Notes on Model Assumptions, No-Arbitrage, and Market Efficiency (Lecture Outline)
Model and Assumptions
The lecturer begins by reinforcing a central theme from the prior class: a model is equivalent to its assumptions. To understand a model, you must understand its assumptions. Testing a model requires testing its assumptions; rejecting the model by rejecting some of its implications is possible, but it doesn’t prove the model correct unless every implication is tested and shown to be consistent with the assumptions. In practice, all implications may not be testable, which is why we test the assumptions themselves. Models are mathematical abstractions of reality; the assumptions are simplifications or approximations, never absolutely correct. Therefore, testing a model is really about testing whether the approximations are good enough for the intended purpose. Goodness of fit is judged by whether errors—the differences between observed market prices and model prices—behave in a reasonable way. Specifically, for a good approximation one prefers errors that are unbiased (on average correct), not autocorrelated over time, and not correlated with important variables included in the model. If the errors are correlated with the level of the price being modeled, it signals that something is missing from the model; with the correct model, the correlation should be zero. If errors are autocorrelated across time, it suggests that some persistence in the price process (the passage of time) isn’t captured. If errors are too large relative to the price scale (e.g., an error of $50 on a price typically around $100), the model may be technically unbiased but practically useless. Consequently, testing a model means testing its assumptions where possible, and ensuring the errors behave well. In practice, not all assumptions will have testable data, so one often relies on experience and common sense for some assumptions. The instructor then introduces a two-type decomposition of assumptions to guide testing: robust vs. critical.
A robust assumption is one where small changes to the assumption do not materially alter the implications. The example given is assuming no transaction costs: if buying something worth $100 incurs a cost of only a penny, and the model’s implications do not change meaningfully when that penny cost is added, the assumption is robust. In continuous-function terms, small changes in the input x lead to small changes in the output along the curve; if you move a little along the curve, the implications stay essentially the same. If the assumption is robust, you feel comfortable using the model with that approximation, knowing that minor frictions won’t derail it.
A critical assumption, by contrast, can cause a discontinuous switch in implications if altered even slightly. Short sale constraints are given as an explicit example of a critical assumption: adding a short-sale constraint can change the implications drastically. If short selling is allowed only to a small extent, the model’s implications can change in a non-smooth way, which makes such assumptions deserving of careful testing when possible. Short-sale constraints are easy to test where markets provide clear regulatory or empirical evidence: if the market bans shorting, a model assuming no short sales is invalid in that context. For critical assumptions, it may be necessary to adjust the model or its constraints to reflect reality, or rely on implication testing rather than full data-driven validation. The instructor emphasizes that this is the art of applying models: rules of thumb and judgment accompany the rigorous framework.
In practice, market frictions are often ignored for two reasons. First, for idealized reasoning and tractable theory, and second, because for many large financial institutions, a frictionless model is a reasonable approximation that still yields useful business decisions. Anecdotal evidence is cited: many industry practitioners use frictionless models to guide decisions successfully, even though frictions exist. This sets the stage for a closer look at specific assumptions, which we address next.
No Credit Risk and Counterparty Risk
The lecture then moves to the assumption of no credit risk (also called no default risk or no counterparty risk, depending on the context). In a default-free world, zero-coupon bonds pay their promise of 100 with probability one. This assumption is described as a critical one: the value of debt will differ markedly if there is a genuine chance of default. The text notes that one cannot apply the no-credit-risk assumption to all instruments; it is appropriate in certain contexts (e.g., risk-free government securities) but not for others (e.g., corporate or mortgage-backed securities).
A subtle distinction is drawn between counterparty risk and default risk. For interest rate derivatives, there are zero-supply securities (a buyer and seller must exist for a derivative to exist), and the standard modeling approach often assumes that both counterparties will complete their obligations. This is another form of no counterparty risk for modeling purposes. The narrative then illustrates what happens when this assumption breaks down with a real-world anecdote: a large Canadian bank (CIBC) engaged in selling credit default swaps (CDS) under the belief that counterparties were risk-free (rated AAA). They hedged by buying the same CDS from another institution, assuming no credit risk on the counterparty. When the credit crisis hit, this assumption proved catastrophically false, resulting in enormous losses. The speaker recounts being an expert witness and notes that the defense from top management claiming “we were hedged” was untenable in hindsight. This underscores the importance of testing and monitoring counterparty risk, particularly in derivatives trading.
