Information-Theoretic Model for the Growth Optimal Portfolio of Stocks
summary
The gist
The paper constructs and models a proxy for the growth-optimal portfolio of stocks using information-theoretic principles to enable benchmark-neutral pricing and hedging.
In short
The model creates a proxy for stocks that grow optimally using information theory to allow for benchmark-neutral pricing and hedging. This method offers an alternative to standard risk-neutral pricing, which is often too high for long-term derivatives. The approach uses principles like surprisal maximization to derive market dynamics.
Key concepts
- Information Theoretic Foundation (A1-A4)
- These are four core assumptions that form the basis of the model. They interpret the stock market as a communication system where distinct trading behaviors can be derived by maximizing 'surprisals.' This framework guides how the market dynamics and portfolio behavior are mathematically structured.
- Minimal Market Model (MMM) Parametrization
- This parametrization models how activity time evolves over time using a specific differential equation. It is inspired by surprisal maximization and assumes the normalized activity follows an ergodic scalar diffusion process, which helps capture trading overreactions in the market.
- Benchmark-Neutral Pricing Measure (QS*)
- This is a new pricing measure derived when using the growth-optimal portfolio as a reference point (numeraire). Using this measure allows for pricing and hedging that are significantly lower than those suggested by traditional risk-neutral pricing, especially for long-term bonds.
- Proxy Construction (EWI163)
- A proxy for the benchmark is built using equal-weighted indices, such as EWI163. This index is theoretically expected to be the best representation of the growth-optimal portfolio. Empirical tests show that stock return densities follow a Student t distribution with slightly less than four degrees of freedom.
Terminology used across episodes
This episode discusses
- Information-Theoretic Model for the Growth Optimal Portfolio of Stocks · Paper Radio
- Free Lunches with Vanishing Risks Most Likely Exist
- Information-Theoretic Approach to Financial Market Modelling
- Benchmark-Neutral Risk-Minimization for insurance products and nonreplicable claims
The paper
Information-Theoretic Model for the Growth Optimal Portfolio of Stocks · Read on arXiv
Eckhard Platen, Kevin Fergusson
University of Technology Sydney · Australian National University · Bond University
Drawing on information-theoretic principles and empirical evidence, a proxy for the growth-optimal portfolio of stocks (the benchmark) is constructed and modeled. The derivative of its activity time captures overreactions in trading behavior and follows a fast-moving, quasi-stationary scalar diffusion. The benchmark's log-returns are predicted to follow a Student t distribution with degrees of freedom slightly less than four, and this prediction is tested empirically. With the benchmark as numeraire, the derived model enables benchmark-neutral pricing and hedging.
Transcript
Introduction to the show: ident: Quantitative Finance. The day in quantitative finance: models of markets, risk and prices.
Ruth: I'm Ruth.
Warren: And I'm Warren, here with Iris.
Ruth: We're looking at the paper: "Information-Theoretic Model for the Growth Optimal Portfolio of Stocks".
Warren: Here's the gist of the paper. The paper constructs and models a proxy for the growth-optimal portfolio of stocks using information-theoretic principles to enable benchmark-neutral pricing and hedging.
Ruth: This approach is important because it provides an alternative to risk-neutral pricing, which has been shown to be significantly overpriced for long-term derivative prices Information Theoretic Foundation The model is built upon four information-theoretic principles that underpin the benchmark approach These principles are A1 through A4, which formulate the core assumptions of the proposed model for the benchmark.
Warren: Specifically, Assumption A3 posits that "the surprisals of the quasi-stationary probability densities of the normalized factors and the normalized activity in their activity times are maximized" The model interprets the market as a communication system where distinct market dynamics are derived in a "surprisal-maximized market".
Ruth: Benchmark Dynamics and Activity Model The benchmark dynamics are modeled in activity time using the Minimal Market Model (MMM) parametrization, which is inspired by surprisal maximization The MMM parametrization uniquely specifies the portfolio’s activity time via the differential equation dτt dt = σ2t e−τt dt.
Title and authors: Ruth: Moving on to how they suggest improving this work, the authors point out several ways the framework can be strengthened. They focus on making the proxy construction more robust and refining how they model those return densities.
Warren: I’m interested in what they propose for constructing that proxy itself, because if the benchmark proxy isn't accurate, the whole pricing method falls apart regardless of how good the underlying theory is.
Iris: They suggest using equal-weighted indices, or EWIs, like EWI163 as a starting point because it’s theoretically expected to be the best proxy for this benchmark.
Ruth: And they go further by testing ten stylized empirical facts against their model, and they found that the log-return density prediction with respect to activity time is a Student t distribution with slightly less than four degrees of freedom.
Warren: That sounds like a strong empirical anchor, but how do we check if that Student t fit is really the best one? They need a rigorous way to compare different density models.
Iris: They suggest implementing a likelihood ratio test to statistically compare different log-return density models, for instance comparing the Student t density against simpler nested models like the Student t itself.
Ruth: So, they aren't just settling on one distribution; they are building a statistical process to determine which one best fits the observed data from stock portfolios in terms of their returns over time.
Warren: That makes sense because if the density is wrong, our estimate of risk is wrong too, and that impacts how much we need to hedge.
