Long Memory / Scaling Laws in Return Volatility

The 25-year accumulation of evidence for long memory / fractional integration / self-similarity / scaling laws in financial asset return volatility continues unabated.  For the latest see this nice new paper from Bank of Portugal, in particular its key Table 6. Of course the interval estimates of the fractional integration parameter "d" are massively far from both 0 and 1 -- that's the well-known long memory. But what's new and interesting is the systematic difference in the intervals depending on whether one uses absolute or range-based volatility. The absolute d intervals tend to be completely below 1/2 (0<d<1/2 corresponds to covariance-stationary dynamics), whereas the range-based d intervals tend to include 1/2 (1/2<d<1 corresponds to mean-reverting but not covariance- stationary dynamics, due to infinite unconditional variance). 

Realized vol based on the range is less noisy than realized vol based on absolute returns. But least noisy of all, and not considered in the paper above, is realized vol calculated directly from high-frequency return data (HFD-vol), as done by numerous authors in recent decades. Interestingly, recent work for HFD-vol also reports d intervals that tend to poke above 1/2. See this earlier post.

Long Memory in Realized Volatility

A noteworthy aspect of long memory in realized asset return volatility is that in many leading cases it's basically undeniable on the basis of a variety of evidence -- the question isn't existence but rather strength.  Hence it's useful to have a broad and comparable set of state-of-the-art (local Whittle) estimates together in one place, as in the interesting paper below.  For the most part it gets d in [.4, .6], consistent with my personal experience of d usually around .45, in the covariance stationary (finite variance) region d<.5, but close to the boundary.

http://d.repec.org/n?u=RePEc:han:dpaper:dp-601&r=ecm

The Memory of Volatility

Date:	2017-07
By:	Wenger, Kai ; Leschinski, Christian ; Sibbertsen, Philipp
The focus of the volatility literature on forecasting and the predominance of the conceptually simpler HAR model over long memory stochastic volatility models has led to the fact that the actual degree of memory estimates has rarely been considered. Estimates in the literature range roughly between 0.4 and 0.6 - that is from the higher stationary to the lower non-stationary region. This difference, however, has important practical implications - such as the existence or non-existence of the fourth moment of the return distribution. Inference on the memory order is complicated by the presence of measurement error in realized volatility and the potential of spurious long memory. In this paper we provide a comprehensive analysis of the memory in variances of international stock indices and exchange rates. On the one hand, we find that the variance of exchange rates is subject to spurious long memory and the true memory parameter is in the higher stationary range. Stock index variances, on the other hand, are free of low frequency contaminations and the memory is in the lower non-stationary range. These results are obtained using state of the art local Whittle methods that allow consistent estimation in presence of perturbations or low frequency contaminations.
Keywords:	Realized Volatility; Long Memory; Perturbation; Spurious Long Memory
JEL:	C12 C22 C58 G15
URL:	http://d.repec.org/n?u=RePEc:han:dpaper:dp-601&r=ecm

Dense Data for Long Memory

From the last post, you might think that efficient learning about low-frequency phenomena requires tall data. Certainly efficient estimation of trend, as stressed in the last post, does require tall data. But it turns out that efficient estimation of other aspects of low-frequency dynamics sometimes requires only dense data. In particular, consider a pure long memory, or "fractionally integrated", process, \( (1-L)^d x_t = \epsilon_t \), 0 < \( d \) < 1/2. (See, for example, this or this.) In a general \( I(d) \) process, \(d\) governs only low-frequency behavior (the rate of decay of long-lag autocorrelations toward zero, or equivalently, the rate of explosion of low-frequency spectra toward infinity), so tall data are needed for efficient estimation of \(d\). But in a pure long-memory process, one parameter (\(d\)) governs behavior at all frequencies, including arbitrarily low frequencies, due to the self-similarity ("scaling law") of pure long memory. Hence for pure long memory a short but dense sample can be as informative about \(d\) as a tall sample. (And pure long memory often appears to be a highly-accurate approximation to financial asset return volatilities, as for example in ABDL.)

Regularization for Long Memory

Two earlier regularization posts focused on panel data and generic time series contexts. Now consider a specific time-series context: long memory. For exposition consider the simplest case of a pure long memory DGP,  \( (1-L)^d y_t = \varepsilon_t \) with  \( |d| < 1/2  \).  This \( ARFIMA(0,d,0) \) process is  is \( AR(\infty) \) with very slowly decaying coefficients due to the long memory. If you KNEW the world was was \(ARFIMA(0,d,0)\) you'd just fit \(d\) using GPH or Whittle or whatever, but you're not sure, so you'd like to stay flexible and fit a very long \(AR\) (an \(AR(100) \), say). But such a profligate parameterization is infeasible or at least very wasteful. A solution is to fit the \(AR(100) \) but regularize by estimating with ridge or a LASSO variant, say.

Related, recall the Corsi "HAR" approximation to long memory. It's just a long autoregression subject to coefficient restrictions. So you could do a LASSO estimation, as in Audrino and Knaus (2013). Related analysis and references are in a Humboldt University 2015 master's thesis.)

Finally, note that in all of the above it might be desirable to change the LASSO centering point for shrinage/selection to match the long-memory restriction. (In standard LASSO it's just 0.)

Long Memory Stochastic Volatility

Check out Mark Jensen's new paper.  Long memory is a key feature of realized high-frequency asset-return volatility, yet it remains poorly understood. Jensen's approach may help change that. Of particular interest are: (1) its ability to handle seamlessly d in [0, 1[, despite the fact that the unconditional variance is infinite for d in ].5, 1[, and (2) closely related, the important role played by wavelets. 

Details:

Robust estimation of nonstationary, fractionally integrated, autoregressive, stochastic volatility

Date:	2015-11-01
By:	Jensen, Mark J. (Federal Reserve Bank of Atlanta)
Empirical volatility studies have discovered nonstationary, long-memory dynamics in the volatility of the stock market and foreign exchange rates. This highly persistent, infinite variance—but still mean reverting—behavior is commonly found with nonparametric estimates of the fractional differencing parameter d, for financial volatility. In this paper, a fully parametric Bayesian estimator, robust to nonstationarity, is designed for the fractionally integrated, autoregressive, stochastic volatility (SV-FIAR) model. Joint estimates of the autoregressive and fractional differencing parameters of volatility are found via a Bayesian, Markov chain Monte Carlo (MCMC) sampler. Like Jensen (2004), this MCMC algorithm relies on the wavelet representation of the log-squared return series. Unlike the Fourier transform, where a time series must be a stationary process to have a spectral density function, wavelets can represent both stationary and nonstationary pr! ocesses. As long as the wavelet has a sufficient number of vanishing moments, this paper's MCMC sampler will be robust to nonstationary volatility and capable of generating the posterior distribution of the autoregressive and long-memory parameters of the SV-FIAR model regardless of the value of d. Using simulated and empirical stock market return data, we find our Bayesian estimator producing reliable point estimates of the autoregressive and fractional differencing parameters with reasonable Bayesian confidence intervals for either stationary or nonstationary SV-FIAR models.
Keywords:	Bayes; infinite variance; long-memory; Markov chain Monte Carlo; mean-reverting; wavelets
JEL:	C11 C14 C22
URL:	http://d.repec.org/n?u=RePEc:fip:fedawp:2015-12&r=ecm