The discussion then connects to the role of exchanges and margining: derivatives traded on exchanges provide a guarantor function through margin requirements, reducing counterparty risk relative to over-the-counter trading. Hence, for exchange-traded derivatives, assuming no counterparty risk is often a reasonable first approximation. The key takeaway is that credit risk is a critical assumption for many derivatives but can be mitigated or managed in certain markets and through central clearing.
Competitive Markets, Liquidity, and Price Impact
The lecturer revisits the notion of competitive markets as markets in which traders are price takers: buying or selling a large quantity does not move the price against you (i.e., no price impact from your own order). In reality, most markets exhibit some price impact, which is tied to liquidity and information asymmetries. Liquidity risk arises from how your trades affect prices in the short run (temporary impact) and possibly in the longer run (permanent impact). Two types of price impact are distinguished:
Temporary (endogenous transaction costs): your trade moves the price today, but the effect may dissipate over time. If these impacts are small, the assumption of a frictionless market is robust and reasonable.
Permanent (affects future price evolution): if your trade permanently shifts the price path, this is a critical assumption because it changes the dynamics of hedging and valuation over time. If your own trading alters future prices in a meaningful way, the model’s implications may be invalid.
The discussion includes a vivid example of a market manipulation strategy called a pump-and-dump. An operator begins by buying a security, pushing the price up, drawing in other investors, and then selling into the rising market. Once arbitrageurs have piled in, the manipulator dumps shares at a higher price, realizing profits while others incur losses. Although illegal in many jurisdictions, such manipulation can invalidate a model’s pricing and risk management if the price process is capturing manipulation as if it were normal trading. The point is that if markets can be manipulated, a model built on a frictionless and competitive assumption may produce misleading prices and hedge ratios. The instructor emphasizes vigilance: trading should not be treated as a black box, and model use should be coupled with ongoing market observation for potential violations of the core assumptions.
An illustrative financial example is provided using European call options. Consider a stock price process that follows a lognormal distribution with constant volatility as maturity approaches. For a call option with strike $K$ and maturity $T$, the payoff is If the stock is currently near but below the strike (out-of-the-money) and there is little time to expiry, the Black-Scholes-type model would assign a near-zero value to the call, implying almost no need to hedge (the delta is near zero). However, if a manipulator buys a large quantity of the stock and drives the price into the money, the true value of the option can jump dramatically. The necessary hedge, to remove risk, would require shorting a large amount of stock (potentially the entire position), which reveals how a model that neglects manipulation can dramatically misprice and mismanage risk. This example underscores the fact that if markets are not truly competitive and manipulation is possible, the model’s assumptions fail in important ways.
The takeaway is that model validity hinges on the assumption of competitive, liquid markets. If this assumption is violated, one must adjust the model, monitor it for signs of manipulation, and be prepared to back off or recalibrate when market conditions change. The lecturer also discusses how practitioners might probabilistically assess the likelihood of such violations, use signals or machine-learning indicators, and calibrate risk management accordingly. Yet, in practice, the use of a model remains contingent on continuous market observation and judgment rather than blind reliance on a black-box approach.
No-Arbitrage and Arbitrage Opportunities
A core assumption introduced is no-arbitrage: in an arbitrage-free market, there are no opportunities to earn a riskless profit with zero net investment. The instructor delineates two concrete types that underpin no-arbitrage reasoning:
Type 1 arbitrage: at a single point in time, across markets, the same asset trades at different prices. In frictionless, competitive markets, this creates an immediate riskless profit by buying cheap and selling dear. This is the classic instantaneous price-difference arbitrage.
Type 2 arbitrage (zero-net investment): you form a portfolio that requires no initial net investment (or a nonpositive one) by combining long and short positions and ensuring the future value is nonnegative with positive probability. The portfolio’s value at maturity is never negative, and there exists a state in which it is strictly positive, guaranteeing a riskless profit if left unexploited. The speaker uses intuitive desert analogies (finding a lottery ticket in the desert; finding a dollar bill) to illustrate the idea of riskless profit with zero net investment.
The no-arbitrage assumption is framed as highly reasonable because markets should quickly eliminate free lottery tickets or obvious mispricings. However, the assumption is also described as a critical one: if arbitrage opportunities exist, prices will adjust until they disappear; if they persist, one would exploit them and earn profits. The lecturer acknowledges that arbitrage opportunities do exist in real markets, citing historical anecdotes (e.g., Thorp’s early arbitrage profits in options markets; high-frequency traders finding rapid, small-arbitrage opportunities that vanish as fast as they appear; sales of mispriced CDOs before the crisis). The point is not that arbitrage never exists, but that it tends to be transient and requires speed and information.
