About Me
I am a third-year PhD student in Economics at the University of Toronto. My research is situated at the intersection of macroeconomics, household finance, and computational economics. I focus on how the decisions of heterogeneous households and entrepreneurs interact with monetary and fiscal policy. My work aims to build more realistic models of the economy to better understand the effects of policy on investment, hiring, and economic stability.
Research
Working Papers
-
Leverage, Default and the Entrepreneurial Channel of Policy Transmission (Work in Progress)
Abstract
This paper develops a heterogeneous-agent New Keynesian model to study the investment channel of monetary policy transmission through risk-averse, credit-constrained entrepreneurs. The model provides a new perspective on fiscal policy, demonstrating that debt-financed expansions can be significantly dampened by a crowding-out effect on entrepreneurial investment.
-
The Retirement Gap: Entrepreneurs, Tax Policy, and the Cost of Liquidity (Work in Progress)
With: Sebastian Dyrda, Baxter Robinson
Conferences: Winter SED (Argentina, 2024); BSE Summer Forum (2025); Summer SED (Copenhagen, 2025)
Abstract
This paper studies how entrepreneur's income risk affects their wealth accumulation and portfolio allocation. We document novel facts about the portfolio allocation of entrepreneurs, showing that entrepreneurs hold substantial financial assets outside their business. We then build a life-cycle model of entrepreneurship that features risky intangible capital. The risky nature of intangible capital helps our model match the portfolio allocation of entrepreneurs of assets inside vs. outside their business. We demonstrate the aggregate importance of this mechanism by studying the aggregate cost of financial frictions in a model with and without intangible capital. We also study two policy applications: an expansion of small business loans and a fiscal stimulus transfer showing that the effects of both policies depend critically on the presence or absence of risky intangible capital.
-
Demand, Spillovers, and Declining Entry (Work in Progress)
With: Mahmood Haddara
Abstract
When a firm invests in increasing its profits, does this benefit other firms? The endogenous growth literature has typically assumed it does, with each firm "standing on the shoulders" of its predecessors. We consider an endogenous growth model with two types of investment: innovation, which generates positive spillovers by advancing the technological frontier, and demand accumulation, which increases profits for a particular firm while leaving the technological frontier unchanged. The key novelty of our framework is to endogenously link this distinction to the level of entry in the economy. Specifically, while new firms can build on the innovations that preceded them, their initial demand stock is fixed. As a result, the distance between a potential entrant and incumbent firms depends on the composition of investment in the economy. We quantify the importance of this mechanism using a dynamic model of oligopolistic competition and endogenous growth. Our framework suggests that while targeted investment subsidies are welfare-improving, indiscriminate subsidies can inadvertently deter entry, lower competition, and slow aggregate growth.
Pre-PhD
-
Common Errors in Machine Learning Projects: A Second Look (2024)
Conferences: Koli Calling '23
Abstract
While machine learning (ML) has proved impactful in many disciplines, design decisions involved in building ML models are difficult for novices to make, and mistakes can cause harm. Prior work by Skripchuk et al. [35] identified common errors made by ML students via qualitative analysis of open-ended ML assessments, but their sample was limited to a single institution, course, and assessment setting. Our work is an extended, conceptual replication of this work to understand the common errors made by machine learning students. We use a mixed-method approach to analyze errors in 30 final project reports in an undergraduate machine learning course. The final reports describe the model-building process for a classification task, where students build models on a complex data set with numerical, categorical, ordinal and text features. Our choice to analyze project reports (rather than code) allows us to uncover design errors via how students justify their methodology. Our project task is to achieve the best test accuracy on an unseen test set; thus, as a way to validate these common errors, we identify the association between these errors and the model’s test accuracy performance. Common errors we find include those consistent with Skripchuk et al. [35], for example issues with data processing, hyperparameter tuning, and model selection. In addition, our focus on design error exposes other common errors, for example where students use certain kinds of features (e.g., bag of words representations) only with particular models (e.g., Naive Bayes). We call these latter types of errors model misconceptions, and such errors are associated with lower test accuracy. Some of these errors are also present in work by practitioners. Others reflect a difficulty by students to make correct connections between ML concepts and achieve the relational level of the SOLO taxonomy. We identify areas of opportunity to improve machine learning pedagogy, particular... [truncated]
-
Modeling Social Learning Using Dyna-Q and Ant Colony Optimization (2023)
Conferences: The Carroll Round XVII
Abstract
This paper introduces a novel way of modeling social learning in macroeconomics using techniques from model-based reinforcement learning and ant colony optimization. The work extends previous works in bounded rationality and social learning by providing tools to complement previously-distinct models in adaptive learning. We test these new techniques using simulations of job search and consumption. Results demonstrate that models fit using the proposed techniques can learn core economic behaviors given no information about the environment, but do not fully fit reward functions in line with rational expectations theory.
Education
- PhD in Economics, University of Toronto (2023 - Present)
- MA in Economics, University of Toronto (2022 - 2023)
- HBSc in Computer Science and Economics, University of Toronto (2018 - 2022)
Experience
-
Course Instructor, University of Toronto
Courses
- Economics
- ECO482: Machine Learning Applications in Macroeconomic Finance
- Computer Science
- CSC258: Computer Organization
- CSC207: Software Design
-
Teaching Assistant, University of Toronto
Courses
- Economics
- ECO2107 (Graduate): Monetary Theory
- ECO2101 (Graduate): Macroeconomic Theory II (PhD)
- ECO2100 (Graduate): Macroeconomic Theory I (PhD)
- ECO2460 (Graduate): Economic Applications of Machine Learning
- ECO482: Machine Learning Applications in Macroeconomic Finance
- ECO375: Econometrics I
- ECO227: Quantitative Methods in Economics
- ECO208: Macroeconomic Theory
- ECO200: Microeconomic Theory
- ECO102: Principals to Macroeconomics
- ECO101: Principals to Microeconomics
- ECO100: Introduction to Economics
- Computer Science
- CSC367: Parallel Programming
- CSC311: Introduction to Machine Learning
- CSC236: Introduction to the Theory of Computation
- CSC148: Introduction to Computer Science
- CSC108: Introduction to Computer Programming
-
Research Assistant, University of Toronto
Details
- Murat Celik
- Re-write and optimize Matlab code in Julia to solve a heterogeneous agent macroeconomic model.
- Achieve 30x speed improvement by leveraging advanced code optimization and numerical procedures.
- Adapt the model to be fully compliant of automatic differentiation to improve estimation efficiency.
- Sebastian Dyrda
- Design, implement and test global nonlinear optimizer in modern object-oriented Coarray (MPI) Fortran.
- Read literature and study techniques to optimize code for high performance computing environments.
- Run and profile massively-distributed models in up to 6000 cores in the Compute Canada supercomputer.
- Xu Tian
- Create high-performance parallel algorithms in C++ to scrape, process and analyze 4TB+ of financial data.
- Design and implement dynamic structural models for research in corporate finance using Matlab and Fortran.
- Accelerate computationally-intensive routines with GPUs and use MPI to parallelise estimation.
- Marlene Koffi
- Implement language processing models using techniques from machine learning and neural networks.
- Parallelize and optimize mathematical code for research in gender economics.
- Profile code used for large scale parallelism in supercomputer clusters.
- Ismael Mourifie
- Translate structural econometric models from Matlab to Fortran and optimize memory and cache efficiency.
- Design and implement distributed implementations of main research algorithms.
- Improve total program execution speed from 1.2 hours (reported by PI) to 6.8 seconds.