This book reflects a significant step in volatility modeling with a clear focus on financial applications. It offers a nice combination of theoretical results combined with fitting the corresponding models to real financial data.
From a practical point of view, it has been long known that (log) asset prices are not adequately described by Brownian Motion (BM) or Fractional BM. For instance, take the fact that typically, low-frequency (weekly or monthly) return distribution has much thinner tails than high-frequency (daily or hourly) return distribution. Such absence of "self-similarity" across frequencies is not consistent with BM or Fractional BM.
To capture this effect (among many others), they propose to model the (log) asset price as a Multifractal Process. Such process is characterized by so-called "scaling function" which can be estimated from the data. One may think of Multifractal Processes as an extended class of stochastic processes that includes self-similar BM / Fractional BM. In particular, for self-similar processes the scaling function has to be linear. However, estimations based on real currency and equity data (see Chapter 8) do not produce a linear scaling function. Therefore, the hypothesis of self-similarity (also called "unifractality") of the (log) asset price doesn't hold.
Apparently, the limitations of self-similar processes have been known for a while, and many popular volatility models (such as GARCH or FIGARCH) address them to a certain degree. However, numerical results show that the Multifractal Model is a better fit to the data in terms of scaling function.
In practice, the multifractal approach is implemented as so-called Markov-Switching Multifractal model (MSM) in discrete time. Markov-Switching models (pioneered by Hamilton, see Time Series Analysis) are based on the idea is that volatility (and possibly drift) are dependent on the unobserved state variable that follows a Markov process. MSM extends that idea by imposing certain restrictions on the transition matrix, thus reducing the dimensionality. The physical meaning of the restrictions is that different economic factors (technology shocks, business cycles, liquidity shocks) affect the volatility on different time scales. The volatility impact from one economic factor can be a lot more lasting than that from another factor.
The authors demonstrate that MSM model accounts for such data features as:
1) short- and long-range dependence in volatility;
2) fat tails of return distribution;
3) volatility jumps.
Again, many previously known models account for these effects to a certain extent, so a comparison to some benchmark models is necessary. Fitting MSM model to daily currency data via Maximum Likelihood (Chapter 3) shows that MSM is superior to:
1) GARCH-t ("t" means that the error term has a t-distribution)
2) Markov-Switching GARCH-t
3) FIGARCH-t
Personally, I would have liked to see how well MSM competes with some models based on Extreme Value Distribution, but no examples are provided.
There have been many complaints in the reviews of the popular book of Mandelbrot (see The Misbehavior of Markets: A Fractal View of Financial Turbulence) that few "implementation details" had been provided. Numerical examples in Calvet and Fisher clearly show how to apply Mandelbrot's ideas to real data and where exactly the new framework surpasses the existing volatility models.
Other chapters include multivariate volatility modeling (again, MSM is superior to multivariate CC-GARCH) and application of MSM to asset pricing theory. Therefore, I can highly recommend this book to people interested in the latest advances in volatility modeling.