←
Simon Scheidegger708 posts
Simon Scheidegger avatar

Simon Scheidegger

@comp_simon

Associate Professor @HECLausanneEcon

◷ Joined June 2018▧ 73 Photos and videos
Simon Scheidegger avatar
Simon Scheidegger@comp_simon·May 13

Thrilled to share a project I've been refining: a complete, open-source repository on "Deep Learning for Solving and Estimating Dynamic Models in Economics and Finance." I've cleaned up the materials from my PhD classes and summer schools into one coherent resource. 🧵 1/6

Post attachment
Andreas von Astiburg avatar
⟳ Simon Scheidegger retweeted
Andreas von Astiburg@Astiburg·Oct 18

Reaktorentwicker will in die Schweiz: Immerhin gibt es noch Leute, die in der Schweiz einen Beitrag für den künftigen Wohlstand in anderen Ländern leisten wollen. Das Projekt selbst ist durchaus eine Herausforderung. tagesanzeiger.ch/copenhagen-…

Post attachment
NZZ Schweiz avatar
⟳ Simon Scheidegger retweeted
NZZ Schweiz@NZZSchweiz·Oct 2

Der Expertenkrach um die AKW-Kosten geht in die nächste Runde: Nach einer Gruppe von ETH-Wissenschaftern übt nun auch ein ehemaliger Spitzenbeamter des Bundes Fundamentalkritik an den Berechnungen der Akademien der Wissenschaften – diese widersprechen... nzz.ch/schweiz/43-milliarden…

Andreas von Astiburg avatar
⟳ Simon Scheidegger retweeted
Andreas von Astiburg@Astiburg·Oct 2

Egal, was ein neues AKW kostet: Die Energiewende ist praktisch unbezahlbar. Der Atomausstieg bricht jeder modernen Volkswirtschaft das Genick. Man muss sich entscheiden: Bin ich ein Narr & Staatsfeind und schade mir und allen anderen, oder eben nicht? nzz.ch/schweiz/43-milliarden…

Simon Scheidegger avatar
Simon Scheidegger@comp_simon·Sep 28

Delighted to see this published with @PhilippJRenner! Moving beyond 2 or 3 types has been a persistent roadblock for ages. By bypassing APS set-valued iterations entirely, we open up rich type distributions for applied work. 💻 Code: github.com/GaussianProcesses…

"Machine Learning for Dynamic Incentive Problems": @comp_simon & @PhilippJRenner introduce a framework combining penalties, Gaussian processes and Bayesian active learning to bypass set-valued dynamic programming. Handles at least 10 persistent types." restud.com/machine-learning-…

Post attachment
Marlon Azinovic-Yang avatar
⟳ Simon Scheidegger retweeted
Marlon Azinovic-Yang@MarlonAzinovic·Sep 15

I remember the first presentation of this. @comp_simon drew Gaussian Processes by hand.

Thrilled to share that "Machine Learning for Dynamic Incentive Problems" (with @PhilippJRenner ) is in Review of Economic Studies (@RestudJournal)! This was truly a career-decisive project for me: my Job Market Paper back when ML and deep learning were not yet a thing 🧵👇

Post attachment
Simon Scheidegger avatar
Simon Scheidegger@comp_simon·Sep 15

Free-access article link: academic.oup.com/restud/adva… Python toolbox & replication code: github.com/GaussianProcesses…

We introduce a penalty relaxation that bypasses computing feasible sets a priori, solving it via value function iteration with Gaussian Process regression and Bayesian Active Learning. The generic framework scales up to at least 10 persistent types.

Simon Scheidegger avatar
Simon Scheidegger@comp_simon·Sep 15

We introduce a penalty relaxation that bypasses computing feasible sets a priori, solving it via value function iteration with Gaussian Process regression and Bayesian Active Learning. The generic framework scales up to at least 10 persistent types.

Dynamic adverse selection with persistent private info typically requires set-valued dynamic programming over high-dimensional, unknown, and irregular state spaces (curse of dimensionality).

Simon Scheidegger avatar
Simon Scheidegger@comp_simon·Sep 15

Dynamic adverse selection with persistent private info typically requires set-valued dynamic programming over high-dimensional, unknown, and irregular state spaces (curse of dimensionality).

Thrilled to share that "Machine Learning for Dynamic Incentive Problems" (with @PhilippJRenner ) is in Review of Economic Studies (@RestudJournal)! This was truly a career-decisive project for me: my Job Market Paper back when ML and deep learning were not yet a thing 🧵👇

Post attachment
Simon Scheidegger avatar
Simon Scheidegger@comp_simon·Sep 15

Thrilled to share that "Machine Learning for Dynamic Incentive Problems" (with @PhilippJRenner ) is in Review of Economic Studies (@RestudJournal)! This was truly a career-decisive project for me: my Job Market Paper back when ML and deep learning were not yet a thing 🧵👇

Post attachment