Scaling in financial prices 3

Cartoon Brownian Motions in Multifractal Time

Scaling Laws
Fractals
Financial Markets
Time series
Reviews
Author

Oren Bochman

Published

Friday, December 20, 2024

Keywords

Mandelbrot, Scaling, Multifractal Time, Power-law distributions, Fractional Brownian Motion, Lévy Stable Processes, Brownian Motion, Volatility Models

Scaling in financial prices: III. Cartoon Brownian Motions in Multifractal Time

TL-DR

In the paper “Scaling in financial prices: III. Cartoon Brownian Motions in Multifractal Time” (Mandelbrot 2001) Mandelbrot continus his exploration of fractal and multifractal geometry to model financial price fluctuations. It introduces “cartoon” functions – simplified, visually illustrative models – to capture key characteristics of financial prices, such as continuously varying volatility, discontinuities, and extreme price changes. These cartoons, parameterized within a “phase diagram,” offer intuitive representations of various existing models, including Mandelbrot’s own earlier work and the standard Brownian motion.

The core concept is Brownian motion in multifractal time (BMMT), a new model that accounts for the observed complexities in financial data better than previous approaches. The paper aims to provide a more accessible understanding of BMMT, emphasizing the limitations and strengths of the cartoon approximations while highlighting the importance of scaling invariance and the concept of trading time in capturing the dynamics of financial markets.

Mandelbrot, B. B. 2001. “Scaling in Financial Prices: III. Cartoon Brownian Motions in Multifractal Time.” Quantitative Finance 1 (4): 427–40. https://doi.org/10.1080/713665836.

Summary of the third paper

Main Themes:

  • Inadequacy of traditional financial models: Classic models like the Brownian motion fail to capture the key features of financial price variations, including fluctuating volatility and large, discontinuous jumps.
  • Fractal and multifractal geometry for financial modeling: Mandelbrot advocates for using fractal and multifractal models to better represent the complex and volatile nature of financial markets.
  • Cartoon Brownian Motions in Multifractal Time (BMMT): Mandelbrot presents BMMT, a new family of random processes, as a promising model for financial prices. BMMT builds upon the concepts of fractional Brownian motion and introduces the idea of “trading time.”
  • Recursive interpolation to create cartoons: The paper focuses on creating simplified, visual representations of BMMT, termed “cartoons,” using recursive interpolation techniques. These cartoons help visualize and understand the complex behaviors of BMMT.
  • Phase diagram for classifying price behaviors: A “phase diagram” is introduced to map the diverse behaviors generated by these cartoons based on two key parameters. Different regions in the phase diagram correspond to different types of price variations, including Fickian, unifractal, mesofractal, and multifractal.

Most Important Ideas/Facts:

  1. Limitations of standard Brownian motion: “Financial prices, such as those of securities, commodities, foreign exchange or interest rates, are largely unpredictable but one must evaluate the odds for or against some desired or feared outcomes, the most extreme being ‘ruin’. Those odds are essential to the scientist who seeks to understand the financial markets and other aspects of the economy. They must also be used as inputs for decisions concerning economic policy or institutional arrangements. To handle all those issues, the first step—but far from the last!—is to represent different prices’ variation by random processes that fit them well.” This quote highlights the need for a model that accurately reflects the inherent volatility of financial markets, which the Brownian motion fails to do.
  2. Recursive interpolation for financial modeling: “In the case of BMMT, the random walk has no direct counterpart. However, splendid cartoons in a very different style were developed and sketched in Mandelbrot (1997), chapter E6, and Mandelbrot (1999a), chapter N1. They are limits of discrete-parameter sequences of successive interpolations drawn on a continually refined temporal grid.” This explains the construction of simplified “cartoon” models using recursive interpolation, offering a visual and conceptual understanding of BMMT.
  3. The importance of the Hölder exponent (H): “This replacement of ratios of infinitesimals by ratios of logarithms of infinitesimals is an essential innovation. It was not directed by trial and error. Neither did its early use in classical ‘fine’ mathematical analysis suggest that H and many variants thereof could become concretely meaningful, quite the contrary. H became important because of its intimate connection with certain invariances.” This passage emphasizes the significance of the Hölder exponent in capturing the scaling and self-affinity properties of financial data, which traditional methods like derivatives fail to address.
  4. Distinction between absence of correlation and statistical independence: “Mathematicians know that whiteness does not express statistical independence, only absence of correlation. But the temptation existed to view that distinction as mathematical nit-picking. The existence of such sharply non-Gaussian white noises proves that the hasty assimilation of spectral whiteness to independence was understandable but untenable. White spectral whiteness is highly significant for Gaussian processes, but otherwise is a weak characterization of reality.” This section debunks the misconception that uncorrelated data implies independence. Multifractal cartoons with uncorrelated increments can still exhibit significant structure and dependence.
  5. Introduction of “trading time” to capture varying market speed: “Less mathematically oriented observers describe the panels at the bottom of figure 1 (both the real data and forgeries) as corresponding to markets that proceed at different ‘speeds’ at different times. This description may be very attractive but remains purely qualitative until ‘speed’ and the process that controls the variation of speed are quantified.” This introduces the concept of “trading time” as a way to quantify the subjective experience of varying market speed, a key element of the BMMT model.

