Broken Symmetry of Stock Returns -- a Modified Jones-Faddy Skew t-Distribution

This paper proposes that the negative skew and positive mean observed in stock returns stem from the broken symmetry of stochastic volatility between gains and losses, a phenomenon effectively modeled by a modified Jones-Faddy skew t-distribution that is validated using daily S&P 500 return data.

The Gist

Here is an explanation of the paper "Broken Symmetry of Stock Returns" using simple language, analogies, and metaphors.

The Big Picture: Why Stocks Don't Play Fair

Imagine the stock market as a giant, chaotic playground where a ball (the stock price) is constantly being kicked around. Most physics models assume this ball is kicked by a perfectly fair wind. If you kick it left, it goes left; if you kick it right, it goes right, with equal force and probability. This is what scientists call symmetry.

However, the authors of this paper looked at 45 years of data (1980–2025) from the S&P 500 (a basket of 500 big US companies) and realized: The wind isn't fair.

The market has a "broken symmetry." It behaves differently when things go up (gains) than when they crash (losses). Specifically:

1. The Trend: Over time, the ball generally rolls uphill (positive average return).
2. The Crash: When it falls, it falls harder and faster than it rises.
3. The Frequency: There are more days of small gains than days of small losses, but the losses that do happen are more violent.

The paper's goal was to build a mathematical "map" that captures this unfairness, rather than pretending the market is a fair coin toss.

---

The Old Map vs. The New Map

1. The Old Map: The "Student-t" Distribution

For a long time, physicists used a standard map called the Student-t distribution.

• The Analogy: Imagine a bell curve that is slightly flatter on top and has "fatter tails" than a normal bell curve. It accounts for the fact that extreme events (crashes or massive rallies) happen more often than a standard bell curve predicts.
• The Problem: This map is perfectly symmetrical. It assumes the chance of a massive gain is exactly the same as a massive loss, just in the opposite direction. It also assumes the average is zero (once you remove the general upward trend).
• Reality Check: Real stock data doesn't look like this. The "loss" tail is heavier (more dangerous), and the "gain" side is slightly shifted. The old map is too rigid.

2. The "Half-Student" Idea (The Split Approach)

The authors first tried a simple fix: "Let's just use two different maps glued together."

• The Analogy: Imagine a seesaw. On the left side (losses), we use a heavy, stiff spring. On the right side (gains), we use a bouncy, light spring.
• The Result: This worked mathematically to describe the different shapes of gains and losses. However, it felt "artificial." It was like saying, "The rules of physics change depending on which side of the line you are standing on." It didn't feel like one single, organic system. Also, it struggled to explain why the average return was positive.

3. The New Map: The "Modified Jones-Faddy" (mJF)

This is the paper's main contribution. They found a more elegant, single formula that naturally bends to fit the data.

• The Analogy: Think of a slingshot.
  • In a normal symmetric world, the rubber bands on both sides are identical.
  • In the stock market, the rubber band on the "loss" side is made of a different, stretchier material (it allows for bigger, more frequent drops). The rubber band on the "gain" side is tighter.
  • The Modified Jones-Faddy (mJF) distribution is the mathematical description of this specific, uneven slingshot. It is one single formula that knows how to stretch differently in two directions.

Why Does This Matter? (The "Broken Symmetry")

The authors argue that this "broken symmetry" comes from Stochastic Volatility.

• The Metaphor: Imagine the stock price is a boat, and volatility is the size of the waves.
• The paper suggests that the "rules" for how big the waves get are different when the boat is going up versus when it is going down.
• When the market is falling, the "wave generator" (volatility) seems to have a different setting (a different parameter called $\alpha$) than when the market is rising. This causes the "loss" tail to be heavier (more extreme crashes) and the "gain" tail to be slightly lighter.

The Results: Does the New Map Work?

The authors tested this new "slingshot" map against the actual S&P 500 data from 1980 to 2025.

