2022
Automatic pricing competition. An experiment with simulated Amazon seller.
Joint work with Jeanine Miklos-Thal and Catherine Tucker
Made in Python with Django
It is a study on how the presence of automation tools on such platforms as Amazon influences the long-term pricing equilibria. At the core of this research project, we are looking at how individuals make their decisions in an online pricing simulation, which imitates Amazon sellers. Each seller is operating in two online marketplaces simultaneously. In every marketplace, the seller has to choose his own price. In one of the marketplaces, sellers are allowed to use automated pricing (provided by the platform) to help them set their prices, but not in another. Each seller is facing two competitors in each marketplace.
Economics keywords: market equilibrium, pricing automation, experimental economics
Computer Science keywords: Cloud Services, AWS EC2, Django, linux KVM
2021
Product Launches with New Attributes: A Hybrid Conjoint-Consumer Panel Technique for Estimating Demand. Neural network estimation.
joint work with Mitchell Lovett and Bhoomija Ranjan
Made in Python, R, Keras
The conjoint study is a great tool for making predictions of individual preferences for new-to-market product attributes. Unfortunately, this tool measures only stated preferences, which prompt of different types of biases. The demand estimation of revealed preferences provides much more accurate results for existing products but can’t be used for non-existing product attributes. This paper provides a hybrid method of demand estimation, which combines the information from both conjoint surveys and consumer panel purchase data. The core idea of the method lies in utilizing neural networks which are built in a structural way (NN+S approach), and further regularized and tuned for best out-of-sample prediction.
2020
Improving the substitution patterns of aggregate demand models
Made in Julia, Python, R
Modern aggregate demand models with random coefficients rely on observed product characteristics
to account for substitution patterns between products.
This project investigates an application of matrix factorization approach for
estimating unobserved product characteristics from online ratings in the beer industry.
Estimated unobserved characteristics have
several advantages over observed characteristics, including relevance, interpretability
and low dimensionality. This technique can also be applied
for estimating unobserved product characteristics to improve the estimates of the substitution between products
in such industries as video games, movies, books.
-
The main internal loop is implemented with Julia, which makes the algorithm lighting fast (zero memory allocation)
-
Overcame the problem of numerical stability which is commonly reported by researchers
-
Custom made gradient-based optimization algorithm
Data Science keywords: data cleaning, PCA, SVD, hyper-parameters tuning, regularization, cross-validation, cluster computing, shell scripting
Economics Keywords: cross-price elasticity, GMM, bootstrapping, BLP
2019
Matrix factorization using neural networks
Made in Python
Matrix factorization algorithms (such as Singular Value Decomposition) are very popular in recommender systems.
The goal of this project is to estimate SVD using a neural network, which improves the convergency to the global minimum.
(Almost any mathematical model can be expressed as a custom neural network and can be trained using standard Neural Network tools, such as Keras, Tensorflow, and PyTorch)
Data Science keywords: web scraping, neural networks, SVD, hyper-parameters tuning, regularization, cross-validation
Economics Keywords: product map, individual tastes heterogeneity, individual scale usage
2019
Human-like web-scraping
Made in R
In some situations, you can not scrape online data as a robot.
To get the information your program needs to behave as a human.
Take some time to think, react to on-screen events, move the mouse cursor as a human. It turns out that RSelenium package
and a lot of work can make it possible.
-
Fault-tolerant web scraping performed by multiple robots in parallel
-
A unique identity of each robot
-
An extra process which orchestrates all the robots to make a stratified sample
2019
Getting insights from text-mining of online reviews
Made in R
Text-mining can be used for investigating product closeness.
An underlying assumption is that products with similar user experience will receive similar user reviews.
Data Science keywords: web scraping, text-mining, stemming, stop words removal, synonyms pooling, multidimensional scaling
2019
Analysis of keywords search money spending
Made in R
There are plenty of online tools for analyzing keywords. None of them is better than the data from your own advertising campaigns.
The idea was to combine the data from individual keywords advertising campaigns and spacial locations of the competitors to increase the conversion rate per dollar spend. This project is still in development.
2018
Quasi-experiment in online reviews
Made in R
Econometric study on the presence of social bias in online user reviews in the context of the video game industry.
The study exploits the fact, that some websites report average ratings for the games by aggregating for each platform separately (PC, Xbox, etc.),
while other websites report average ratings for the games by pooling all the platforms together.
Economics keywords: diff-in-diff, quasi-natural experiment
2017
Continuous web-scraping
Made in C#, MySQL
In some cases, you need to visit a page multiple times to get a single observation.
The goal of the project is to develop a library for gathering such observations. The library was used in several further projects.
- Fault-tolerant custom made ORM
- 100'000 requests per day from a single computer
2013
Software for semi-automatic parsing huge text of data
Made in Java
The idea was to quantify the information from a particular book. The book was scanned and recognized with a lot of mistakes.
The number of mistakes didn't allow to develop a fully automatic parser.
The solution was to develop a semi-automatic parser, which was doing all the heavy lifting automatically while asking an operator (human) to resolve the most fault critical situations.
2011
LAN based games with stabilo-cheers synchronization
Made in C++ using RakNet, Irrlicht, .NET
The goal of the project was to develop software, which allows conducting different experiments in the Laboratory of Experimental Economics.
The software was able to track the movement of the center-mass of individuals and mouse movements while individuals were playing economic games by LAN.