Solorio M.
Following R.A. Fisher’s (1935) randomization methods of statistical inference to test the null hypothesis of no treatment effects, this paper analyzes the repercussions that a stock market crash has on belief formations with respect to future stock returns using Safford et al. (2018) behavioral finance experiment data. Safford et al. (2018), among others (Malmendier and Nagel, 2011; Weber et al., 2013; Cameron and Shah, 2015), find that experiencing a crash causes a significant difference in the overall belief distributions between individuals who experienced a market crash at the beginning of their investment career, and individuals who never experienced one. The randomization results presented in this paper, in contrast, indicate that experiencing a market crash has no effect on altering future expected returns to an extent that the post-belief distribution of a cohort who experienced a crash is not statistically different in its mean, standard deviation, skewness, and kurtosis from a belief distribution of a cohort who never experienced a crash. This is an indication that the correlation between market crashes and changes in risk aversion passes mainly through a preference channel rather than a belief one just as Bucciol and Zarri (2015); Voors et al. (2012); Callen et al. (2014); Kim and Lee (2014) have found.
Ridinger G., Sundali J., Guerrero F., Solorio M., Fan M., Sevindik I., Chen Q., Achoka D.
An “intergenerational” investment experiment is conducted in which investment advice is passed from one generation to the next. Participants make asset allocation decisions for 30 years to a safe and risky asset and provide annual forecasts (beliefs) on the return on the risky asset. Risky asset returns are drawn from the price returns on the S&P 500 from 1921-2010. Results show that negative investment advice passed from Generation 1 to Generation 2 leads to: 1) significantly lower allocations to the risky asset compared to Generation 1; and 2) a 19% difference in allocations between Generation 2 cohorts between those receiving positive vs. negative advice. A second experiment examines the effect on Generation 3 from receiving consistent and mixed advice from Generations 1 and 2. The results from the first experiment are replicated showing that positive (negative) advice received from prior cohorts leads to higher (lower) investment beliefs and portfolio allocations to the risky asset. Statistical analyses are conducted to determine if the large differences in allocations are driven by changes in beliefs about future returns or changes in risk aversion. The preliminary takeaway is that you might not want to listen to your parents if they tell you to stay away from investing in the stock market.
Solorio M., Cao-Noya J. A., Nichols M., Guerrero F., Sundali J.
This study examines the predictive capabilities of machine learning models in identifying distinct gambling behaviors among slot machine players. By utilizing an extensive dataset of 38, 299 unique slot machine players with over 24 million gambles, our research investigates the reasons behind prolonged gambling sessions versus shorter engagements. By integrating insights from Conlisk’s (1993) theoretical framework on the utility of gambling, modern psychology theory, and advanced machine learning techniques, we developed two predictive models that can classify gamblers based mainly on their interaction with the slot machine within the first 15 minutes of play and at the end of their initial gambling session. These models, one based on a Random Forest and the other on a Logistic Regression classifier, demonstrate robust predictive power, with the Random Forest model achieving an accuracy score of up to 0.912 and AUC of 0.903, and the Logistic Regression model reaching accuracy scores of 0.814 and AUC values of up to 0.842. Our analysis reveals that initial betting amount, gambling pace, players’ age, number of changes in the machines played, number of unique machines utilized, the account depletion rate, and the utility generated by each bet are significant predictors of a player's likelihood to engage in prolonged gambling sessions. Our findings underscore the potential of machine learning in developing early intervention strategies that could lead to responsible gambling practices and the creation of predictive tools for gambling disorder prevention.
Solorio M., Guerrero F., Sundali J.