Spectral calibration of exponential Levy models ??
19
∣F ( O) -O )( u ) F ( O) -O )( v )∣2
Λ ~~ ~~ ɪ-"
IIF(Oi - O)H4oo + E[F(O - Oi)(u)F(O - Oi)(v)12],
which together with estimates (30), (35) yields that (40) is bounded in order
by
∆8
-U -U
+ E hlll X δkδiεkεiFbk (u)Fbi (v)lll i´
u 4 WU ( u ) v 4 WU ( v )
κ(u)2κ(v)2 du dv
U U N 4U 4U
u W (u)v W (v)
∆ ∆8 + ∑ δkδj ∣Fbk (u) |2 ∣Fbl (v) |2 ----σ ( V) ( σ( ) du
U -U k i 1 κ(u)2κ(v)2
k,i=1
U
∆4
-U
u4WU (u) 2 U N 2 2 u4 WU (u) 2
(\2 d u + / δLFδbk F bk ( u ) 12 (22 d u
κ(u)2 -U k=1 κ(u)2
∆8U4+∆4U4∣δ∣i22 E(Tσm2axU2)2.
Putting all estimates together and using U .∆-1 as well as∆∣δ∣i22 . ∣δ∣i2∞
we obtain (32) and consequently the rate for σ2 .
6.4. Upper bound for γ and λ
Since the claimed risk bound for γ is larger than for σ2, we only need to
σ2
estimate the risk of γ + σ2- instead of that for γ. Equally, we can restrict
to λ - σ2— γ instead of λ. Then the proof follows exactly the lines of the
proof for σ2, the only difference being the different norming in estimate
(30) giving rise to a factor U for γ and a factor U2 for λ. It remains to
note that we obtain the bounds in the compound Poisson case by setting
σ = σmax = 0 and considering the continuous extension of the bounds for
that case. For γ we obtain as bias
lʃ F μ ( u ) WU ( u )d u∣ . U - ( s+2).
(41)
The linear error term is estimated by
U2
E -U L(u)WγU(u) du .E(Tσ2U2) U2∆4+U∆∣δ∣i2∞
(42)
and the remainder satisfies
E
2
R(u)WγU (u) du∣∣ .E(Tσm2axU2)2∆8U6+∆4U6∣δ∣i22 .
(43)
For λ we obtain the same asymptotic error bounds as for γ, but multiplied
by U when regarding the root mean square error. With the rate-optimal
choice (23) of U this gives the asserted risk bounds for ^ and λ.