Ανάρτηση Ερευνητικών Δοκιμίων no 13/25, no 14/25 & no 15/25

    Ερευνητικό Δοκίμιο no 13/25 με τίτλο "Updating Density Estimates using Conditional Information Projection: Stock Index Returns and Stochastic Dominance Relations"

    των Στέλιου Αρβανίτη, Richard McGee και Thierry Post

    Περίληψη

    We propose, analyze, and apply a Conditional Information Projection Density Estimator (CIPDE) that estimates the conditional density function by projecting a prior time-series estimator onto the set of distributions that satisfy given conditional moment conditions with a functional nuisance parameter. A statistical theory is developed based on asymptotic theory, information geometry and variational analysis. Theoretically, CIPDE is shown to reduce the limiting divergence to the latent population density, if the prior is inconsistent and the moment conditions are well specified. A Monte Carlo simulation experiment and an empirical analysis apply the proposed method to stock index returns using pricing restrictions for index options. The simulations show large potential improvements in the density estimates in a controlled but realistic environment. In the empirical analysis, CIPDE is shown to enhance monthly S&P500 index return density forecasts and improve
    the out-of-sample investment performance of one-month S&P500 index option strategies by better timing of protective put purchases and covered call writing.

    Ο Στέλιος Αρβανίτης είναι Καθηγητής στο τμήμα Οικονομικής Επιστήμης του Οικονομικού Πανεπιστημίου Αθηνών, o Richard McGee είναι Επίκουρος Καθηγητής στο University Dublin College, Ireland και ο Thierry Post είναι Καθηγητής στο Graduate School of Business of Nazarbayev University, Kazakhstan.

    Ερευνητικό Δοκίμιο no 14/25 με τίτλο "Measuring Textual Cohesion via Graph Curvature, Weighted Dominance, and Embedding-Based Coherence"

    του Στέλιου Αρβανίτη

    Περίληψη

    Evaluating textual cohesion is an important component of information retrieval, summarization, and automated content quality assessment. We present a framework that combines semantic embeddings with graph-based indicators of discourse structure. Sentence similarity graphs are built from SBERT embeddings, and Ollivier–Ricci curvature is applied to capture structural connectivity. We introduce a Weighted Utopia Index (UI), derived from a dominance relation among curvature embeddings, to quantify proximity to cohesive discourse. To enhance robustness, a Nash-style multiobjective aggregation balances the UI with embedding-based coherence scores. Experiments  on the publicly available Cohesentia dataset show that the proposed metrics achieve higher alignment with human coherence judgments than embedding-only baselines, with consistent gains in agreement and nDCG. The approach is domain-independent, interpretable, and adaptable to a range of evaluation scenarios, including summarization quality control and discourse analysis
     

    Ο Στέλιος Αρβανίτης είναι Καθηγητής στο τμήμα Οικονομικής Επιστήμης του Οικονομικού Πανεπιστημίου Αθηνών.

    Ερευνητικό Δοκίμιο no 15/25 με τίτλο "Universal Preference Spaces for Expected Utility via a Banach-type Functorial Fixed Point"

    των Στέλιου Αρβανίτη, Παντελή Αργυρόπουλου και Σπύρου Βασιλάκη 

    Περίληψη

    This paper develops universal preference spaces in the Expected Utility paradigm using functorial fixed point theory in enriched categories. We work within the CMS-enriched category of compact metric spaces with embedding retraction morphisms and define the endofunctor vNMλ, which maps each  object to a space of Lipschitz-continuous von Neumann–Morgenstern utilities. Under conditions, vNMλ admits a canonical fixed point, representing a universal space of recursively structured preferences. This construction is extended  to multi-utility preferences by composing vNMλwith the Hausdorff functor. Both universal spaces are also characterized as colimits of their respective ω-chains, starting from the terminal object. These constructions provide a  categorical framework for representing layered decision models under ambiguity. Possible extensions include dropping embedding requirements and organizing  preferences via coend-based constructions.
     

    Ο Στέλιος Αρβανίτης είναι Καθηγητής στο τμήμα Οικονομικής Επιστήμης του Οικονομικού Πανεπιστημίου Αθηνών, ο Παντελής Αργυρόπουλος είναι teaching Assistant στο MSc in Economics του Οικονομικού Πανεπιστημίου Αθηνών και ο Σπύρος Βασιλάκης είναι Καθηγητής στο τμήμα Διεθνών και Ευρωπαϊκών Οικονομικών Σπουδών του Οικονομικού Πανεπιστημίου Αθηνών.