What was Migrations.ml Inc?

I started Migrations.ml in February 2019, after quitting my job at RBC, where I managed a team of BAs and QAs serving the Finance pillar.

My thesis was that credit analysis was a great use case for machine learning: the data used in credit analysis is audited, or at least attested to by company executives under Sarbannes-Oxley in the US. Furthermore, there are well-defined credit events that could be used as the dependent variable or label in a supervised learning model.

From this analysis, it looked like the first area to focus on was the US corporate bond market. Financial fundamentals data was available via EDGAR, market data was available via FRED, and there are almanacs of default events published by Moody’s, S&P and Fitch.

As they say, the devil is in the details. I thought that I could buy a Bloomberg data license, and use it as the data source for a home-brewed supervised learning model. Learning experience #1: Bloomberg doesn’t play that game. After the salesperson referred me to their legal review team, I was told that they viewed my two-week old company as a competitor, and they wouldn’t sell me a license. After experimenting with xbrl files, I eventually landed on a service that re-formatted EDGAR filings into csv files. This wasn’t a perfect fit, as the aggregations used by the service didn’t match the granularity of the model we were progressing towards. Solving this puzzle cut into resources across the board, and another discovery was made: time is a big source of stress. Even if you have a high-flying team, and loads of financing (to be clear, I didn’t have loads of financing!), if you can’t get your product in a presentable state, let alone bringing it to market on a timely basis, those potential early adopters will lose interest.