Research
Dissertation
Decision Making of a Homo Silicus
∇ Abstract
Recent studies show that human behavior and its decision-making can be replicated by Large Language Models (LLMs) such as Generative Pre-trained Transformers (GPTs) and suggest that they can be used to explain real world economics. We created virtual agents, 'Homo Silicus', using LLMs that play a role in a dating simulation to provide an example of how simulation models with LLM based agents can help solving unobserved heterogeneity issues such as individual personality. Specifically, we adopted Myers-Briggs Type Indicator (MBTI) as a personality measure and assigned it to each of the silicon samples. We generated a total of 200 male and 200 female virtual agents and assigned them various traits including age, education, wealth, political affiliation, physical attractiveness and MBTI. Each group consists of 10 males and 10 females, and they choose their partner only based on the profiles in the first round, have conversations with all agents of the opponent sex in the second round, and make final decisions based on the profiles plus the conversations in the third round. Our findings show that MBTI as a personality measure weakens the impact of other traits such as partner's age, wealth and physical attractiveness. Additionally, controlling for unobserved heterogeneity, which is represented by MBTI in this paper, resulted in significant influence on the impacts of other observed variables. For such circumstances where economists are not able to obtain the information about unobserved heterogeneity variables from the data or where moral or ethical issues arise, we suggest a new approach of simulation with virtual agents generated by LLM that can help in enriching data analysis and obtaining better implications from the results.AI, AI, Who is the Fairest of Them All? Beauty Premium in Dating and Marriage among Celebrities
∇ Abstract
This research explores the association between the stability of romantic relationships and physical attractiveness among celebrities. Using deep learning and face recognition technology, we quantified the beauty of celebrities from their facial images, as well as the face similarities of couples. Our results show that more beautiful people are more likely to have more partners and the impact is much higher for women compared to less beautiful women. However, for both marriage and other relationships, the role of women’s beauty is short-expiry. The results strongly indicate that education and other unobservable values might dominate the beauty effect over time. Thus, beauty for female celebrities emerged as a dichotomous factor: lending an edge in some scenarios, but undermining in others. Looking deeper, we uncover the subtle role of sexual preferences and racial combinations, particularly when contextualized within occupations. Our study also suggested that history matters, with a richer market of relationships leading to markedly shorter durations with the latest pair. There is no evidence supporting the notion that similar-face-to-me or astrological compatibility. Our study can be used to help understand complex interactions in relationships and provide insight into how physical attraction can work for the stability of the relationship. Our study also acknowledges many limitations, and the use of deep learning technologies may have limitations when capturing the sparks of love between people.Genetic Assortative Mating and Its Impact on Household Economics
∇ Abstract
This paper uses polygenic scores (PGSs) from the Health and Retirement Study, derived from a genome-wide association study (GWAS) to identify the existence of genetic assortative mating and investigate its impact on household economies such as household wealth, family planning, and divorce as well as the possibility of its inter-generational inheritance. Key findings show that PGSs related to Subjective Well-being, Age at First Birth, and Number of Children Ever Born indicate positive assortative mating, with 35.9% of couples in the sample being similar to each other. Specifically, PGSs for Subjective Well-being and Age at First Birth are positively associated with household wealth while the PGS for Number of Children is negatively associated. For family plans, the PGS for Subjective Well-being leads to fewer children whereas the PGS for Age at First Birth and the Number of Children correspond with more children. Regarding divorce, PGSs for Subjective Well-being and Number of Children are negatively related, but the relationship between Age at First Birth and divorce remains uncertain. Additionally, due to the fact that genes are subject to being inherited by the next generations, the results show that these PGSs may play a role in explaining widening income inequality, declining birth rate, and divorce issues.Peer Reviewed Publications
Park, S. L., Le Marchand, L., Cheng, G., Balbo, S., Chen, M., Carmella, S. G., . . . & Hecht, S. S. (2022). Quantitation of DNA adducts resulting from acrolein exposure and lipid peroxidation in oral cells of cigarette smokers from three racial/ethnic groups with differing risks for lung cancer. Chemical Research in Toxicology, 35(10), 1914–1922.
