BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Mathcal MA TUM//V1
BEGIN:VEVENT
UID:20180307102202-ID1181@mathcal.ma.tum.de
DTSTART:20180507T121500Z
DTEND:20180507T130000Z
DTSTAMP:20180404T082941Z
SUMMARY:
Prof. Michael Ludkowski: Marrying Stochastic Control and Machine Learnin
g: from Bermudan Options to Natural Gas Storage and Microgrids
LOCATION:
B349 (Theresienstr. 39\, 80333 München)
DESCRIPTION:
Simulation-based strategies bring the machine learning toolbox to numeri
cal resolution of stochastic control models. I will begin by reviewing t
he history of this idea\, starting with the seminal work by Longstaff-Sc
hwartz and through the popularized Regression Monte Carlo framework. I w
ill then describe the Dynamic Emulation Algorithm (DEA) that we develop
ed\, which unifies the different existing approaches in a single modular
template and emphasizes the two central aspects of regression architect
ure and experimental design. Among novel DEA implementations\, I will di
scuss Gaussian process regression\, as well as numerous simulation desig
ns (space-filling\, sequential\, adaptive\, batched). The overall DEA t
emplate is illustrated with multiple examples drawing from Bermudan opti
on pricing\, natural gas storage valuation\, and optimal control of back
-up generator in a power microgrid. This is partly joint work with Adity
a Maheshwari (UCSB).
END:VEVENT
BEGIN:VEVENT
UID:20180131103457-ID1154@mathcal.ma.tum.de
DTSTART:20180507T130000Z
DTEND:20180507T134500Z
DTSTAMP:20180309T100547Z
SUMMARY:
Dr. Tobias Kley\, Berlin: Quantile-Based Spectral Analysis of Time Seri
es
LOCATION:
B 349 (Theresienstr. 39\, 80333 München)
DESCRIPTION:
Classical methods for the spectral analysis of time series account for c
ovariance-related serial dependencies. This talk will begin with a brief
introduction to these traditional procedures. Then\, an alternative met
hod is presented\, where\, instead of covariances\, differences of copul
as of pairs of observations and the independence copula are used to quan
tify serial dependencies. The Fourier transformation of these copulas is
considered and used to define quantile-based spectral quantities. They
allow to separate marginal and serial aspects of a time series and intri
nsically provide more information about the conditional distribution tha
n the classical location-scale model. Thus\, quantile-based spectral ana
lysis is more informative than the traditional spectral analysis based o
n covariances. For an observed time series the new spectral quantities a
re then estimated. The asymptotic properties\, including the order of th
e bias and process convergence\, of the estimator (a function of two qua
ntile levels) are established. The results are applicable without restri
ctive distributional assumptions such as the existence of finite moments
and only a weak form of mixing\, such as alpha-mixing\, is required.
END:VEVENT
BEGIN:VEVENT
UID:20180131103225-ID1153@mathcal.ma.tum.de
DTSTART:20180507T140000Z
DTEND:20180507T144500Z
DTSTAMP:20180309T100627Z
SUMMARY:
Dr. Gregor Kastner\, Wien: Bayesian Time-Varying Covariance Estimation i
n Many Dimensions using Sparse Factor Stochastic Volatility Models
LOCATION:
B349 (Theresienstr. 39\, 80333 München)
DESCRIPTION:
We address the curse of dimensionality in dynamic covariance estimation
by modeling the underlying co-volatility dynamics of a time series vecto
r through latent time-varying stochastic factors. The use of a global-lo
cal shrinkage prior for the elements of the factor loadings matrix pulls
loadings on superfluous factors towards zero. To demonstrate the merits
of the proposed framework\, the model is applied to simulated data as w
ell as to daily log-returns of 300 S&P 500 members. Our approach yields
precise correlation estimates\, strong implied minimum variance portfoli
o performance and superior forecasting accuracy in terms of log predicti
ve scores when compared to typical benchmarks. Furthermore\, we discuss
the applicability of the method to capture conditional heteroskedasticit
y in large vector autoregressions.
END:VEVENT
END:VCALENDAR