Time-Changed Lévy Process

levy-processtime-changestochastic-volatilityrealized-volatilityhigh-frequencyjump-diffusionou-processlong-memoryquasi-likelihoodquadratic-variationbipower-variation

Definition

A time-changed Lévy process is a log-price model of the form Yt=mt+ZτtY_t = mt + Z_{\tau_t} where ZZ is a Lévy process (independent and stationary increments, Var(Z1)<∞\text{Var}(Z_1) < \infty, Z0=0Z_0 = 0) and τt=∫0tσu du\tau_t = \int_0^t \sigma_u\,du is a non-decreasing stochastic time-change, assumed independent of ZZ (Z⊥τZ \perp \tau). The time-change is driven by a non-negative spot variance process σt\sigma_t. When ZZ is Brownian motion this reduces to the standard diffusion SV model; non-Brownian ZZ introduces infinite-activity or finite-activity jumps.

Key Ideas

How It Works

Second-order properties of RV

Let [Y]d,i∗∗=[Y]d,i[Y]_{d,i}^{**} = [Y]_{d,i} be realized variance for period ii. Ignoring O(M−2)O(M^{-2}) terms: E([Y]d,i)≈k2Δξ,Cov([Y]d,i,[Y]d,i+s)≈k22ω2r~Δs∗∗E([Y]_{d,i}) \approx k_2\Delta\xi, \quad \text{Cov}([Y]_{d,i}, [Y]_{d,i+s}) \approx k_2^2\omega^2\tilde{r}^{**}_{\Delta s} Var([Y]d,i)≈2ω2k22rΔ∗∗+k4Δξ⋅{(4k1k3+2k22)(2ω2Mrd∗∗+M−1Δ2ξ2)}\text{Var}([Y]_{d,i}) \approx 2\omega^2 k_2^2 r_\Delta^{**} + k_4\Delta\xi\cdot\{(4k_1k_3 + 2k_2^2)(2\omega^2 Mr_d^{**} + M^{-1}\Delta^2\xi^2)\} where rs=Cor(σt,σt+s)r_s = \text{Cor}(\sigma_t, \sigma_{t+s}), ω2=Var(σt)\omega^2 = \text{Var}(\sigma_t), and rs∗∗r_s^{**}, r~∗∗\tilde{r}^{**} are integrated autocorrelation functions.

Quasi-likelihood estimation

Build a Gaussian QL from the above second-order properties; constrain k2=1k_2 = 1 (otherwise k2k_2 and ξ\xi are not separately identified from Var([Y]d,i)\text{Var}([Y]_{d,i})). Sandwich standard errors I−1JI−1I^{-1}JI^{-1}.

Computation: (i) Durbin algorithm (O(n2)O(n^2)): exploits Toeplitz structure of Cov([Y]d∗∗)\text{Cov}([Y]_{d}^{**}) when τ\tau is stationary; works for any parameterised autocorrelation function. (ii) Kalman filter (O(n)O(n)): applies when τi−Δξ=x′$i\tau_i - \Delta\xi = x'\$_i and $\$ satisfies a linear state-space recursion; ~20× faster than Durbin for n=3000n=3000 and 2-dimensional state.

Model hierarchy

Model rsr_s State dim Algorithm
Single OU e−λse^{-\lambda s} 2 Kalman
JJ-factor superposition ∑jwje−λjs\sum_j w_j e^{-\lambda_j s} 2J2J Kalman
Log-normal OU (LNOU) [eσlog⁡2e−λ∣s∣−1]/[eσlog⁡2−1][e^{\sigma^2_{\log}e^{-\lambda|s|}}-1]/[e^{\sigma^2_{\log}}-1] approx. 2×102\times 10 Kalman (truncated Poisson approx.)
Gamma long-memory (1+s/a)−2H(1 + s/a)^{-2H}, H<1/2H < 1/2 ∞\infty Durbin

For single OU: Cov(τi,τi+s)\text{Cov}(\tau_i, \tau_{i+s}) gives ARMA(1,1) for τi\tau_i with AR root ϕ=e−λΔ\phi = e^{-\lambda\Delta} and analytic MA root from the ratio r1=ϕ(Cor(τi,τi+1)−ϕ)/[(1+ϕ2)−2ϕ Cor(τi,τi+1)]r_1 = \phi(\text{Cor}(\tau_i, \tau_{i+1}) - \phi)/[(1+\phi^2) - 2\phi\,\text{Cor}(\tau_i, \tau_{i+1})].

Empirical results (USD/DEM, Olsen data, 1986–1996, M=288M=288)

Why It Matters

Time-changed Lévy processes provide a unified framework for non-Gaussian volatility models that encompasses both continuous (diffusion) and jump components. The RV inconsistency result quantifies the cost of jumps for realized variance estimation: the signal that RV recovers is the quadratic variation (continuous + jump), not the integrated variance (continuous only). The QL estimation framework connects high-frequency second-order properties directly to continuous-time model parameters without requiring distributional assumptions beyond moments, making estimation computationally tractable even for complex memory structures. The ACF inequality provides a formal justification for the claim that "true" volatility predictability exceeds what realized variance series reveal.

Open Questions

Related