The paper's summary: Ruth: To summarize what they suggest improving about the model structure itself, it’s really about making the underlying assumptions more flexible and ensuring the empirical fit is rigorously tested.
Warren: I mean, if this information-theoretic framework can be adapted to handle different market regimes, that would be a big win for practical application.
Iris: They also suggest modeling the normalized activity of the benchmark using an ergodic scalar diffusion process where the stationary density is a gamma density with four degrees of freedom. This model is designed to capture those overreactions in trading behavior we talked about earlier.
Ruth: That gamma distribution for activity time seems like a specific way to handle how quickly things move, and it links back into that assumption A3 about maximizing surprisals.
Warren: So, they’re trying to get a more detailed picture of the benchmark's movement by using this diffusion process instead of just the simpler models they started with.
Iris: And when you consider what this means for identification, it helps confirm that the variation in trading behavior is indeed what drives the observed dynamics in a way that aligns with their theoretical framework.
The paper's improvements: Ruth: So, wrapping up on this Information-Theoretic Model for the Growth Optimal Portfolio of Stocks, we’ve seen how they construct a proxy using EWIs and how they derive a benchmark-neutral pricing measure.
Warren: And the key result is that this method allows for prices significantly lower than risk-neutral pricing suggests for long maturities, which is what we were hoping to see when we looked at the overpricing issue.
Iris: It’s important to remember their limitation here: they say their model relies on these four assumptions or principles, and the divergence between the benchmark measure and the real-world probability measure is minimized under those specific conditions.
Ruth: And that’s where it stops working; if you change the market dynamics in a way that violates those core information-theoretic principles, this benchmark approach might not give you as good an answer.
Warren: So, for someone just listening to the show, what does this mean? It means we have a new theoretical tool to get better prices for long-term financial instruments by using a proxy derived from market activity rather than relying on the standard risk-neutral assumptions that might be too optimistic.
Iris: Essentially, it offers an alternative way to value things when the standard methods are theoretically unsatisfactory for extreme maturities.
Conclusion: Ruth: So to wrap up, this paper builds a model for finding the growth optimal portfolio of stocks using information theory to price things better than standard risk-neutral pricing would allow, especially for long maturities.
Warren: Yeah, it’s about using a proxy for that optimal portfolio—a benchmark—and then doing benchmark-neutral pricing and hedging. So we get prices significantly lower than what the usual methods suggest for those long zero-coupon bonds.
Iris: From an econometric perspective, they build this whole thing on four core information-theoretic principles, starting with this idea that in a surprisal-maximized market, you’re looking at how the probability densities of normalized factors are behaving.
Ruth: That sounds like a big assumption about how the market communicates; they interpret the market as some kind of communication system where distinct dynamics emerge from maximizing surprise.
Warren: And they model this benchmark in activity time using something called the Minimal Market Model, which specifies a differential equation that dictates how much activity happens over time.
Iris: They use that differential equation to define the normalized activity as an ergodic scalar diffusion, and they claim the stationary density for this activity is a gamma distribution with four degrees of freedom.
Ruth: So that gamma distribution for the benchmark’s activity is supposed to capture those overreactions in trading behavior, which makes sense if we think about how quickly things move.
Warren: And they build a proxy for this benchmark using equal-weighted indices, like EWI163, because they expect that index to be the best stand-in for the actual growth optimal portfolio.
Iris: They tested ten different empirical facts against their model and found that the log-return density with respect to activity time is predicted to be a Student t distribution with slightly less than four degrees of freedom.
Ruth: That Student t fit, with that specific number of degrees of freedom, seems to be the main empirical anchor for their findings on how stock returns behave over time.
Warren: I mean, what does this actually change for someone looking at hedging? It suggests you can use this benchmark-neutral measure to get prices that are way more conservative than the standard risk-neutral price when you’re dealing with long-term instruments.
Ruth: Exactly. It gives a way to price these things that seems much more realistic for long time horizons, moving away from those risk-neutral assumptions we often use.
Iris: The limitation they mention is that their model relies on those four specific information-theoretic principles being true, and if the market dynamics shift in a way that breaks those rules, the whole benchmark approach loses its strength.
Warren: Right. So it’s a powerful tool under these specific assumptions, but you gotta be careful about where it breaks down in real markets.
Ruth: It’s definitely a different lens for looking at asset pricing, but we still need to see how robust this framework is when people actually start changing their minds about those underlying market assumptions.
Warren: Speaking of changing minds, next up, we’ve got some papers on how firms are responding to opportunities in language model space and strategic crowding.
More episodes
- The day in conversation for 2026-11-19
- Daily Summary for 2026-11-19
- The day in conversation for 2026-11-18
- Daily Summary for 2026-11-18
- 2601.05274-On the use of case estimate and transactional payment data in neural networks for individual loss reserving
- 2009.00868-Loss-Given-Default Modeling by Post-Last Passage Time Process
- 2606.30070-Financial Resilience Evaluation: From Conditional Expectations to Dynamic Convex Risk Measures
- 2212.12687-Measuring price impact and information content of trades in a time-varying setting
- Daily Summary for 2026-11-17
- 2401.08323-Dynamic portfolio selection under generalized disappointment aversion