Choosing Stochastic Volatility and Jump Models for Equity Returns
Summary
The document discusses model features that can better represent equity-market behavior than a basic constant-volatility diffusion. It proposes a flexible stochastic-volatility framework in which log prices follow Brownian motion, volatility varies over time and tends to revert, and volatility can be negatively associated with price moves. These features are linked to return tails, volatility clustering, and skew. Random jumps in prices and volatility, often represented with compound Poisson processes, are suggested for abrupt market moves.
The answer emphasizes that model choice depends on the intended use. For real-world forecasting, estimating or calibrating the model to historical observations is difficult. For risk-neutral pricing, fitting vanilla option prices does not ensure that the model will produce plausible changes in the implied-volatility smile or reliable prices for exotic products. The document offers a qualitative modeling recommendation rather than a comparison of calibrated models or evidence that one specification is universally best; the appropriate model depends on the application and validation criteria.
Key ideas
- Stochastic volatility can represent changing return dispersion and volatility clustering.
- Negative association between volatility and spot returns can help explain equity volatility skew.
- Jump processes can represent abrupt moves in prices or volatility.
- A model suitable for historical forecasting may differ from one calibrated for option pricing.
- Matching vanilla option prices alone does not establish reliable exotic-option prices.
Tags
Full text
# Modeling the Stock Market # Modeling the Stock Market Hi I was wondering what is the model that best describes the price movement of the stock market? - A Brownian motion Process with drift? - An Ornstein Uhlenbeck_process? (where the long term mean is described by an ax+b function for the drift component) - A model that uses stochastic volatility? - Jump Diffusion Models? - Or something else? (I placed these 4 points to act as multiple choice bullets. I don't know how to make this question more specific, that is what I want to know, which model best fits the stock market. Which model more realistically fits the stock market, which one will show volatility clustering, price appreciation, market shocks(jump diffusion), show leptokurtic distributions, mean reversion etc. I see many models used but I haven't yet seen which one is standard for realistic simulations. I placed the 4 points up there as a guide to which class of mathematical models I'm referring to. I realize that there can be many good ways to answer this question. I believe though that only a few models will approximate the way the stock market moves and behaves. So IDK.) Thanks ## Answer by Quantuple (score 1, accepted) https://quant.stackexchange.com/a/26273 Given the well-known stylised facts of equity markets, I would go for a generic stochastic volatility model where - log-asset prices, hence geometric returns, are driven by a standard Brownian motion (although this would explain the lack of returns' auto-correlation, it would also boil down to assuming their independence, which is a stronger assumption). - volatility of returns is stochastic (would explain heavy tails), mean-reverting (would explain volatility clustering), negatively correlated to the spot evolution (would explain skew). Some may even consider modeling (log-)volatility using fractional Brownian motions instead of standard Brownian motions (better emprical behaviour). - volatility and price should be subject to random jumps both in timing and in size (blame it on Murphy's law). Common practice is to use compound Poisson processes to model that feature. Now, if you are working under the real-world measure and looking at using such a model for forecasting purpose, the complex parts is how exactly would you calibrate such a model to historical data? If you are on the other hand working under the risk-neutral measure and that you (manage to decently) calibrate your model to observed vanilla option prices, the question is then, how to can you guarantee that your model will also embed a plausible forward smile dynamics so that it produces decent prices for more exotic structures?
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.