1. The Fit: The new map (mJF) hugged the real data much better than the old symmetrical maps. It correctly predicted:
   • The positive average return (the boat drifts uphill).
   • The negative skew (the boat is more likely to hit a huge wave going down).
   • The "fat tails" (the extreme crashes).
2. The Tails: They looked specifically at the "tails" of the distribution (the rare, extreme events). They found that the new map accurately describes how often these "Dragon Kings" (massive outliers) happen.
3. Simplicity: Interestingly, the version of the map that assumed the "average wave size" was the same for gains and losses (mJF1) worked just as well as the more complex version. This suggests that while the shape of the volatility changes, the average intensity might be similar, but the distribution of that intensity is what breaks the symmetry.

The Takeaway

In plain English:
Stock markets aren't fair. They have a built-in bias where crashes are more violent and frequent than rallies, even though the market generally goes up over time.

Old mathematical models tried to force the market into a symmetrical box, which didn't work well. This paper introduces a new, flexible mathematical shape (the Modified Jones-Faddy distribution) that acts like a one-sided slingshot. It acknowledges that the market has different rules for gains and losses, allowing us to model risk and returns much more accurately.

Why you should care:
If you are an investor or a risk manager, using the "old map" might make you think a crash is as rare as a lottery win. Using this "new map" tells you that the "loss" side of the slingshot is stretchier, meaning you need to be more prepared for the possibility of a sudden, deep drop than the old models suggested.

Technical Summary

Here is a detailed technical summary of the paper "Broken Symmetry of Stock Returns - a Modified Jones-Faddy Skew t-Distribution" by Shao et al.

1. Problem Statement

The paper addresses a fundamental discrepancy between standard financial models and empirical market data regarding the distribution of stock returns.

• The Symmetry Issue: Standard models based on stochastic volatility (specifically the multiplicative model leading to a Student's $t$-distribution) predict a symmetric distribution of returns. However, empirical data (specifically S&P500 daily returns from 1980–2025) exhibits broken symmetry:
  • Positive Mean: The distribution has a positive drift.
  • Negative Skew: The distribution is skewed to the left.
  • Asymmetric Tails: Losses exhibit "heavier" tails (slower power-law decay) than gains.
  • Frequency Imbalance: There are more gain events than loss events.
• The Limitation of Existing Models:
  • The standard Student's $t$-distribution is symmetric and cannot capture skewness or the positive mean.
  • A "Half Student-t" approach (mixing two separate Student's $t$ distributions for gains and losses) can capture the tail asymmetry but fails to produce a positive mean and is not an "organic" single distribution derived from first principles.

2. Methodology

The authors propose a framework that moves from stochastic differential equations (SDEs) to a modified probability distribution.

A. Theoretical Foundation (SDEs)

The authors start with the standard SDE framework for stock returns ($x_t$) and stochastic volatility ($v_t = \sigma_t^2$):

1. Return Dynamics: $dx_t = \sigma_t dW$ (where $dW$ is a Wiener process).
2. Volatility Dynamics: A mean-reverting process where $g(v_t)$ determines the noise term. The authors utilize the multiplicative model ($g(v_t) = \kappa v_t$), which yields an Inverse Gamma distribution for variance and a Student's $t$-distribution for returns.

B. Proposed Solution: Broken Symmetry via Parameter Differentiation

To explain the broken symmetry, the authors hypothesize that the parameters governing stochastic volatility differ for gains and losses.

• Half Student-t: They first explore a mixture model where gains and losses are fitted separately with different parameters ($\alpha_g, \theta_g$ vs. $\alpha_l, \theta_l$). While this fits the tails, it fails to capture the positive mean naturally and lacks a unified functional form.
• Modified Jones-Faddy (mJF) Distributions: The core contribution is the application of a Modified Jones-Faddy Skew $t$-distribution. This is a single, organic distribution derived from the Jones-Faddy family, adapted to reflect the broken symmetry of volatility parameters.

The authors analyze two specific variants:

1. mJF1: Assumes a single mean stochastic volatility parameter ($\theta$) for both gains and losses, but distinct shape parameters ($\alpha_g$ for gains, $\alpha_l$ for losses). This allows for different tail exponents while maintaining a unified structure.
2. mJF2: A generalization allowing distinct mean volatility parameters ($\theta_g$ and $\theta_l$) for gains and losses.

