AI Forecasting Analytics Lab

Publications

Selected Publications

Grouped by research area. Click any paper to expand its topic, methodology, and results — each lists its published journal.

Text & Media Big-Data Analytics8

AI-Data-Based Art-Therapy Convergence MajorJ. of Learner-Centered Curriculum & Instruction, 20(19), 2020

TopicDesigning a convergence major that fuses art therapy with AI and data skills for the 4th industrial revolution.

MethodologySurveyed 113 art-therapy students with frequency and regression analysis, plus morphological keyword analysis of open-ended responses.

ResultsStudents saw AI-data convergence as a key opportunity; prior awareness and AI-data knowledge shaped major choice, yielding a proposed six-semester curriculum.

Have the Elderly Really Changed? The 'Active Senior'Korean J. of Communication Studies, 24(2), 2025

TopicWhether an economically- and digitally-capable 'active senior' cohort is emerging in Korea.

MethodologyMerged 10 years (2014–2023) of Financial-Panel and media-audience data across six age groups for time-series comparison.

ResultsThe 55–74 groups narrowed leisure/telecom-spend and portal/TV-news gaps versus 45–54 while wealth-income gaps widened — evidence of active seniors.

Older Adults' Political Orientation in the Digital-Transformation EraKorean J. of Journalism & Communication Studies, 69(2), 2025

TopicThe interplay of digital literacy, online participation, and political orientation among older adults.

Methodology2020–2023 Korea Media Panel and Press-Foundation data analyzed with statistical matching.

ResultsOlder adults (especially 55–64) improved digital competencies and narrowed the gap with younger cohorts, with associated shifts in engagement and political preference.

China-Market Distribution & Tourism Strategy via Big-DataKorea Logistics Review, 27(2), 2017

TopicStrategy to attract Chinese tourists to Korea's distribution and tourism sectors.

MethodologyText-network (semantic) analysis of 2015 news containing 'China' and 'tourism', extracting centrality and topic clusters.

ResultsMarketing centered on duty-free/hospitality (Seoul-biased) while destination and cultural value were under-leveraged — recommends stronger linkage, regional diversification, and active Hallyu use.

Positive/Negative Milk-Consumption News Topics via Generative AIHealth Communication Research, 23(2), 2024

TopicMapping positive and negative issues around milk consumption in the news.

MethodologyGenerative-AI semantic-network and topic modeling over 1,338 milk-related articles across five years.

ResultsPositive topics (health, safety, sustainable/subscription consumption, protein products, logistics) and negative ones (declining consumption, plant-based substitutes, price rises, low birthrate, COVID) — insight for dairy policy.

Semantic-Network Analysis of Community-Sports News in Aged RegionsJ. of Korea Society for Wellness, 2026

TopicCommunity-sports news topics across ten aging and super-aged cities in northern Gyeonggi Province.

MethodologySemantic-network analysis of community-sports news coverage.

ResultsSurfaces latent resident policy demands beyond the usual supplier-centric view, supporting region- and lifecycle-tailored community-sports policy.

Role of K-Culture for the Vietnam Market after COVID-19J. of Product Research, 39(6), 2021

TopicLeveraging Hallyu (K-culture) to appeal to the Vietnamese market after COVID-19.

MethodologyMorphological and text-network analysis of five years of 'Vietnam tourist' news.

ResultsKorea–Vietnam is strongly linked via football/sports, but the Hallyu–product/retail linkage is weak — recommends Hallyu-linked hit products and systematic promotion.

Organizational Communication in Early COVID-19: 34 UniversitiesJ. of Practical Advertising & PR, 16(3), 2023

TopicCrisis-response communication of organizations in the early COVID-19 period.

MethodologySMCRE-model analysis of 1,761 messages from 34 Korean universities over six months.

ResultsMetro-area and large universities communicated crisis response and stakeholder safety more promptly, via president-level sources and dedicated COVID channels.

