Chapkovski, Khapko, and Zoican (2023)
"Trading Gamification and Investor Behavior"

Experimental replication package

General instructions

General Instructions

To run the experiments and generate all Tables and Figures in the paper, follow these setup instructions based on your operating system:

For Windows Users:

1. Set Up Stata Path: Ensure your local Stata path is correctly set up in conf/config.yaml. For Windows, it should look like this (uncomment this line and comment out the Mac path):

   # Stata path for windows
   stata_path: "C:/Program Files/Stata17/StataMP-64.exe"

2. Run the Experiment: Use the executable file run_experiment.bat to execute all code in order. This script generates all Tables and Figures.

For macOS and Ubuntu Users:

1. Set Up Stata Path: Update your local Stata path in conf/config.yaml. For macOS, uncomment the relevant line and comment out the Windows path. Adjust the path based on your Stata version (e.g., StataSE, StataMP):

   # Stata path for mac
   # You may need to change from StataSE to StataMP depending on your version
   stata_path: "/Applications/Stata/StataSE.app/Contents/MacOS/StataSE"

2. Make the Script Executable (if needed): Before running the script for the first time, make it executable by running:

   chmod +x ./run_experiment.sh
   

3. Run the Experiment: Execute the script run_experiment.sh using:

   ./run_experiment.sh
   

   This script will execute all code in order and generate all Tables and Figures.

Note: Ensure that the paths specified in config.yaml correctly match the location of your Stata installation.

• All required raw data is in the raw_data_dump folder. You can safely empty the following folders before running the code:
  • data_processed, data_prolific, figures, and tables.

Data cleaning and pre-processing

1. code/select_sample.py: Takes the raw Heroku data (which includes all development tests) in the raw_data_dump folder and selects the relevant experimental sessions.
   1. Run with parameter 0 to extract data for main sample and with parameter 1 for the University of Toronto sample used in Online Appendix G.
   2. Running the code with parameter 0 populates 3 data files in the data_prolific folder: pretrade_data.csv, trader_actions.csv, and trader_metadata.csv.
   3. Running the code with parameter 1 populates 3 other data files in the data_prolific folder: pretrade_data_uoft.csv, trader_actions_uoft.csv, and trader_metadata_uoft.csv.
   4. Saves Figure 4 in the main paper as figures/main_figure_4.png.
   5. Saves Table 2 in the main paper as tables/main_table_2.tex.
2. code/figure_price_path: Renders Figure 3 with a sample price path and saves it as figures/main_figure_3.png.
3. code/optimal_trades_theory.py: Uses the price paths in price_paths folder to compute the optimal strategy of a Bayesian trader in any given round. Saves two files:
   1. data_prolific/optimal_trade_flags.csv, for price paths featuring momentum in Sessions I-III
   2. data_prolific/markov_trade_flags.csv, for martingale price paths in Session IV.
4. code/self_reflection.py: Uses post-experimental data to encode preference dummies.
   1. Run with parameter 0 for main sample, saves panel as /data_processed/self_reflection.csv
   2. Run with parameter 1 for University of Toronto sample, saves panel as /data_processed/self_reflection_uoft.csv.
5. code/build_main_panels.py takes the raw data from from data_prolific and generates three panels for regression analysis:
6. data_processed/panel_trader_round.csv: a panel with participant-round observation (aggregate measures)
7. data_processed/panel_tick_data.csv: a panel with tick-by-tick actions of participants (including idle ticks)
8. data_processed/panel_trades.csv: a panel with all participant trades.
9. code/generate_figures.py: Generate figures in the main paper. Run with parameter 0 for main experiment and parameter 1 for University of Toronto data. List of figures, all saved in figures subfolder:
   1. Figures 5 and G1: Preferences for gamification and financial literacy
   2. Figures 6 and G2: Trading activity and preferences for gamification
   3. Figures 7 and G3: and Distribution of $\pi_t$ by trading activity
   4. Figure 8: Tick-by-tick trading decisions and notifications
   5. Figures 9 and G4: Trading decision on notification ticks
10. code/jackknife_financialquiz.py: Builds a jack-knife samples measure of financial quiz score (dropping questions one by one):
    1. Saves resulting panel of quiz scores as data_processed/jackknife_finquiz.csv.
    2. Saves Figure H.1 in the figures subfolder.

Regression code

8. regressions/execute_stata.py:
   1. Run with argument regression_trader_round.do to generate trader-round tables in tables folder: Tables 4, 5, and 7 in the main paper; Tables F1, F2, F3, F4, F5, F6 in the Online Appendix.
   2. Run with argument regression_trader_panel.do to generate trader panel tables: Table 3 in the main paper and Table H1 in the Online Appendix.
   3. Run with argument regression_tick_data.do to generate tick-level panel tables: Table 6 in the main paper and Table F7 in the Online Appendix.
   4. Run with argument regression_robustness_H2_H6.do to generate robustness Tables H2 through H6 in the Online Appendix.
   5. Run with argument regression_quiz_jackknife.do to generate Table H.7 with jack-knife measure of the financial quiz score.