A deeper empirical point is made through the CDO experience: misratings by rating agencies created structural inefficiencies that produced arbitrage profits for those who understood the mispricing. When the market corrected, those opportunities evaporated or led to large losses for misinformed counterparties, highlighting the practical importance of model assumptions and the risk of relying on mispriced inputs.
No-arbitrage remains the foundational pricing principle in the absence of market frictions, but there is a distinction to be drawn between no-arbitrage as a condition and market efficiency as a stronger, information-based property. The discussion emphasizes that arbitrage opportunities, while possible, should not be expected to persist; if they do, the market is not in equilibrium, and riskless profits are available to those who can exploit the discrepancy.
Efficient Markets and Information
The lecturer then connects the no-arbitrage framework to the broader notion of market efficiency, a topic often debated in finance. Efficient markets are those in which prices fully and accurately reflect all relevant information. There are longstanding debates about whether markets are efficient, and if so, to what degree. The classical debate has two dominant strands: some argue that markets are efficient (no arbitrage suffices to price assets correctly), while others point to abundant evidence of inefficiencies, especially in certain periods or markets (e.g., after the advent of high-frequency trading, during new-market openings, or during crises).
The professor emphasizes that efficiency is not an automatic consequence of no-arbitrage; efficiency is a stronger condition. Efficiency is defined (in the course) via Eugene Fama’s perspective: a market is efficient with respect to a given information set if there exists an equilibrium model consistent with the observed prices given that information set. This definition relies on two key ideas: existence vs. identification. Existence means there is some equilibrium model that could generate the observed prices; identification would require characterizing that model explicitly, which is often not possible. The upshot is that one can argue for efficiency by showing that no arbitrage opportunities exist and that the market’s pricing is dominated by well-behaved price dynamics (no dominance). The talk then clarifies three formalizations of efficiency based on information sets:
Weak form efficiency: prices already reflect all information contained in past prices and past volumes and dividends. If a market is weak-form efficient, no arbitrage opportunities can be found by analyzing past price history.
Semi-strong form efficiency: prices reflect all publicly available information; even novel public information cannot be used to earn abnormal profits in a consistent way.
Strong form efficiency: prices reflect all information, including insider information; even those with private information cannot consistently beat the market.
The speaker notes that there is no single, universally settled empirical verdict on efficiency. In practice, weak-form efficiency is often observed as a reasonable baseline, but counterexamples exist (e.g., certain high-frequency or newly opened markets) that challenge it. For semi-strong efficiency, the evidence is mixed, and the literature is replete with contradictory findings, in part because empirical tests depend on instrumentation, time period, and market context. Strong-form efficiency is widely regarded as false in most real-world settings because insiders can and do earn abnormal profits in some cases, despite laws against such trading.
A crucial point is that efficiency is conditioned on the information set being considered. When evaluating efficiency, one must specify which information is relevant. The literature classifies efficiency according to three information sets: weak (past prices, volumes, dividends), semi-strong (public information), and strong (all information). The discussion stresses that market efficiency is a stronger condition than no-arbitrage, and if one truly believes markets are efficient, then the modeling exercise of constructing a pricing model based on no-arbitrage is less compelling. Conversely, if markets are not efficient, arbitrage opportunities exist and can be exploited, which is precisely what practitioners seek to do.
In 2012, the instructor formalized a rigorous definition of market efficiency under frictionless and competitive markets: a market is efficient with respect to an information set if there exists an equilibrium model consistent with the observed prices given that information set. The distinction between existence and identification is highlighted: existence means such a model could exist, but identifying or characterizing it may be impossible in practice. The operational takeaway is that efficiency can be argued via no-arbitrage plus a no-dominance condition; if no arbitrage opportunities exist and no dominance is observed, the market can be deemed efficient with respect to the considered information set. This framework clarifies why some researchers claim markets are efficient and others argue that markets are not—both positions can be partly true depending on the information set and time period under study.
The lecturer then connects efficiency theory to practical pricing and model testing. If efficiency holds, there should be no arbitrage; if inefficiencies exist, they can be exploited for profit. However, even if markets are efficient in a formal sense, pricing models may still be useful for risk management and for understanding the distribution of potential hedging outcomes—so long as one remains mindful of the assumptions about information flow and liquidity. The discussion also distinguishes three forms of efficiency with respect to the information that matters for pricing and arbitrage: weak, semi-strong, and strong, as described above.
