A Socio-Finance Model: Inference and empirical application

Jørgen Vitting Andersen, Ioannis D. Vrontos, Πέτρος Δελλαπόρτας, Serge GalamHAL 2014

TL;DR

This paper introduces and applies a novel "socio-finance" model that links social interactions and opinion dynamics to financial market behavior. By incorporating how people influence each other's beliefs, the model can better explain and predict market returns and volatility, especially in situations with non-normal return distributions.

What Problem Does This Paper Address?

Traditional financial models often struggle to capture the full complexity of market dynamics, particularly phenomena like volatility clustering and fat tails (i.e., extreme events occurring more frequently than predicted by normal distributions). These models typically treat market participants as rational, independent agents. However, real-world markets are influenced by social interactions, herd behavior, and the spread of opinions, which can significantly impact asset prices and their volatility. This paper addresses this gap by proposing a model that explicitly accounts for these socio-psychological factors.

Methods

The core of the paper is the development and application of a "socio-finance" model. This model builds upon existing opinion formation models (like the Galam model) where individuals in a population can hold either a "bullish" or "bearish" opinion. The key innovation is how these opinions evolve and influence market outcomes:

  • Opinion Dynamics: Individuals interact in random subgroups, and their opinions can be swayed by the majority opinion within the group. The probability of adopting a particular opinion is not static but can change based on market performance.
  • Market Impact: The model links changes in the aggregate "bullishness" proportion to the market's price return. It also models time-varying conditional volatility, where the market's reaction to news is amplified or dampened by the prevailing sentiment.
  • Statistical Inference: The authors employ a maximum likelihood estimation (MLE) approach to estimate the model's parameters. They consider both normal and Student-t distributions for the error process, with the latter being more appropriate for capturing fat tails.
  • Simulation and Application: The model's performance is evaluated through extensive simulations and then applied to real-world data from the European Union Banks index.

Datasets

The paper uses two main types of data:

  • Simulated Data: The authors generate synthetic time series data from their socio-finance model under various parameter settings and sample sizes (T=2000 and T=5000) to test the estimation procedure.
  • Real-World Data: The empirical application uses daily price data for the European Union Banks five-year index from January 1, 2008, to October 7, 2013 (1505 data points).

Key Insights and Ideas

  • Social Influence Matters: The study strongly suggests that social interactions and the dynamics of opinion formation are crucial for understanding financial markets, going beyond purely rational agent models.
  • Opinion Dynamics Predict Volatility: Changes in the aggregate "bullishness" proportion are shown to be a significant driver of both market returns and time-varying conditional volatility.
  • Fat Tails are Real: The empirical application highlights that financial return series, like the EU Banks index, often exhibit fat tails, making the Student-t distribution a more appropriate choice than the normal distribution for modeling their behavior.
  • Model Captures Phenomena: The socio-finance model, particularly with Student-t errors, successfully captures key market phenomena like volatility clustering.

Key Results

  • Simulation Performance: The maximum likelihood estimation procedure generally provides estimates that are close to the true simulated parameter values, with good convergence properties, especially as sample size increases.
  • Empirical Application (EU Banks Index):
    • The model, under both normal and Student-t error assumptions, can explain price movements and time-varying volatility.
  • The Student-t error distribution provides a better fit, as indicated by statistically significant parameter estimates and a more appropriate capture of the data's tail properties. The estimated degrees of freedom for the Student-t distribution (v=2.312) confirm the presence of heavy tails.
  • The estimated conditional volatilities from the Student-t socio-finance model were higher than those from the normal model, suggesting it better captures periods of extreme market movement.
  • Tests for normality on the standardized residuals rejected the null hypothesis, reinforcing the need for a non-normal distribution like the Student-t.

Limitations and Future Work

  • Limitations: The paper acknowledges that the normality assumption is often violated in financial applications, which is why they propose and validate the Student-t distribution. The study also implicitly highlights the limitations of standard GARCH models in capturing the full dynamics of the EU Banks index data, especially regarding tail behavior.
  • Future Work: The authors suggest that this new class of socio-finance models offers a promising avenue for empirical applications. They believe many interesting questions remain open and that further research can explore various topics within this framework.

Summary updated

Abstract

In this report we show the empirical application of our socio-finance model introduced in Andersen, Vrontos, Dellaportas and Galam (2014).
BibTeX
@inproceedings{andersen2014socio,
  author = {Jørgen Vitting Andersen and Ioannis D. Vrontos and Πέτρος Δελλαπόρτας and Serge Galam},
  title = {{A Socio-Finance Model: Inference and empirical application}},
  booktitle = {HAL},
  year = {2014},
}
Figure