Blogs›Shopify›Quantile Regression for Merchant Funding

Quantile Regression for Merchant Funding

posts

2017–2021

This feature thread tracks the development and application of machine learning techniques, specifically quantile regression, to assess merchant eligibility for cash advances. The goal is to accurately predict future sales and offer funding that balances risk for both merchants and Shopify. The technique allows for nuanced predictions that account for varying levels of sales volatility, ensuring that advances are structured for a high probability of repayment. This post details the use of propensity score matching (PSM) to evaluate the impact of Shopify Capital on merchant growth by comparing US merchants who received funding with a matched cohort of Canadian merchants who did not have access at the time. The methodology involved defining treatment and control groups, selecting relevant features for matching, employing a caliper matching algorithm, and assessing matching quality through standardized mean differences, visual diagnostics, and variance ratios.

2021

Using Propensity Score Matching to Uncover Shopify Capital’s Effect on Business Growth - Shopify

11/11/2021

This post details the application of propensity score matching (PSM) to evaluate the causal impact of Shopify Capital on merchant growth. It outlines the research question, the challenges of A/B testing in this context, and the methodology used to create comparable treatment (US Capital recipients) and control (Canadian non-recipients) groups. Key aspects include defining the groups, selecting features for matching, using a caliper matching algorithm, and assessing the quality of the match using standardized mean differences, visual diagnostics, and variance ratios.

2017

How Shopify Capital Uses Quantile Regression To Help Merchants Succeed - Shopify

7/11/2017

This post introduces the application of quantile regression within Shopify Capital's risk-assessment algorithms. It explains the limitations of traditional regression for predicting sales distributions with varying variances and details how quantile regression addresses this by predicting specific quantiles of future sales. The post provides theoretical background, mathematical formulations, and practical examples using Python to illustrate the concept and its application in offering tailored merchant cash advances.