Price-Evolution, Completeness, and the HJM Agenda
The final major topic before the upcoming session is the evolution of price processes and the notion of completeness. The instructor flags one more strong assumption: the evolution of the price process (for example, interest rates and zero-coupon bonds). This evolution must be consistent with no-arbitrage; otherwise, the chosen dynamics could permit arbitrage opportunities. In practical terms, the evolution of prices is often modeled via a stochastic process (for zero-coupon bonds, interest rate models), and a core part of model building is ensuring that the chosen dynamics are arbitrage-free and, in many contexts, complete. Completeness means that every contingent claim can be replicated by trading in a self-financing strategy using traded securities. If the market is incomplete, some payoffs cannot be replicated exactly, which complicates pricing and hedging.
The concept of synthetic construction is introduced: a derivative can be priced by creating a portfolio of traded securities that replicates its payoff. The feasibility of this replication depends on the chosen price evolution and on market completeness. The Wardrobe of this course’s methodology centers on ensuring the evolution is arbitrage-free and can support a complete market, enabling derivative pricing via replication. The Hedging/Forward-Measure framework (the HJM model) is presented as a key contribution to the literature because it provides a systematic approach to specifying the drift and diffusion terms of the forward rate curve in a way that guarantees no-arbitrage for an arbitrary evolution of the term structure. The instructor previews that the next session (Monday) will begin constructing the HJM model and its pricing machinery, starting from the evolution of interest rates and the term structure.
Practical Takeaways and Connections
Modeling philosophy: always tie the model to its assumptions; use the assumptions as the primary test targets; be mindful of when some assumptions cannot be tested with data and rely on experience.
Robust vs. critical assumptions: robust assumptions are safer to ignore or approximate; critical assumptions require careful testing or model modification when violated (e.g., no short selling, no counterparty risk in certain contexts).
Market frictions matter: even small frictions can change pricing and hedging; temporary vs. permanent price impact distinguishes robust from critical assumptions.
No-arbitrage is fundamental but weaker than market efficiency; arbitrage opportunities can exist but are typically fleeting and require fast action; inefficiencies imply potential profits and drive market dynamics.
Efficiency concepts (weak, semi-strong, strong) formalize how information is reflected in prices; efficiency is a stronger condition than no-arbitrage and depends on the information set considered.
Price evolution and market completeness are central to asset pricing; ensuring no-arbitrage and completeness is the backbone of replication-based pricing; advanced models like the HJM framework address broad classes of evolutions while maintaining arbitrage-free and complete-market properties.
Key Definitions, Concepts, and Formulas (glossary)
Model: a mathematical representation of a real phenomenon built from assumptions; a model is identified with its assumptions.
Assumptions: simplifications used to make modeling tractable; can be robust or critical.
Robust assumption: small changes to the assumption leave the implications largely unchanged (e.g., negligible transaction costs).
Critical assumption: small changes can cause large, discontinuous changes in implications (e.g., short-sale constraints).
No-arbitrage: there are no portfolios with nonnegative future value that have strictly positive probability of positive payoff and require zero net investment; arbitrage opportunities exist only if there is a strategy with zero net investment and nonnegative payoff that is strictly positive with positive probability.
Arbitrage types:
Type 1: price discrepancies across markets for the same asset at a single point in time.
Type 2: zero-net investment strategies (long/short combos) with nonnegative future payoff; sometimes there exists a state with strictly positive payoff.
Efficient market (Fama, 1970s): a market is efficient with respect to information set I if there exists an equilibrium model consistent with observed prices given I. Existence vs identification: existence means such an model could exist; identification requires characterizing it.
Information forms of efficiency:
Weak form: prices reflect all past price information and volume/dividends.
Semi-strong form: prices reflect all publicly available information.
Strong form: prices reflect all information, including insider information.
Price evolution and complete markets: pricing requires a consistent evolution of prices (e.g., interest rates for zero-coupon bonds) that does not admit arbitrage and allows replication of contingent claims (completeness).
Synthetic construction: pricing a derivative by replicating its payoff with a portfolio of traded securities.
HJM framework: an approach to specify the evolution of the forward rate curve in arbitrage-free ways for a broad class of term structures; its contribution is ensuring the chosen evolution is arbitrage-free and compatible with complete markets.
Next steps
On Monday, the course will begin constructing the HJM model and detailing the evolution of interest rates and the term structure, while continually checking for consistency with no-arbitrage and market completeness. Have a nice weekend, and prepare questions if you have any.