Loading field map
How should one value a contract whose payoff depends on the future price of a stock, a currency, or a commodity? For most of the twentieth century, the answer seemed to require knowing investors' attitudes toward risk. The central intellectual achievement of derivatives pricing was to show that, under certain conditions, the price of a derivative can be determined without any reference to risk preferences at all—by constructing a portfolio of other assets that exactly replicates the derivative's payoff. This insight, known as no-arbitrage pricing, transformed finance. Yet the conditions under which perfect replication is possible are restrictive, and much of the subsequent history of the subfield has been about extending the no-arbitrage logic to more realistic settings while grappling with the limits it imposes.
The earliest systematic attempt to price an option was Louis Bachelier's 1900 doctoral thesis, which treated stock prices as following a random walk and valued an option as the expected value of its payoff under the true probability distribution of the underlying asset. Bachelier's approach was actuarial: it discounted the expected payoff at the risk-free rate, implicitly assuming that investors do not require a premium for bearing risk. This assumption was not justified by any economic argument, and it produced prices that could violate the commonsense constraint that an option cannot be worth less than its immediate exercise value.
In the 1960s, economists such as Paul Samuelson and James Boness developed equilibrium models that grounded option prices in investors' utility functions. In these models, the expected payoff was discounted at a rate that reflected the asset's risk, which depended on the representative investor's risk aversion. The resulting prices were logically consistent, but they required knowing the utility function—something that could not be observed in market data. The practical problem was clear: option pricing needed a method that used only observable market prices, not unobservable preferences.
Fischer Black, Myron Scholes, and Robert Merton broke through this impasse in 1973 by showing that an option's payoff could be replicated by a continuously rebalanced portfolio of the underlying stock and a risk-free bond. If the replicating portfolio costs the same as the option at every instant, then any discrepancy between the option price and the cost of the replicating portfolio would create an arbitrage opportunity—a riskless profit. To rule out arbitrage, the option price must equal the cost of the replicating portfolio. This argument eliminated any dependence on risk preferences.
The Black-Scholes-Merton framework assumed that the underlying asset follows a geometric Brownian motion with constant volatility, that trading is continuous, and that there are no transaction costs or dividends. Under these assumptions, the model produced a closed-form formula for European call and put options. The formula depended only on the current stock price, the strike price, the time to expiration, the risk-free interest rate, and the volatility of the stock. The first four inputs were observable; volatility had to be estimated, but it was a property of the asset, not of investors' preferences.
The timing was fortuitous: the Chicago Board Options Exchange opened in April 1973, creating a centralized market for listed options. The Black-Scholes-Merton model quickly became the industry benchmark, and it remains the starting point for most practical option pricing today.
Almost immediately, researchers identified two important limitations of the Black-Scholes-Merton model. First, asset prices sometimes move discontinuously—they jump—in response to unexpected news. Robert Merton's 1976 jump-diffusion model added a Poisson jump component to the continuous diffusion process. The jump risk could not be hedged by trading the underlying stock alone, because the jump size was random and the stock position could not be adjusted during the jump. Merton showed that if the jump risk was diversifiable (uncorrelated with the market), the option price could still be expressed as a weighted average of Black-Scholes prices conditional on the number of jumps. The jump-diffusion model preserved the no-arbitrage logic while acknowledging that not all risks can be perfectly hedged.
Second, the continuous-time mathematics of Black-Scholes-Merton was inaccessible to many practitioners and did not easily handle American options, which can be exercised before expiration. In 1979, John Cox, Stephen Ross, and Mark Rubinstein introduced the binomial lattice model, which replaced continuous trading with a discrete-time tree of possible asset prices. At each node, the option value was computed by taking the risk-neutral expected value of the option at the next nodes and discounting at the risk-free rate. The binomial model's risk-neutral probabilities were a direct discrete-time instantiation of the martingale measure idea that would soon be formalized. The model was computationally flexible: by adding more steps, it could approximate the Black-Scholes-Merton result, and it could easily price American options by checking at each node whether early exercise was optimal.
Both the jump-diffusion and binomial models extended the no-arbitrage approach rather than replacing it. They showed that the replicating-portfolio logic could be adapted to more complex settings, but they also hinted at a deeper structure underlying all no-arbitrage pricing.
In 1979 and 1981, J. Michael Harrison and David Kreps (and later Harrison and Stanley Pliska) provided the mathematical foundation that unified the earlier models. They showed that the absence of arbitrage is equivalent to the existence of a probability measure—called a risk-neutral or equivalent martingale measure—under which discounted asset prices are martingales. Under this measure, the expected future payoff of any derivative, discounted at the risk-free rate, gives its current price. The actual probabilities of future price movements are irrelevant; only the risk-neutral probabilities matter.
