The Common Task Framework is Live - Submit Your Pricing Model

A new chapter in asset pricing research begins: the Common Task Framework (CTF) is now open for submissions.

Ready to compete? Submit your model, see how it ranks, and contribute to the cutting edge of empirical finance.

What is the Common Task Framework?

The Common Task Framework is a collaborative and competitive research paradigm that has driven breakthroughs across science and technology. The framework rests on three pillars: shared training data, a common task with predefined metrics, and a leaderboard that controls for overfitting. This approach has produced remarkable results:

• DARPA Speech Recognition (1970s-90s): DARPA's standardized evaluations allowed researchers to systematically compare approaches on common datasets. Over two decades, word error rates dropped dramatically, laying the groundwork for modern voice assistants (Huang et al., 2014).
• ImageNet (2012): The ImageNet Large Scale Visual Recognition Challenge provided 1.3 million labeled images across 1,000 categories. When AlexNet won the 2012 competition by a dramatic margin using deep convolutional neural networks, it ignited the deep learning revolution (Krizhevsky, Sutskever, & Hinton, 2012).
• CASP / Protein Folding (1994-2020): The Critical Assessment of protein Structure Prediction (CASP) competition ran for 25 years with incremental progress. Then DeepMind's AlphaFold2 effectively solved the problem at CASP14 in 2020. Demis Hassabis and John Jumper received the 2024 Nobel Prize in Chemistry for this breakthrough.

As Donoho (2017) put it, CTF is the "secret sauce" of machine learning. Now we're bringing this proven methodology to finance.

The Challenge: Finding the Pricing Kernel

The goal is simple yet profound: construct the stochastic discount factor (SDF) that maximizes out-of-sample Sharpe ratio.

You have access to 402 characteristics from the JKP dataset covering global equities (with more to come). Build your model, submit your methodology, and see how it ranks against the field.

How to Participate

1. Develop your model using the JKP characteristics data
2. Submit your files:
   • A Python or R script that generates portfolio weights
   • A CSV file with your model's weight predictions
   • A PDF describing your methodology
   • If needed, a file that outlines your package dependencies
3. Track your performance on the public leaderboard

Your code runs in a controlled sandbox environment with standardized dependencies. See our detailed submission rules for technical specifications.

Early Results

Our initial evaluation of 18 candidate models reveals that machine learning methods consistently outperform traditional factor models, and ensemble approaches outperform individual methods. But the leaderboard is just getting started and there's plenty of room at the top.

Get Started

• Submit Your Model: jkpfactors.com/ctf/submit
• Starter Code & Examples: github.com/theisij/common-task-framework-SDF
• CTF Dataset Access: jkpfactors.com/ctf/dataset-access
• Submission Rules: jkpfactors.com/ctf/rules
• Full Paper: The Power of the Common Task Framework

Why Participate?

The CTF methodology has a proven track record of accelerating scientific progress by orders of magnitude. By participating, you're contributing to a rigorous, reproducible evaluation of asset pricing models, and potentially shaping the future of empirical finance.

Whether you're a PhD student with a novel idea, a practitioner with production-tested models, or a researcher exploring new machine learning techniques, we want to see what you can build.

The leaderboard awaits. Show us what your model can do.

---

Questions? Join the discussion or contact us.