TL;DR
This paper introduces a powerful framework for analyzing financial derivatives by framing problems as convex optimization over the set of risk-neutral probabilities. This approach allows for the efficient computation of bounds on various financial quantities like cumulative distribution, VaR, and CVaR, and has been demonstrated on real-world S&P 500 and Bitcoin data.
What Problem Does This Paper Address?
In financial markets, the absence of arbitrage (the possibility of risk-free profit) is a fundamental assumption. The "arbitrage theorem" states that either arbitrage exists, or there's a "risk-neutral probability distribution" over future asset prices. This risk-neutral distribution is crucial because it not only represents a possible future state but also serves as a certificate that no arbitrage opportunities exist. However, in many realistic scenarios, there isn't a single, unique risk-neutral probability distribution; instead, there's a whole set of them. This multiplicity makes it challenging to derive precise values for financial quantities. The paper addresses this by developing methods to work with this set of risk-neutral probabilities, enabling the computation of bounds rather than single point estimates.
Methods
The core of the paper's methodology lies in formulating financial problems as convex optimization problems over the set of risk-neutral probabilities. This set, denoted by $\Pi$, is a convex polyhedron. By discretizing the possible future prices of an underlying asset into $m$ outcomes, the problem becomes a finite-dimensional convex optimization problem. This is highly tractable, scaling linearly with $m$. The authors leverage modern convex optimization solvers and domain-specific languages (DSLs) like CVXPY to efficiently solve these problems. They demonstrate how to compute bounds on:
- Expected values: For functions of the underlying asset's price at expiration.
- Ratios of expected values: Including probabilities of events.
- Cumulative Distribution Function (CDF): And its related measures like Value-at-Risk (VaR) and Conditional Value-at-Risk (CVaR).
- Estimation: Finding the maximum entropy or minimum KL-divergence risk-neutral distribution.
- Bounds on costs: For new investments.
- Sensitivities: How changes in costs affect optimal values.
Datasets
The paper evaluates its approach using real-world financial data:
- Standard & Poor's 500 Index (SPX): End-of-day bid and ask prices for SPX options on June 3, 2019, along with the index price.
- Bitcoin: Prices for Bitcoin options and futures from the Deribit exchange on March 27, 2020.
Key Insights and Ideas
- The power of bounds: When a unique risk-neutral probability doesn't exist, bounds derived from the set of possibilities provide valuable, actionable information. This is more robust than relying on a single, potentially arbitrary, choice of distribution.
- Convex optimization as a unifying framework: The paper shows that a wide range of financial problems, from pricing derivatives to estimating distributions, can be elegantly cast as convex optimization problems over the risk-neutral probability set.
- Tractability of high-dimensional problems: By discretizing the underlying price, the authors can handle a large number of outcomes (tens of thousands), making the approach practical for real-world applications.
- Interpretability: The methods are not only computationally efficient but also provide interpretable results, such as tight bounds on option prices.
Key Results
The paper demonstrates the effectiveness of its approach through numerical examples:
- SPX Index:
- Computed tight bounds on the expected expiration price, CDF, VaR, and CCDF.
- Estimated risk-neutral distributions, finding that the SPX appears to have heavy tails to the left (more probable large price decreases).
- Calculated bounds on put and call option prices that tightly bound the observed bid and ask prices.
- Bitcoin:
- Similar to the SPX, computed bounds on expected price, CDF, CCDF, and VaR.
- Found that Bitcoin's risk-neutral distribution appears to be heavy-tailed to the right (more probable large price increases), contrasting with the SPX.
- Demonstrated tight bounds on Bitcoin option and futures prices.
- Sensitivities: For the Bitcoin example, identified specific investments (like shorting the underlying or trading certain calls) that have the most significant impact on the maximum entropy risk-neutral probability.
Limitations and Future Work
The authors acknowledge several limitations:
- Finite outcomes: The approach requires discretizing the underlying asset's price, which means it cannot directly handle continuous price movements or infinitely many outcomes. This limits its direct application to American-style options, which can be exercised at any time.
- Static investments: The current framework assumes static investments that are fixed until expiration, precluding multi-period investment models and dynamic hedging strategies.
Future work suggested includes:
- Extending the framework to handle continuous outcomes or more complex option types like American options.
- Investigating the use of risk-neutral probabilities for multiple expiration dates to connect price movements over time.
Summary updated