C. Analytical Framework

The paper derives analytical expressions for:

• Probability Density Function (PDF) and Cumulative Distribution Function (CDF).
• Statistical moments: Mean ($m_1$), Variance ($m_2$), Mode ($m$), and Median ($e_m$).
• Skewness coefficients ($\zeta_1, \zeta_2$).
• Power-law tail exponents for the Complementary CDF (CCDF).

3. Key Contributions

1. Unified Distribution for Asymmetric Returns: The paper introduces the mJF1 distribution as the most effective single organic distribution to model stock returns. It successfully captures:
   • The positive mean (via a location parameter $\mu$).
   • The negative skew.
   • The distinct power-law exponents for gains and losses (heavier tails for losses).
2. Physical Interpretation of Skewness: The authors argue that the negative skew and heavier loss tails are physically driven by the broken symmetry of stochastic volatility, specifically where the parameter $\alpha_l$ (governing losses) is smaller than $\alpha_g$ (governing gains). A smaller $\alpha$ results in a slower power-law decay (heavier tail).
3. Statistical Validation: The study provides rigorous statistical testing (U-tests, confidence intervals, p-values) comparing the mJF models against the S&P500 data, demonstrating that mJF1 outperforms the standard Student's $t$ and the Half Student-t mixture.

4. Results

The authors fitted the models to daily S&P500 returns (1980–2025).

• Parameter Fitting (Table 1):
  • mJF1 yielded a location parameter $\mu \approx 8.465 \times 10^{-4}$, successfully capturing the positive mean.
  • The shape parameters showed $\alpha_g > \alpha_l$ (specifically $7.924 \times 10^{-5}$vs$6.416 \times 10^{-5}$), confirming the hypothesis that losses have a "slower" decay (heavier tail).
• Statistical Moments (Table 2):
  • Mean ($m_1$): mJF1 produced a positive mean ($4.06 \times 10^{-5}), closely matching the empirical S&P500 mean (4.38 \times 10^{-5}$). In contrast, the Half Student-t model produced a negative mean, failing to match reality.
  • Skewness ($\zeta_1$): mJF1 achieved a skewness of $-3.96 \times 10^{-2}$, aligning well with the empirical value of $-7.70 \times 10^{-3}$ (noting that the second skewness coefficient $\zeta_2$ was even closer).
  • Tail Exponents: The power-law exponents for mJF1 were $-3.12$ (gains) and $-2.90$ (losses), closely matching the empirical $-3.14$ and $-2.91$. This confirms the model captures the asymmetry in tail thickness.
• Model Comparison:
  • mJF2 (with separate $\theta_g, \theta_l$) offered no significant improvement over mJF1, suggesting a single mean volatility parameter is sufficient.
  • Half Student-t failed to capture the positive mean and is considered less "organic."
  • Standard Student-t failed to capture any asymmetry.

5. Significance and Conclusion

• Best Candidate: The mJF1 distribution is identified as the optimal candidate for modeling daily stock returns. It balances physical interpretability (rooted in volatility asymmetry) with statistical accuracy.
• Mechanism of Asymmetry: The paper provides a compelling argument that the "broken symmetry" of stock returns is not merely a statistical artifact but stems from the underlying stochastic volatility governing gains and losses having different parameters. Specifically, the lower $\alpha$ value for losses explains the "fat tails" of market crashes.
• Future Work: The authors suggest extending this analysis to other indices (NASDAQ, Russell, international markets) and investigating accumulated returns (realized volatility) to see if the symmetry breaking persists over longer time horizons. They also highlight the need for a first-principles derivation of the mJF distribution directly from SDEs, which remains an open challenge.

In summary, the paper successfully bridges the gap between stochastic volatility theory and empirical market data by proposing a modified skew $t$-distribution that naturally accounts for the positive mean, negative skew, and asymmetric tail behavior of stock returns.