• Archivi categoria: News

    Seminario Steven Scott (Google)

    Il giorno venerdì 2 settembre alle ore 17, presso l’Aula Martini al piano interrato dell’edificio U6,  Steven Scott, Senior Economic Analyst at Google, terrà un seminario su

    Predicting the Present with Bayesian Structural Time Series

    This article describes a system for short term forecasting based on an ensemble prediction
    that averages over different combinations of predictors. The system combines a structural
    time series model for the target series with regression component capturing the contributions of contemporaneous search query data. A spike-and-slab prior on the regression coefficients induces sparsity, dramatically reducing the size of the regression problem. Our system averages over potential contributions from a very large set of models and gives easily digested reports of which coefficients are likely to be important. We illustrate with applications to initial claims for unemployment benefits and to retail sales. Although our exposition focuses on using search engine data to forecast economic time series, the underlying statistical methods can be applied to more general short term forecasting with large numbers of contemporaneous predictors (joint work with Hal Varian).

    Tutti gli interessati sono invitati a partecipare; sarà anche possibile seguire in streaming. L’evento è organizzato in collaborazione con Innovation Pub.

    Steve sarà in Italia per una scuola di una settimana sul Lago di Como alla quale parteciperanno anche i nostri dottorandi. Vi segnalo anche la conferenza per il grande pubblico su Google data and Public Sentiment che si terrà a Como il giorno 30 agosto alle 18.00. Per approfondimenti consultate la pagina web di Scott.

    Suboptimality in portfolio CVaR optimization

    Giovedì 21 gennaio 2016 ore 11.00

    (edificio U7, 4° piano, stanza 4026)
    Via Bicocca degli Arcimboldi, 8 – 20126 Milano

    Suboptimality in portfolio CVaR optimization

    Edgars Jakobsons
    ETH Zurich

    Abstract
    We consider the portfolio optimization problem with conditional value-at-risk as the objective. Summarizing commonly used methods of solution, we note that the linear programming approximation is the most generally applicable and easy one to use (the LP uses a Monte Carlo sample from the true asset returns distribution). The suboptimality of the obtained approximate portfolios is then analyzed using a numerical example, with up to 101 assets and Student-t distributed returns, ranging from light to heavy tails. The results can be used as an estimate of the portfolio suboptimality for more general asset returns distributions, based on the number of assets, tail-heaviness, and the fineness of the discretization. Computation times using different techniques available in the literature are also analyzed.

    Tutti gli interessati sono invitati a partecipare

    Per ulteriori informazioni: fabio.bellini@unimib.it

    Allegato: LOCANDINA

    Seminari Prof. Emilio Porcu – “Short course” su processi stocastici spazio-temporali

    Data: 
    27/01/2016

    S’informa che:

    Il prof. Emilio Porcu (Department of Mathematics, University Federico Santa Maria, Valparaiso, Chile) terrà presso l’aula seminari del Dipartimento di Statistica e Metodi Quantitativi dell’Università degli studi di Milano-Bicocca (stanza 4026) i seguenti quattro seminari:

    –  Mercoledì 27/01/2016 ore 14:00: ” Space-Time covariance functions for planet Earth
    – Mercoledì 27/01/2016 ore 15:15: “Dynamically compactly supported space-time covariance functions
    – Giovedì 28/01/2016 ore 9:00: “Composite Likelihood Approaches for Space-Time Gaussian fields
    – Giovedì 28/01/2016 ore 10:15: “Equivalence of Gaussian measures for some Gaussian fields“.

    Tutti gli interessati, in particolare i dottorandi di ogni ordine e ciclo, sono invitati a partecipare.

    Seguono gli abstract.

    1) Wednesday, 27 january 2016 14:00
    Space-Time covariance functions for planet Earth

    Abstract:
    In this paper, we propose stationary covariance functions for processes that evolve temporally over a sphere, as well as cross-covariance functions for multivariate random fields defined over a sphere. For such processes, the great circle distance is the natural metric that should be used in order to describe spatial dependence. Given the mathematical difficulties for the construction of covariance functions for processes defined over spheres cross time, approximations of the state of nature have been proposed in the literature by using the Euclidean (based on map projections) and the chordal distances. We present several methods of construction based on the great circle distance and provide closed-form expressions for both spatio-temporal and multivariate cases. A simulation study assesses the discrepancy between the great circle distance, chordal distance and Euclidean distance based on a map projection both in terms of estimation and prediction in a space-time and a bivariate spatial setting, where the space is in this case the Earth. We revisit the analysis of Total Ozone Mapping Spectrometer (TOMS) data and investigate differences in terms of estimation and prediction between the aforementioned distance-based approaches. Both simulation and real data highlight sensible differences in terms of estimation of the spatial scale parameter. As far as prediction is concerned, the differences can be appreciated only when the interpoint distances are large, as demonstrated by an illustrative example.

