User behavior prediction

Organized by profrowe - Current server time: March 24, 2018, 12:11 a.m. UTC


Training and Evaluation
Oct. 25, 2016, midnight UTC


Competition Ends

Predict which traders respond to firm-up requests

Trumid is an all-to-all electronic credit trading platform. It is designed to unlock liquidity and make bond and CDS trading easier.

This competition supplements the Trumid interview process for Data Science roles in Trumid Labs (the data science and quantitative analysis group). Trumid has a meritocratic hiring process and invites you to participate in the challenge if you are interested in a data science role at Trumid. Competition participants with a high score automatically qualify for a phone screen despite prior experience (although you should be in the United States, as we don't cover relocation).

Trumid is a trading platform for bonds. Like an exchange, buyers and sellers use Trumid to buy and sell bonds. However, Trumid is a dark pool, which means that market participants cannot see each other's orders. One reason for this is that bonds don't have much liquidity, so placing an order gives away too much information, and that information can be used against you. So how do traders know if there are potential buyers/sellers of a specific bond? Traders can enter an order to buy or sell a bond. If an order exists on both sides and are close enough together, then a "lit" trading session begins, with a proposed mid price between the bid (buy price) and ask (sell price).

Trumid gives traders another option. Traders can enter non-binding IOIs (indications of interest) that are only visible to the system. IOIs give traders a safe way to show some interest without having to commit to a trade. Now, if another trader enters a firm order on the opposing side, the platform will send the original trader a "firm-up request". This message simply asks the trader to "firm up" their IOI into an order. The goal of this competition is to predict which firm-up requests lead to firm orders.


See Participate for some important data considerations.


The challenge essentially boils down to a binary classification for each firm-up request. For a given request, predict whether the order_placed field is TRUE, meaning the trader firmed up the IOI into an order, or FALSE, meaning they ignored the request. Submissions will be evaluated based on the F1 score as well as its precision and recall constituents.

For applicants currently in the Trumid interview process, the purpose of this competition is to provide a setting for discussing your model development process. During the on-site interview, you should be prepared to present your approach to the problem, what assumptions you made, how you selected and tuned your model, etc. Be prepared to defend your methodology and know where the approach might break-down. This discussion is more important than the actual score. That said, if your F1 is worse than random, you probably need to put more time into it.

If you are currently not in the Trumid interview process, then the score is more important. Specifically, participants with a scores will have the opportunity to interview with Trumid Labs, irrespective of your experience. We believe in meritocracy, and this is a great way for capable and passionate but less experienced individuals to get noticed.

Challenge Rules

  • The work on a submission must be the work of a sole individual. No teams are allowed.
  • Invitations to submit an entry are made at the sole discretion of Trumid
  • Evaluation of results and phone interview offers are made at the sole discretion of Trumid

Training and Evaluation

Start: Oct. 25, 2016, midnight

Competition Ends


You must be logged in to participate in competitions.

Sign In