∇Abstract
The Multiethnic Cohort Study has demonstrated that the risk for lung cancer in cigarette smokers among three ethnic groups is highest in Native Hawaiians, intermediate in Whites, and lowest in Japanese Americans. We hypothesized that differences in levels of DNA adducts in oral cells of cigarette smokers would be related to these differing risks of lung cancer. Therefore, we used liquid chromatography-nanoelectrospray ionization-high resolution tandem mass spectrometry to quantify the acrolein-DNA adduct (8R/S)-3-(2'-deoxyribos-1'-yl)-5,6,7,8-tetrahydro-8-hydroxypyrimido[1,2-a]purine-10(3H)-one (γ-OH-Acr-dGuo, 1) and the lipid peroxidation-related DNA adduct 1,N6-etheno-dAdo (εdAdo, 2) in DNA obtained by oral rinse from 101 Native Hawaiians, 101 Whites, and 79 Japanese Americans. Levels of urinary biomarkers of nicotine, acrolein, acrylonitrile, and a mixture of crotonaldehyde, methyl vinyl ketone, and methacrolein were also quantified. Whites had significantly higher levels of γ-OH-Acr-dGuo than Japanese Americans and Native Hawaiians after adjusting for age and sex. There was no significant difference in levels of this DNA adduct between Japanese Americans and Native Hawaiians, which is not consistent with the high lung cancer risk of Native Hawaiians. Levels of εdAdo were modestly higher in Whites and Native Hawaiians than in Japanese Americans. The lower level of DNA adducts in the oral cells of Japanese American cigarette smokers than Whites is consistent with their lower risk for lung cancer. The higher levels of εdAdo, but not γ-OH-Acr-dGuo, in Native Hawaiian versus Japanese American cigarette smokers suggest that lipid peroxidation and related processes may be involved in their high risk for lung cancer, but further studies are required.Lee, Y. H., & Chu, H. Y. (2015). An Analysis of Stock Price with the Remaining Volume. Asian Journal of International Studies (AJIS), 20, 96–116.
∇ Abstract
The most important factor that determines the stock price is the demand and the supply for a stock. Even if the value of a stock is very low and its price is overestimated, the price would still go up as long as there are more people willing to buy the stock. Even if the technical analysis gives a sign to sell a stock, the price may still go up as long as more people do not think the price will go down. This paper looks into how the demand and the supply for a stock would change, and from what we can define its changes. It is not the firm’s intrinsic value or the economic variable which directly drives the changes in the stock price in the future but it is the stock market trader (one that can affect the stock market) that takes action according to their expectations, whose behavior directly affects the stock price equilibrium. Traders’ expectations on the future stock price can be represented by the remaining volume of the asking price and the bid price. It does not matter whether the expectation is right or wrong. The set of traders’ expectation builds up the price in the future like ‘self-fulfilling prophecy’.Working Papers
Walsh, C. P., Shariff-Marco, S., Lee, Y., Wilkens, L. R., Le Marchand, L., Haiman, C. A., ... & Park, S. L. (2024). Joint Association of Education and Neighborhood Socioeconomic Status with Smoking Behavior: The Multiethnic Cohort Study.
∇ Abstract
BackgroundCigarette smoking is the leading cause of preventable mortality. Both neighborhood- and individual-level socioeconomic status (SES) are inversely associated with smoking. However, their joint effect on smoking behavior has not been evaluated.In Progress
Bond-Smith, S., and Lee, Y. Estimating a CPI-based Relative Price Parity Index For US Cities.
∇ Abstract
We use metropolitan area Consumer Price indexes (CPI) and the US CPI to calculate a CPI-based Regional Price Parity Index (CPI-based RPP) for the period from 1969 to the most recent CPI observation. The Bureau of Economic Analysis (BEA) uses the Personal Consumption Expenditures (PCE) Price Index to estimate a similar Regional Price Parity Index (BEA RPP)—which is the gold standard for regional price comparison indexes—but only for 2008-2022. Prior to our work, there was no suitable price comparison index for adjusting economic data for regional price differences for periods outside this range. We draw on the BEA’s RPP methodology to calculate a CPI-based RPP. This method requires estimating a reliable relative price level difference in a baseline (or average year). However, differences between the CPI and PCE Price Index require weighting adjustments. We use a regression of CPI components to weight the implicit regional price deflators of components of the BEA RPP to estimate a CPI-based average relative price level difference for 2008-2022. We use this estimate to re-base the average regional CPIs and the US CPI and we use changes in regional CPIs to estimate a CPI-based RPP for any period with local CPI data. Subsequently, changes in the CPI can be used to calculate the relative price level differences in any year where CPI data is available. We demonstrate possible applications for the CPI-based RPP by applying it to various measures of metropolitan economic activity.
Bond-Smith, S., and Lee, Y. Regional Price Differences and Measuring Economic Distress.
∇ Abstract
This paper surveys measures of economic distress in the US and adjusts these measures for regional cost-of-living differences using the BEA’s RPP index and a CPI-based RPP index. By accounting for regional price differences, this re-evaluation reveals a more accurate account of economic distress as experienced by residents in those regions. The regional-price-parity adjustment also allows for lower income regions to be considered not in distress, and even thriving, in spite of lower incomes, due to lower costs of living. Perhaps more importantly, it reveals places that are facing economic distress but are otherwise masked in nominal economic performance data by high local prices. Such regions may be able to use this information to appear for ‘A special need’ when applying for Federal economic development support.
Bond-Smith, S., and Lee, Y. Club Convergence.
∇ Abstract
This paper uses Personal Per Capita Income data from Bureau of Economic Analysis (BEA) to analyze club convergence in Metropolitan Statistical Area and County levels.