Limitations of Cartoon Models:

  • Inability to fully predict power-law tails: The cartoons, being based on multinomial cascades, struggle to accurately represent the long-tailed distributions observed in real financial data.
  • Coupling of H and multifractal time: Unlike their continuous-time counterparts, the cartoon models impose a dependence between the Hölder exponent (H) and the multifractal time.
  • Singular perturbation in mesofractal cartoons: The specific construction of the mesofractal cartoons introduces an undesirable singular perturbation, highlighting a limitation of the three-interval symmetric generators.

Q&A

  1. What are Cartoon Brownian Motions and why are they used to model financial prices?

    Cartoon Brownian Motions (CBMs) are simplified, recursively generated functions designed to mimic the complex behavior of financial prices. They are called “cartoons” because they intentionally emphasize and distort certain features of real market data while remaining computationally simple and easy to visualize. These cartoons offer an intuitive way to understand the more complex model of Brownian Motion in Multifractal Time (BMMT), which is a more accurate but mathematically challenging model.

  2. What makes financial price data challenging to model, and how do CBMs address those challenges?

    Financial price data exhibits several characteristics that defy traditional modeling approaches:

    • Continually varying volatility: The magnitude of price fluctuations changes over time, exhibiting periods of high activity interspersed with periods of relative calm.
    • Discontinuity or concentration: Prices can jump abruptly, creating sharp spikes in price charts.
    • Non-normality: Many price changes fall far outside the expectations of the bell curve, signifying fat-tailed distributions.

    CBMs address these features by using a recursive interpolation scheme. A simple geometric shape, called the “generator,” is used to repeatedly refine a starting trend line, producing increasingly complex patterns that capture the roughness, variability, and discontinuity observed in financial markets.

  3. What is the “phase diagram” and how does it relate to different types of CBMs?

    The “phase diagram” is a two-dimensional map representing the space of possible CBM generators. Each point in this diagram corresponds to a unique generator shape, and different regions of the diagram give rise to distinct classes of CBMs:

    • Fickian: This corresponds to the classic Brownian Motion, where volatility is constant.
    • Unifractal: These CBMs exhibit long-range dependence or persistence, meaning past price changes influence future ones.
    • Mesofractal: These CBMs incorporate discontinuous jumps in prices.
    • Multifractal: This most general class combines features of the previous types, capturing the full complexity of financial price behavior.
  4. What are the key parameters controlling CBM behavior, and how do they manifest in price charts?

    CBMs are controlled by two main parameters:

    • H (Hölder exponent): Determines the degree of roughness or smoothness of the price curve. Higher H values indicate smoother trends, while lower values indicate more jagged, volatile behavior.
    • Generator shape: Dictates the pattern of price fluctuations. Different shapes lead to varying degrees of volatility clustering, jumps, and long-term trends.
  5. What is “trading time” and how does it explain varying volatility in multifractal CBMs?

    Trading time is a concept used to explain the non-uniform speed at which multifractal CBMs evolve. It contrasts with the regular “clock time” of physics. Multifractal CBMs move uniformly in their own subjective trading time, which can speed up or slow down relative to clock time. This variation in speed gives rise to the observed periods of high and low volatility in financial markets.

