I like finding ways to use machine learning to solve problems.
As a researcher, I thrive on using machine learning—particularly deep learning—to explore and answer complex questions across finance, economics, and business applications. My work is driven by the challenge of applying advanced neural network models to real-world problems that often differ from standard large-model settings like vision or language processing. Whether it's modeling the dynamics of household behavior to understand broader economic trends or optimizing portfolio choices in response to transaction costs, I focus on harnessing machine learning techniques to uncover valuable insights.
Some of the key questions I've tackled include:
- How does individual household behavior move economies?
- How do portfolio choices vary with transaction costs?
- Do tick sizes affect asset prices?
- Can unpaired images be used to make cross-domain production recommendations?
- How can nonlinear estimators be used for causal inference?
These questions often involve unique challenges—like approximating non-linear functions with high precision—requiring a deep understanding of neural networks and their application to financial data. My background in econometrics, rooted in my master's studies, complements my engineering skills in implementing algorithms and building efficient data pipelines. This blend of theory and practical expertise allows me to design and execute sophisticated data-driven solutions, while staying hands-on with algorithm development and implementation. No matter the task—whether conceptualizing new algorithms or working through the finer points of implementation—I'm passionate about finding elegant, efficient solutions to complex problems.
Published Research and Working Papers
Deep Equilibrium Nets
Marlon Azinovic, Luca Gaegauf, and Simon Scheidegger
Published in International Economic Review, 63(4).
Abstract: We introduce deep equilibrium nets (DEQNs)—a deep learning-based method to compute approximate functional rational expectations equilibria of economic models featuring a significant amount of heterogeneity, uncertainty, and occasionally binding constraints. DEQNs are neural networks trained in an unsupervised fashion to satisfy all equilibrium conditions along simulated paths of the economy. Since DEQNs approximate the equilibrium functions directly, simulating the economy is computationally cheap, and training data can be generated at virtually zero cost. We demonstrate that DEQNs can accurately solve economically relevant models by applying them to two challenging life-cycle models and a Bewley-style model with aggregate risk.
Download PDF GitHub Repo tl;dr
A Comprehensive Machine Learning Framework for Dynamic Portfolio Choice With Transaction Costs
Luca Gaegauf, Simon Scheidegger, and Fabio Trojani
Working paper on SSRN.
Abstract: We introduce a comprehensive computational framework for solving dynamic portfolio choice problems with many risky assets, transaction costs, and borrowing and short-selling constraints. Our approach leverages the synergy between Gaussian process regression and Bayesian active learning to efficiently approximate value and policy functions with a novel, formal way of characterizing the irregularly-shaped no-trade region; we then embed this into a discrete-time dynamic programming algorithm. This combination allows us to study dynamic portfolio choice problems with more risky assets than was previously possible. Our results indicate that giving the agent access to more assets may alleviate some illiquidity resulting from the presence of transaction costs.
Download PDF tl;drOngoing Research
Empirical Causal Asset Pricing with Trading Costs
Luca Gaegauf and Vincent Wolff
Abstract: Do stock prices of publicly listed companies respond to changes in trading costs? We document a significant and asymmetric effect of tick-size changes on prices by leveraging a novel policy framework that allows for a causal randomized control trial difference-in-differences analysis. The doubling of the tick size leads to a decrease in prices by 0.9% to 1.3%, whereas halving the tick size results in an increase of 3.3% to 3.5%. The price effect is more pronounced in smaller firms and tick-unconstrained stocks. We report substantial excess returns on the day before the tick-size change attributable to quote and price clustering and, with caution, to strategic short-term price manipulation.
View PDF tl;dr
Which Shoes Fit This Dress? Using Product Images to Infer “Perfect Pairings” Across Categories Without Supervision
Luca Gaegauf and Markus Meierer
Abstract: Product recommendations are omnipresent and contribute significantly to the revenue of online retail businesses. Nevertheless, customers often complain about the quality of recommendations. A possible explanation is that the requirements of existing approaches are not always met. The increasing speed of product assortment changes challenges collaborative filter techniques, which are subject to the cold start problem. Alternative approaches such as preference- or content-based recommendations systems require detailed structured information (e.g., on product characteristics or previous customer purchase patterns), which is often not available. Addressing this challenge, the authors propose an unsupervised approach to infer recommendations across domains solely based on product images. With the increasing interest in deep learning techniques, convolutional neural networks have been used to infer single-domain recommendations based on the intrinsic information of product images (e.g., recommend alternative dresses based on the image of a dress). Albeit promising, approaches to infer cross-domain recommendations are largely unexplored. Using generative adversarial networks and convolutional neural networks to leverage the implicit information in product images, the authors propose an unsupervised approach to make recommendations across product categories (e.g., recommend shoes based on an image of a dress). The performance of the approach is assessed by benchmarking to various empirical baselines as well as surveying potential customers on the perceived quality of the recommendations. Further, the authors discuss two extensions: (1) Making recommendations for multiple other domains based on a single product image as input and (2) making cross-domain recommendations based on multiple product images as input. Concluding, theoretical and managerial implications are discussed.
View Workshop Slides tl;dr