Ali Furkan Kalay

Language models like ChatGPT have become an integral part of my daily life, both personally and professionally, ever since I started using them. One particular area where I’ve seen a significant improvement is in my prompt engineering skills. Here is one of the ways I have been using ChatGPT to save time. [Read More]

I have presented the Linked Administrative Data and Estimating Marginal Willingness to pay for Amenities (joint work with James Hansesn and Alicia N. Rambaldi) at Western Economic Association International @ Melbourne.

I presented Peer Effects in the Demand for Private Health Insurance (joint work with Alicia Rambaldi and Chris Rose) paper in Brown Bag Series at the University of Queensland.

I have been working on a project to create synthetic data for a long while. I have realized that the synthpop package was producing identical values, whereas it was supposed to be producing values from a predictive posterior distribution. I have been using the CART algorithm for its flexibility, but the model must be overfitting, even though I do not have too many variables. So, how can one prevent creating identical values? The answer was given in the article: “synthpop: An R package for generating synthetic versions of sensitive microdata for statistical disclosure control”. [Read More]

This is an attempt to estimate Double Machine Learning with XGboost algorithm in R. The purpose is to create a benchmark estimation with DML. The user can choose various machine learning algorithms, where optimizing hyperparameters can be time-consuming. XGboost is a very useful in this regard. This script can be used to produce substantially accurate preliminary results. Repository is here. [Read More]