study. Proceedings of the National Academy of Sciences, 95(2), 756–761. https://www.pnas.org/doi/10.1073/pnas.95.2.756
• Caginalp, G., Porter, D., & Smith, V. L. (2001). Financial bubbles: Excess cash, momentum, and incomplete information. The Journal of Psychology and Financial Markets, 2(2), 80–99. https://doi.org/10.1207/S15327760JPFM0202_3
• Camerer, C. F., & Ho, T. H. (1999). Experience-weighted attraction learning in normal form games. Econometrica, 67(4), 827–874. https://www.jstor.org/stable/2998573
• Camerer, C. F., Ho, T. H., & Chong, J. K. (2004). A cognitive hierarchy model of games. The Quarterly Journal of Economics, 119(3), 861–898. https://academic.oup.com/qje/article/119/3/861/1938834
• Fama, E. F. (1970). Efficient capital markets: A review of theory and empirical work. The Journal of Finance, 25(2), 383–417. https://www.jstor.org/stable/2325486
• Haruvy, E., Lahav, Y., & Noussair, C. N. (2007). Traders’ expectations in asset markets: Experimental evidence. American Economic Review, 97(5), 1901–1920. https://www.aeaweb.org/articles?id=10.1257/aer.97.5.1901
• Haruvy, E., & Noussair, C. N. (2006). The effect of short selling on bubbles and crashes in experimental asset markets. The Journal of Finance, 61(3), 1119–1157. https://doi.org/10.1111/j.1540-6261.2006.00868.x
• Kirchler, M., Huber, J., & Stöckl, T. (2012). Market design, information, and bubbles: An experimental asset market. American Economic Review, 102(6), 2898–2916. https://www.aeaweb.org/articles?id=10.1257/aer.102.6.2898
• McKelvey, R. D., & Palfrey, T. R. (1995). Quantal response equilibria for normal form games. Games and Economic Behavior, 10(1), 6–38. https://doi.org/10.1006/game.1995.1023
• McKelvey, R. D., & Palfrey, T. R. (1998). Quantal response equilibria for extensive form games. Experimental Economics, 1(1), 9–41. https://doi.org/10.1023/A:1009905800005
• Miller, E. M. (1977). Risk, uncertainty, and divergence of opinion. The Journal of Finance, 32(4), 1151–1168. https://doi.org/10.1111/j.1540-6261.1977.tb03317.x
• Nagel, R. (1995). Unraveling in guessing games: An experimental study. American Economic Review, 85(5), 1313–1326. https://www.jstor.org/stable/2951770
• Porter, D. P., & Smith, V. L. (1995). Futures contracting and dividend uncertainty in experimental asset markets. The Journal of Business, 68(4), 509–541. https://www.jstor.org/stable/2353140
• Simon, H. A. (1955). A behavioral model of rational choice. The Quarterly Journal of Economics, 69(1), 99–118. https://doi.org/10.2307/1884852
• Smith, V. L. (1962). An experimental study of competitive market behavior. Journal of Political Economy, 70(2), 111–137. https://www.jstor.org/stable/1861810
• Smith, V. L., Suchanek, G. L., & Williams, A. W. (1988). Bubbles, crashes, and endogenous expectations in experimental spot asset markets. Econometrica, 56(5), 1119–1151. https://www.jstor.org/stable/1911361
• Stahl, D. O., & Wilson, P. W. (1994). Experimental evidence on players’ models of other players. Journal of Economic Behavior & Organization, 25(3), 309–327. https://doi.org/10.1016/0167-2681(94)90103-1
• Stahl, D. O., & Wilson, P. W. (1995). On players’ models of other players: Theory and experimental evidence. Games and Economic Behavior, 10(1), 218–254. https://doi.org/10.1006/game.1995.1031