    2) Wednesday, 27 january 2016 15:15
    Dynamically compactly supported space-time covariance functions

    Abstract:
    Compactly supported covariance functions have been very popular in geostatistics in the last years. For instance, they are the cornerstone of the covariance tapering technique for both estimation and prediction. Here, we explore compact support in space-time covariance functions. We propose a general class of nonseparable and stationary covariance functions with dynamical temporal support; that is, the compact support in space is a decreasing function of the temporal lag. A special case of our general class is dynamical Wendland functions, which preserve the same properties of the original Wendland functions in space. Specifically, we find that this new class allows for preserving and parameterizing the differentiability at the origin when multiplied with Gneiting functions coupled with a Mat ́ern spatial margin. As an application of the proposed class, we explore covariance tapering for estimation in the space-time context using dynamical tapers. In particular, we focus on large datasets with special features (a few spatial location sites and many observations over time and vice versa). The effectiveness of the method is illustrated with a simulation study and by analysing Irish wind speed data.

    3) Thursday, 28 january 2016 9:00
    Composite Likelihood Approaches for Space-Time Gaussian fields.”

    Abstract
    In the recent years there has been a growing interest in proposing covariance models for multivariate Gaussian random field. Some of these covariance models are very flexible and can capture both the marginal and the cross spatial dependence of the components of the associated multivariate Gaussian random field. However, effective estimation methods for these models are somehow unexplored. Maximum likelihood is certainly a useful tool but it is impractical in all the circumstances where the number of observations is very large. In this work we consider two possible approaches based on composite likelihood for multivariate covariance model estimation. We illustrate through simulation experiments and numerical examples that our methods offer a good balance between statistical efficiency and computational complexity. Asymptotic properties of the proposed estimators are described under increasing domain. Finally we apply the method in analyzing a bivariate dataset on chlorophyll concentration and sea surface temperature in the Chilean coast.

    4) Thursday, 28 january 2016 10:15
    Equivalence of Gaussian measures for some Gaussian fields

    Abstract
    Equivalence of Gaussian measures represent the key for understanding optimal prediction under infill asymptotics. We show that some covariance functions are compatible with the Matérn class according to Yadrenko’s theory. We analyze the implications from the point of view of optimal prediction.

    Allegato: LOCANDINA

    How is elicitability relevant for backtesting?

    Lunedì 08 febbraio 2015 ore 16.00

    Aula seminari, edificio U7, 4° piano, stanza 4026
    Via Bicocca degli Arcimboldi, 8 – 20126 Milano
    How is elicitability relevant for backtesting?”

    Prof.ssa Johanna Ziegel
    University of Bern

    Abstract
    Independently, Weber [2006] and Gneiting [2011] have shown that Expected Shortfall (ES) is not elicitable in contrast to Value at Risk (VaR). Roughly, elicitability of a risk measure means that it can be obtained as the minimizer of an expected loss function. This negative result continues to hold for all spectral risk measures (except for the mean) and the only coherent risk measures that are elicitable are certain expectiles. However, we were able to show recently that ES is jointly elicitable with VaR, and, more generally, a large class of spectral risk measures is elicitable of higher order [Fissler and Ziegel, 2016].
    There is little debate that elicitability is a useful property for model selection, estimation, generalized regression, forecast comparison, and forecast ranking. But the non-elicitability of ES has lead to a lively debate about the relevance of elicitability for backtesting [Acerbi and Szekely, 2014, Davis, 2016, Emmer et al., 2015].
    Contributing to this discussion, we would like to clarify that elicitability is not important for the traditional approach to backtesting. However, we argue that elicitability is crucial to achieve the objectives of backtesting [Fissler et al., 2015].
    We illustrate the proposed approach for VaR and ES jointly and for VaR alone.

    Tutti gli interessati sono invitati a partecipare

    Per ulteriori informazioni: fabio.bellini@unimib.it

    Allegato: LOCANDINA