  6. What are the limitations of using CBMs to model financial prices?

    While offering valuable insights, CBMs are simplifications and possess certain limitations:

    • Constrained tail behavior: Unlike real price data, CBMs based on simple generators do not exhibit the extreme power-law tails associated with rare but significant market events.
    • Interdependence of parameters: In some CBMs, the choice of H and the multifractal trading time are linked, restricting the model’s flexibility compared to continuous-time models.
    • Singular perturbations: Certain mesofractal CBMs exhibit a peculiar behavior where slight changes in parameters lead to drastic changes in price patterns, which may not accurately reflect real market dynamics.
  7. How do CBMs relate to other fractal models of financial prices?

    CBMs serve as stepping stones to understand more complex continuous-time fractal models:

    • Fractional Brownian Motion (FBM): Unifractal CBMs are simplified versions of FBM, which incorporates long-range dependence.
    • Lévy Stable Processes (LSP): Mesofractal CBMs relate to LSP, which features discontinuous jumps in prices.
    • Brownian Motion in Multifractal Time (BMMT): This sophisticated model, for which multifractal CBMs are cartoons, combines FBM with a multifractal trading time to capture the full complexity of financial price dynamics.
  8. What are the practical implications of using CBMs in finance?

    CBMs, despite their limitations, offer a powerful tool for:

    • Visualizing market complexity: They provide an intuitive way to understand and communicate the irregular and multi-scale nature of financial price behavior.
    • Testing hypotheses: Their computational simplicity allows for rapid exploration of various market scenarios and model parameter sensitivity.
    • Developing trading strategies: While not directly predictive, CBMs can inform the design of trading algorithms that are robust to varying volatility and extreme price movements.

A Study Guide

Quiz

Instructions: Answer the following questions in 2-3 sentences each.

Question 1

Explain the concept of self-affinity and its relevance to financial market charts.

Self-affinity refers to the property of a geometric shape where its parts resemble the whole when scaled differently along different axes. Financial market charts exhibit self-affinity, meaning they appear similar when zoomed in or out, with time and price scales adjusted accordingly.

Question 2

Describe the construction process of a Fickian cartoon function using recursive interpolation.

The construction starts with a linear “initiator.” A three-interval “generator” replaces the initiator, creating an oscillation. Each generator interval is then recursively replaced by a scaled and potentially reflected version of the generator, continuing indefinitely.

Question 3

Why is the “square-root rule” relevant in the context of Fickian diffusion and Brownian motion?

The “square-root rule” states that the standard deviation of a sum of independent random variables scales with the square root of their number. In Fickian diffusion and Brownian motion, this rule manifests as the displacement of a particle being proportional to the square root of time.

Question 4

Define unifractality and explain how it differs from the Fickian case in terms of the Hölder exponent.

Unifractality implies a single Hölder exponent (H) governs the scaling behavior of the function at all scales. While Fickian behavior is a specific case of unifractality with H=1/2, other values of H within 0<H<1 lead to different types of unifractal behavior.

Question 5

Discuss the concept of persistence in unifractal cartoons and its implications for market behavior.

Persistence describes the tendency of a function to continue its current trend. In unifractal cartoons, H>1/2 signifies positive persistence, implying trends are more likely to continue. H<1/2 indicates anti-persistence, implying frequent trend reversals.

Question 6

How do mesofractal cartoons incorporate price discontinuity, and what is the role of the exponent α?

Mesofractal cartoons incorporate discontinuity by assigning a zero Hölder exponent (H2=0) to the middle interval of the generator. The exponent α (1/H̃) governs the distribution of jump sizes, with larger α indicating smaller and more frequent jumps.

Question 7

Explain the limitations of the Lévy stable exponent α exceeding 2.

When α>2 in L-stable processes, the sum of absolute values of jumps and moves diverges, leading to unbounded variation. Upon randomization, this divergence creates infinities that cannot be renormalized away, rendering the model mathematically inconsistent.

Question 8

What is the condition for multifractality, and how does it differ from unifractality and mesofractality?

