Evaluating Multifractal Volatility Models for FX Regimes
Summary
The document frames a research question about whether Markov-switching multifractal models are suitable for detecting volatility regime shifts in foreign exchange rates across multiple time horizons, and whether newer models improve on them. It lists the desired properties: multifrequency behavior with long-memory-like persistence and abrupt volatility changes, components with differing persistence, asymmetric learning about rising and falling risk, sensitivity to information quality, and daily seasonality in high-frequency FX data.
It also highlights parsimony, likelihood-based or moment-based estimation, discrete- and continuous-time formulations, and useful volatility forecasts and value-at-risk estimates as comparison criteria. The text describes claimed MSM features, including representing many states with a small parameter set, but presents no comparative results or recommended alternative. It is therefore a checklist for model evaluation rather than evidence that a particular model performs best, especially in real-time applications.
Key ideas
- The research question concerns volatility regime shifts in FX across multiple time horizons.
- The listed criteria include long-memory-like persistence, abrupt volatility changes, and daily seasonality.
- The document calls for models that capture asymmetric updates to perceived volatility risk.
- Parsimony and likelihood- or moment-based estimation are among the stated evaluation criteria.
- The text asks about newer alternatives but supplies no comparative evidence or recommendation.
Tags
Full text
# Markov-Switching Multifractal and FX Rates # Markov-Switching Multifractal and FX Rates Is there a better model than Markov-Switching Multifractal (MSM) for detecting regime shifts in FX rates across multiple time horizons? I am especially interested in the different aspects of the question that have been covered in Multifractal Volatility: Theory, Forecasting, and Pricing: #### Multifrequency Equilibrium Ability to detect regime switches across multiple time horizons, from seconds to years, and to "imitate one of the defining features of long memory, a hyperbolic decline of the autocovariogram." The combination of long-memory behavior with sudden volatility movements is one of the qualities that make MSM very attractive. #### Volatility persistence Ability to deal with volatility persistence components that have different degrees of persistence. #### Persistence skewness “We observe that investors may learn quickly about volatility increases, because a single extreme fluctuation is highly improbable with low volatility. By contrast, learning about reduced risk takes time because observations near the mean are a relatively likely outcome regardless of the true state. Thus bad news about volatility is incorporated into prices quickly, while good news is assimilated slowly.” The MSM model takes this skewness into account. #### Information quality “Skewness increases and kurtosis falls as information quality deteriorates.” #### Daily seasonality High-frequency data for FX rates show strong patterns of daily seasonality. #### Parsimony Ability to model the process with very few variables. For example, MSM can model $2^k$ states where $k$ is the number of time horizons with just 4 parameters. This is a direct result of the observation that volatility shocks have the same magnitude at all time scales. #### Closed-form parameters estimation Ability to estimate parameters by maximixing the closed-form likelihood of the return series. Alternatively, return moments can be used to quickly calibrate or estimate the model. #### Discrete and continuous-time versions Ability to easily switch between discrete and continuous-time versions of the model. #### Good performance Ability to provide good forecasts of volatility and generate reliable estimates of the value-at-risk in a position or portfolio of assets. Multifractal Volatility was written in 2008. Have there been more recent developments that should be taken into account? Do significantly-different models provide much better results along the lines outlined above, especially when dealing with real-time data?
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.