Talking to Machines

Description

This is the replication package for the paper 'Evidence for Countercyclical Risk Aversion: An Experiment with Financial Professionals' by Cohn et al., (2015), as part of the 'Talking to Machines' collaborative project.

Project Structure

The basic project structure is:

├── code
├── data
├── docs
├── experiment_results
├── prompts
└── README.md

where:

• code contains the R script for the preprocessing of data generated by the package to match the data by Cohn et al.
• data contains the pre-processed LLM data
• docs contains notes on how the replication was informed (notes.txt) and includes all relevant information from the original paper (codebook, experimental instructions and manuscript).
• experiment_results contains the raw outputs generated after running the package
• prompts contains the prompt sheet used

Installation

Firstly, download the repository or clone it by running the command:

git clone https://github.com/sohaamir/ttm

To install the relevant packages, please refer to the Talking to Machines GitHub page, which describes how to set-up the project environment.

Specifically, you will need to set-up a conda environment, install a version of python > 3.10 and export an OpenAI API key.

Replication guide

To run the talkingtomachines package on the study simply run in the terminal from the project root:

talkingtomachines prompts/cohn_prompt_exp1

and select full when prompted.

This will generate the raw output within the experiment_results folder.

Then simply open talkingtomachines.Rproj and run exp1.Rmd. This will generate the preprocessed data exp1.csv which matches the structure of the original paper.

For more information on how the replication was structured, please consult docs/notes.txt

Acknowledgements

This replication was created by:

• Aamir Sohail (University of Birmingham)
• Qianze Zhang (University of Gröningen)
• Hayat Zouiten (Mohammed VI Polytechnic University)