Amir Sani, PhD

Thank you for visiting my site.

My research interests are on the crossroads of machine learning, resource allocation and macroeconomic forecasting. In particular, I am interested in time-varying forecast combinations, online surrogate modeling, conformal regression and high-dimensional sampling. My current research,

  • attempts to address the Lucas Critique by calibrating macroeconomic agent-based models to real (eventually microfoundation) data. As part of this work, I have designed a high-dimensional sampling algorithm that enhances online surrogate modeling, which minimize the evaluations of costly simulators.
  • establishes an information theoretic dependent data Bootstrap for the most general class of dependent data processes possible (only assuming that the processes are Stationary-Ergodic). This allows for Monte-Carlo simulations, statistical analysis and Bootstrapping of highly dependent series, such as those produced by decision-making processes (e.g. online optimization/learning) and information cascades.

  • provides an online optimization algorithm that allows one to test a new decision-making policy, while avoiding the risk of underperforming an existing policy by more than a known constant. Further, the same algorithm allows one to combine a pool of heterogeneous forecasts into a point forecast that can be leveraged with the safety of never performing worse than the existing policy, or the difficult to outperform mean combination (also referred to as the so-called "forecast combination puzzle").

Background

I am a postdoctoral research fellow working with Antoine Mandel at the Centre d'Économie de la Sorbonne, Université Paris 1, Panthéon-Sorbonne, Paris School of Economics. My research is part of the European Union Horizons 2020 Future and Emerging Technologies Distributed Global Financial Systems for Society (DOLFINS) project, which addresses the global challenge of making the financial system better serve society.

I am also an an External Researcher at the CFM Imperial Institute of Quantitative Finance, Visiting Professor at HEC Paris, Visting Professor at the University of Paris II Panthéon-Assas, Affiliate Researcher at School of Advanced Studies Pisa and an Associate Researcher at RiskLab Finland.

I completed my PhD, Machine Learning for Decision Making Under Uncertainty, under the supervision of Rémi Munos and Alessandro Lazaric at INRIA-Lille Nord Europe as part of the SequeL team.

Working Papers

Publications

Workshop Papers/Posters

  • Inferring Complex Networks of Influence: Understanding Green Investment Tipping Points, with Antoine Mandel, NIPS Workshop on Inference and Learning of Hypothetical and Counterfactual Interventions in Complex Systems, 2016
  • Macroeconomic Agent Based Model Calibration using Iterated Surrogates, with Francesco Lamperti, Antoine Mandel and Andrea Roventini, NIPS Workshop on Inference and Learning of Hypothetical and Counterfactual Interventions in Complex Systems, 2016
  • Information Theoretic Bootstrapping for Dependent Time Series, with Alessandro Lazaric and Daniil Ryabko, NIPS Workshop on Modern Nonparametric Methods in Machine Learning, 2013
  • The Universal Bootstrap for Dependent Data, with Alessandro Lazaric and Daniil Ryabko, Statistical modeling, financial data analysis and applications, 2013
  • Risk Averse Multi-Arm Bandits, with Alessandro Lazaric and Rémi Munos, NIPS Workshop on Markets, Mechanism and Multi-Agent Models, 2012

Visits

- January 23-27, 2017, Monash Business School Department of Econometrics and Business Statistics, Melbourne, Australia
- January 10-11, University of Zurich, Department of Banking and Finance
- September 25-30, 2016, Institute of Economics - Scuola Superiore Sant'Anna, Pisa, Italy
- September 16, 2016, Alan Turing Institute, London, England - April 17-24, 2016, Institute of Economics - Scuola Superiore Sant'Anna, Pisa, Italy
- February 5, 2016, Center for Data Science Paris-Saclay

Talks

Teaching

Reviewer

Conferences: NIPS 2016, ICML 2016, AI-Stats 2016, 2017, COLT 2013
Journals: Quantitative Finance 2016, Journal of Evolutionary Economics 2016, Computational Economics 2017

Events

  • November 2015 through February 2016, Data Driven Economics and Complexity Seminar Series
  • February 10th, 2016, Collaborative Hackathon for Macroeconomic Agent-Based Model Surrogates
  • February 9th, 2016, Macroeconomic Surrogates for Agent-Based Models Workshop

Contact Information

I can be reached via email at reachme@amirsani.com.