This framework absorbed the Black-Scholes-Merton model, the jump-diffusion model, and the binomial model as special cases. In the Black-Scholes-Merton setting, the risk-neutral measure is unique because the market is complete: every derivative can be perfectly replicated. In the jump-diffusion model, the risk-neutral measure is also unique if the jump risk is diversifiable; otherwise, there are multiple risk-neutral measures, and the price is not uniquely determined by no-arbitrage alone. The binomial model's risk-neutral probabilities are simply the discrete-time version of the same idea.
The risk-neutral and martingale framework introduced the distinction between complete and incomplete markets as a central organizing concept. In a complete market, every contingent claim can be replicated, and the no-arbitrage price is unique. In an incomplete market, replication is imperfect, and no-arbitrage only restricts prices to an interval. This distinction became the lens through which all subsequent pricing models were understood.
By the mid-1980s, empirical evidence had accumulated that the Black-Scholes-Merton model's assumption of constant volatility was false. When traders used the model to compute implied volatilities from market prices, they found that implied volatility varied with strike price and time to expiration—a pattern known as the volatility smile or skew. This pattern indicated that the market was pricing options as if volatility itself was random.
John Hull and Alan White's 1987 model treated volatility as a separate stochastic process, typically mean-reverting. Because volatility is not a traded asset, the volatility risk cannot be hedged by trading the underlying stock alone. The market is therefore incomplete, and the model requires specifying a market price of volatility risk—the extra return investors demand for bearing volatility risk. This market price of risk is not determined by no-arbitrage and must be estimated from data or calibrated to option prices.
Steven Heston's 1993 model provided a closed-form solution for European options under stochastic volatility, assuming that the variance process follows a mean-reverting square-root process (the CIR process) and that the asset price and variance are correlated. The Heston model became the most widely used stochastic volatility model in practice because it captured the volatility smile while remaining computationally tractable. Today, stochastic volatility models are the industry standard for equity and currency options, and they are often combined with jumps to capture both the volatility smile and the short-term skew.
Once the market is incomplete—whether because of stochastic volatility, jumps, transaction costs, or illiquidity—no-arbitrage alone does not determine a unique price. The pricing problem then splits into two broad approaches.
The first approach, used within stochastic volatility and jump-diffusion models, is to specify a market price of risk for each non-traded risk factor. This market price of risk is often chosen by calibrating the model to a set of liquid option prices, effectively letting the market tell us how the risk is priced. This approach is pragmatic and widely used in trading desks, but it is theoretically incomplete: the market price of risk is a free parameter that must be supplied from outside the model.
The second approach, utility-based pricing, derives the derivative price from an investor's optimal portfolio choice. The investor maximizes expected utility of terminal wealth, and the derivative's reservation price is the price at which the investor is indifferent between adding the derivative to the portfolio and not adding it. This approach makes the dependence on preferences explicit: different investors with different risk aversions will assign different prices to the same derivative. Utility-based pricing provides a theoretical foundation for understanding the limits of arbitrage—why prices can deviate from the no-arbitrage bounds when markets are incomplete—but it is rarely used in practice because it requires specifying the utility function and the investor's existing portfolio.
Today, the risk-neutral and martingale framework provides the common language for all derivatives pricing. Practitioners and academics alike think in terms of risk-neutral expectations, even when the market is incomplete. The Black-Scholes-Merton model remains the baseline, but it is almost always adjusted: stochastic volatility models handle the volatility smile, jump-diffusion models handle short-term skew, and binomial or finite-difference methods handle American features and path-dependence.
The leading frameworks agree on the central role of no-arbitrage and risk-neutral pricing. They disagree on how to handle market incompleteness. The stochastic volatility camp treats the market price of volatility risk as a calibration parameter, while the utility-based pricing camp insists that preferences must be modeled explicitly. A growing body of research tries to bridge the gap by deriving the market price of risk from equilibrium conditions or from the prices of other traded assets.
A second area of disagreement concerns the role of jumps. Some researchers argue that stochastic volatility alone can explain the volatility smile, while others maintain that jumps are necessary to capture the behavior of short-dated options. The debate is partly empirical and partly methodological: adding jumps increases model complexity and makes the market even more incomplete.
A third area of active research is the pricing of derivatives in markets with transaction costs, funding constraints, and other frictions. These frictions make perfect replication impossible even in principle, and they push pricing toward the utility-based framework. The challenge is to develop models that are both theoretically sound and computationally practical.
Derivatives pricing has come a long way from Bachelier's actuarial expected value. The no-arbitrage revolution showed that preference-free pricing is possible under ideal conditions. The subsequent development of stochastic volatility and incomplete-markets models has shown how to relax those conditions while preserving the core insight that arbitrage is the ultimate constraint on prices.