Craving money? Evidence from the laboratory and the field
Data files
Dec 15, 2023 version files 371.61 MB
-
Data.zip
-
ISE_data.zip
-
README.md
-
SImulated_Data.zip
Abstract
We propose a new model of choice under repeated exposure to gambles. In it, the agent may come to choose a negative expected value, negative skew gamble, due to a behavioral bias that has a neurobiological foundation. We run laboratory experiments as a first step in testing the model and supplement the experimental findings with suggestive evidence from observational data. In the process of doing so, we identify a new asset pricing anomaly. The findings bring novel insights into the motivations underlying investor decisions and the impact of temptation and self-control in contexts of repeated risk-taking.
README: Craving Money? Evidence from the Laboratory and the Field
https://doi.org/10.5061/dryad.1c59zw42z
All the data and code used to generate the findings for the paper can be found here. You can find the code to generate the lab/theoretical results and the data for the option prices and transaction level data.
1st.) For the experimental and Lab results:
All experimental data needed to evaluate the conclusions reported in the article and the Supplementary Materials are available at:
The code used to generate the experimental findings reported in the article and its Supplementary Materials can be found here in the repository.
The code covers all the models listed below.
- Base model (Risk neutral, Deterministic):
- Base model (Risk neutral, softmax):
- Base model (prospect theory, Deterministic):
- Base model (prospect theory, softmax):
- Base model with Gambler's Fallacy:
- Base model augmented with CbD (Risk neutral, Deterministic):
- Base model augmented with CbD (Risk neutral, softmax):
- Base model augmented with CbD (prospect theory, Deterministic):
- Base model augmented with CbD (prospect thoery, softmax):
- Base model augmented with CbD (modified probability function, Risk neutral, softmax):
- Base model with CbD(Risk neutral,Deterministic,multiplicative):
- Base model with CbD model (Prospect theory,Deterministic, same alphas):
- Time-Varying DA model (risk neutral, softmax):
- Time-Varying DA model (prospect theory, softmax):
- Time-Varying DA model (prospect theory, softmax, same alphas):
- Data Generator for Time-varying DA model (Prospect theory, softmax):
- Data Generator for Base model (prospect theory, softmax):
Folder containing all the simulated datasets:
Each File starts wither a B (Base model) or TV (Time Varying). There are 120 Base and Time-Varying datasets, respectively.
Pre-Requisites
For any given dataset, the user needs to run all the cells in a model (in respective order) to get the results. Parameter optimization and obtaining results are separate parts in each model: One can look at the output of the hyperparameter optimization step in a code to get the values of all the free parameters. The results for a dataset (accuracy score of a particular model in %) can be obtained from the output of the bottom-most code cell.
In some datasets, the 'betting column' (whether or not the subject chooses to bet, denoted by 1 or 0 respectively) is col.No.4 and in some datasets, it is col.No.5, so while training the model, variable 'Y' (which is actually the betting column) needs to be changed (to df[4] or df[5]) depending on datasets.
The user needs to change the drive link in each code to whatever is the location of dataset.
In CbD models, n-value of 4 is used, meaning that DA is zero for any trial t such that 1<= t <= 4
There are no separate codes for the varying n-value test (as everything is exactly the same except for n-value), so the user may change the value of n in the CbD model (Prospect theory, Deterministic) code for that test. Simulated participants' datasets are similar to original datasets, only they contain 2 columns instead of 6/7 columns, so while testing the models on these datasets for model recovery, df[4] should be replaced by df[2], everything else remaining the same. For model recovery procedure 1, Both time-varying DA model (prospect theory, softmax) and Base model (prospect theory, softmax) are trained and tested on simulated participants for Time-varying model. For model recovery procedure 2, Both time-varying DA model (prospect theory, softmax) and Base model (prospect theory, softmax) are trained and tested on simulated participants for Base model.
2nd. )Option Empirics:
For the empirical analysis of the options prices, two data sources were used. First, all the option price data came from Optionmetrics. Second, the transaction-level data is from the International Securities Exchange (ISE). This data covers the period from May 2005 through 2019.
The columns represent: Option ID, Date, Call or Put, Option Expiration date, Open Interest, firm CUSIP, best bid, best offer, Volume, Strike Price, Implied Volatility, Firm Open Buy, Firm Closing Buy, Firm Open Sell, Firm Closing Sell, Customer Open Buy, Customer Closing Buy, Customer Open Sell, and Customer Closing Sell.