Multifractality occurs when the Hölder exponents (Hm) associated with different generator intervals are all non-zero and distinct. This implies a multiplicity of scaling behaviors across different time scales, contrasting with the single H of unifractality and the two exponents of mesofractality.

Question 9

Explain the concept of “trading time” and its role in relating unifractal and multifractal cartoons.

“Trading time” is a subjective time scale that maps a unifractal cartoon onto a multifractal one. It allows for varying “speeds” of price changes, accounting for the observed volatility clustering in financial markets.

Question 10

Describe one limitation of multifractal cartoons compared to continuous-time multifractal models.

Multifractal cartoons generated from a three-interval symmetric generator constrain the choice of the unifractal oscillation and the multifractal time. They cannot be chosen independently, unlike continuous-time models where they can be independent random variables.

Essay Questions

  1. Compare and contrast the three fractal models of price variation proposed by Mandelbrot: the M 1963 model, the M 1965 model, and the M 1972/97 model. Discuss their strengths, weaknesses, and applicability to real financial data.
  2. Explain the “baby theorem” and its significance in understanding multifractal cartoons. How does the concept of trading time contribute to this understanding?
  3. Discuss the limitations of traditional “root-mean-square” volatility as a measure of price fluctuations in the context of multifractal models. What alternative measures of volatility are more appropriate, and why?
  4. Critically evaluate the use of cartoons as representations of complex financial phenomena. Discuss their advantages, limitations, and the potential pitfalls of relying solely on cartoon models.
  5. Elaborate on the concept of “spontaneous resonances” in financial markets and how multifractality might provide insights into understanding these resonances. How might this understanding contribute to improved economic policy and institutional arrangements?

Glossary of Key Terms

Self-affinity
A property of a geometric shape where its parts resemble the whole when scaled differently along different axes.
Recursive Interpolation
A process of repeatedly subdividing and interpolating a function using a predefined “generator” shape.
Fickian Diffusion
A type of diffusion characterized by the square-root rule, where the displacement of a particle is proportional to the square root of time.
Brownian Motion
A random process exhibiting continuous, erratic movement, often used to model financial price changes.
Hölder Exponent
(H)A measure of the local scaling behavior of a function, quantifying its roughness or smoothness.
Unifractality
A property of a function where a single Hölder exponent governs its scaling behavior at all scales.
Persistence
The tendency of a function to continue its current trend. Positive persistence indicates trends are more likely to continue, while negative persistence (anti-persistence) implies frequent trend reversals.
Mesofractality
A property of a function exhibiting discontinuities with a specific scaling behavior governed by an exponent α.
L-stable Process
A type of random process with heavy-tailed distributions, characterized by the exponent α, which determines the tail behavior.
Multifractality
A property of a function exhibiting a range of Hölder exponents across different time scales.
Volatility Clustering
The tendency of large price fluctuations to be followed by other large fluctuations, and small fluctuations to be followed by small fluctuations.
Trading Time
A subjective time scale that accounts for the varying “speeds” of price changes in financial markets.
Compound Function
A function created by composing two or more functions, where the output of one function becomes the input of another.
Subordination
A specific type of compounding where the inner function is a random process with independent increments.
Multinomial Cascade
A specific mathematical construction used to generate multifractal measures with a limited range of possible outcomes.
Lacunarity
A measure of the distribution of gaps or holes in a fractal or multifractal structure.
Singular Perturbation
A mathematical concept where a small change in a parameter leads to a large and discontinuous change in the solution of a problem.
Dimension Anomalies
Deviations from the expected relationship between fractal dimension and other properties of a fractal, often arising from complex scaling behaviors.LocalizationThe concentration of a function’s values within a specific range or region.

Citation

BibTeX citation:
@online{bochman2024,
  author = {Bochman, Oren},
  title = {Scaling in Financial Prices 3},
  date = {2024-12-20},
  url = {https://orenbochman.github.io/posts/2024/2024-11-28-misbahaviour-of-markets/part3/},
  langid = {en}
}
For attribution, please cite this work as:
Bochman, Oren. 2024. “Scaling in Financial Prices 3.” December 20, 2024. https://orenbochman.github.io/posts/2024/2024-11-28-misbahaviour-of-markets/part3/.