Paper: arXiv 2407.10659
Authors: Carsten H. Chong, Viktor Todorov
Abstract
We develop a nonparametric test for deciding whether volatility of an asset follows a standard semimartingale process, with paths of finite quadratic variation, or a rough process with paths of infinite quadratic variation. The test utilizes the fact that volatility is rough if and only if volatility increments are negatively autocorrelated at high frequencies. It is based on the sample autocovariance of increments of spot volatility estimates computed from high-frequency asset return data. By showing a feasible CLT for this statistic under the null hypothesis of semimartingale volatility paths, we construct a test with fixed asymptotic size and an asymptotic power equal to one. The test is derived under very general conditions for the data-generating process. In particular, it is robust to jumps with arbitrary activity and to the presence of market microstructure noise. In an application of the test to SPY high-frequency data, we find evidence for rough volatility.
Complexity vs Empirical Score
- Math Complexity: 8.0/10
- Empirical Rigor: 7.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced stochastic calculus, central limit theorems, and high-frequency econometrics with heavy formal proofs and abstract probability theory, giving it high math complexity. It demonstrates strong empirical rigor through the application of the test to real high-frequency SPY data, robustness to market microstructure noise and jumps, and discussion of finite-sample performance.
Research Flowchart
flowchart TD A["Research Goal<br>Determine if volatility is<br>Rough or Semimartingale"] --> B["Key Methodology<br>Compute sample autocovariance<br>of spot volatility increments"] B --> C["Data Inputs<br>High-frequency asset returns<br>SPY data"] C --> D["Computational Processes<br>Feasible CLT under Null<br>Robust to jumps & noise"] D --> E["Key Findings<br>Evidence of Rough Volatility<br>in SPY data"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes a nonparametric test for rough volatility based on the sample autocovariance of increments of spot volatility estimates from high‑frequency data.
Source sentence
We develop a nonparametric test for deciding whether volatility of an asset follows a standard semimartingale process, with paths of finite quadratic variation, or a rough process with paths of infinite quadratic variation. More specifically, we propose a test based on the sample autocovariance of increments of spot volatility estimates.
Assumes 6
- The test is robust to jumps of arbitrary activity in the price process. model
Source sentence
In particular, it is robust to jumps with arbitrary activity and to the presence of market microstructure noise. In particular, all three processes are allowed to have jumps of arbitrary activity.
- The test is robust to market microstructure noise in observed prices. model
Source sentence
In particular, it is robust to jumps with arbitrary activity and to the presence of market microstructure noise. By contrast, we show in this section that the test statistic bT n from (3.4) is naturally robust to many forms of microstructure noise considered in the literature, without the need to employ classical noise-reduction techniques such as two-scale estimation (Zhang et al. (2005)), realized kernels
- Under the alternative hypothesis, the stochastic volatility process is a rough process with low regularity, modeled as ct = f(vt) with vt driven by a fractional kernel g(t). model
Source sentence
6 2.2 The Alternative Hypothesis Under the alternative hypothesis, the stochastic volatility process c ceases to be a semimartingale and is given by a rough process with a low degree of regularity. More precisely, we assume under the alternative hypothesis that ct = f(vt), where vt = v0 + Z t 0 g(t −s)(σv sdWs + σv sdW s) + evt, (2.13) v0 is F0-measurable and f is a C2-function.
- The sampling interval is fixed while the observation frequency increases (high‑frequency asymptotics). statistical
Source sentence
The test utilizes the fact that volatility is rough if and only if volatility increments are negatively autocorrelated at high frequencies. By showing a feasible CLT for this statistic under the null hypothesis of semimartingale volatility paths, we construct a test with fixed asymptotic size and an asymptotic power equal to one.
Show 2 more
- Spot volatility estimates are constructed from the empirical characteristic function of price increments in local blocks. model
Source sentence
The spot volatility estimates are constructed from the empirical characteris-tic function of price increments in local blocks of high-frequency data. By choosing values of the characteristic exponent away from zero, we mitigate the impact of finite variation jumps on the spot volatility estimators.
- The observed log‑price may be contaminated with additive i.i.d. normal noise. data
Source sentence
Finally, we assume that the observed log-price is contaminated with noise, i.e., instead of observing x, we observe yi∆n = xi∆n + σnoise p Vi∆nεi, i = 1, . . . , n, (4.6) where {εi}i=1,...,n is an i.i.d. sequence of standard normal random variables defined on a product extension of the sample probability space and independent from F. We set σ2 noise = 0.5 × 1 252 × 15 (1.0548 −1)/(4620 −77 × 1.0548) = 2.40 × 10−8.
Data
- SPY high‑frequency price data market-prices public
Source sentence
On each trading day, we sample the SPY price at five-second frequency from 9:35 AM ET until 4:00 PM ET. We drop days with more than 20% zero five-minute returns over the entire trading period of 6.5 hours.
- Monte Carlo simulated data synthetic unknown
Source sentence
4 Monte Carlo Study In this section, we evaluate the performance of the test on simulated data. 14 4.1 Setup We use the following model for x: dxt = p VtdWt + Z R x(µ(dt, dx) −dtνt(dx)), (4.1) where Wt is a standard Brownian motion, µ is an integer-valued random measure counting the jumps in x with compensator measure dtνt(dx) and νt(dx) is given by νt(dx) = c e−λ|x| |x|α+1 Vtdx.
Costs
Not applicable
Finds 3
- Applying the test to SPY high‑frequency data provides evidence for rough volatility. statistical
Source sentence
In an application of the test to SPY high-frequency data, we find evidence for rough volatility. They provide evidence for existence of rough volatility.
- In Monte Carlo simulations, the test rejects the null hypothesis for roughness parameters H = 0.3 and H = 0.1, with higher rejection rates for H = 0.1. statistical
Source sentence
For the volatility specification, we consider three values of the roughness parameter: H = 0.5 (this is the standard Heston model), H = 0.3 and H = 0.1, with the first value corresponding to the null hypothesis and the last two to the alternative hypothesis. Recall that lower values of H imply rougher volatility paths and the rough volatility literature argues for value of H around 0.1 (Gatheral et al. (2018)).
- The autocorrelations of differenced volatility increments are negative up to lag 7, consistent with rough volatility. statistical
Source sentence
The reported autocorrelations up to lag 7 are all negative, with the highest in mag-nitude being the first one. This is consistent with existence of rough volatility.
Checked by others
Supported by
- Roughness Analysis of Realized Volatility and VIX through Randomized Kolmogorov-Smirnov Distribution
Compare the findings
This paper: Applying the test to SPY high‑frequency data provides evidence for rough volatility.
That paper: Both volatility measures exhibit Hurst exponents significantly below 1/2, confirming pronounced roughness in volatility dynamics.
1 paper in the archive : A unified theory of order flow, market impact, and …
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.