Bio-Healthcare Machine Learning2

Multi-Task Gradient Boosting with Multi-Modal Molecular RepresentationsPLOS One, 21(4), 2026

TopicPredicting two pharmacokinetic parameters — drug clearance and volume of distribution — simultaneously.

MethodologyMulti-task gradient boosting over multi-modal molecular representations.

ResultsAccurate, scalable, and interpretable simultaneous PK prediction directly from chemical structure.

Multicenter Validation of a Scalable, Interpretable, Multitask Clinical Prediction Modelnpj Digital Medicine (Nature), 8:583, 2025

TopicPredicting multiple post-operative complications together rather than one at a time.

MethodologyTree-based multitask learning on 16 routine preoperative EHR features, externally validated across cohorts.

ResultsAUROCs ~0.81 (AKI), ~0.89–0.93 (respiratory failure), ~0.85–0.91 (mortality), with interpretable per-variable contributions.

Generative-AI Applications2

Domestic Applicability of SASB ESG Issues via Generative-AI EmbeddingJ. of the Korean OR/MS Society, 50(3), 2025

TopicWhether SASB's industry-specific ESG issues fit Korean industries, using healthcare as the case.

MethodologyGenerative-AI sentence embeddings to classify corporate healthcare news and surface recurrent issues.

ResultsNews-based ESG salience aligns significantly with SASB (p<0.05), yet some domestically distinct issues reveal industry- and region-specific peculiarity.

Trust-by-Design of an Avatar Chatbot for Major/Career AdvisingJ. of Digital Contents Society, 26(12), 2025

TopicIntegrating multi-layered trust signals into a multimodal RAG avatar chatbot for major and career advising.

MethodologyDerived trust signals across identity, response quality, interaction naturalness, and operational-policy transparency, then measured user perception.

ResultsPolicy-transparency signals were perceived most strongly while affective response quality lagged the cognitive side, pinpointing where trust design breaks down.

Pension & Policy Microsimulation4

Empirical Analysis on the Elasticity of Tax Revenue in KoreaKorean Economic Forum (formerly Yonsei Economic Studies), 22(1), 2015

TopicEstimating the elasticity of national tax revenue with respect to its economic base.

MethodologyEmpirical elasticity analysis of Korean tax-revenue and macroeconomic data.

ResultsQuantifies how responsive tax revenue is to the base, informing revenue forecasting and fiscal-policy design.

Effects of Basic-Pension Changes on Income Distribution and PovertyKorean Association of Public Finance, 2018 Autumn Conference Proceedings

TopicDistributional and poverty effects of Korea's Basic Pension reforms.

MethodologyMicrosimulation on the 10th Financial-Panel, comparing no-pension, current, and reformed scenarios.

ResultsThe Basic Pension improved income distribution (Gini, quintile ratio) and reduced poverty; benefit-level hikes helped only modestly.

Crowding-Out of Livelihood Benefits on the Basic Pension's EffectJ. of Fiscal Policy, 23(4), 2021

TopicWhether livelihood-benefit crowd-out weakens the Basic Pension's redistribution and poverty relief.

MethodologyMicrosimulation on the 13th Financial-Panel.

ResultsThe Basic Pension substantially improves redistribution and reduces poverty; recent reforms and the crowd-out effect are small because few households receive both benefits.

Probabilistic Estimation of the National Pension's Unfunded LiabilityJ. of Pension Studies, 15(1), 2025

TopicQuantifying the uncertainty in the National Pension's unfunded liability.

MethodologyMonte-Carlo simulation over policy scenarios (replacement and contribution rates) and environment scenarios (wage growth, benefit growth, discount rate).

ResultsProduces a full probability distribution instead of a single point estimate — a risk-based tool for pension-reform design.

Corporate Finance, Accounting & ESG5

Effects of Temporary Earnings on Dividends in KoreaApplied Economics Letters (Taylor & Francis), 29(21), 2044–2046, 2022

TopicHow temporary (transitory) earnings affect corporate dividend policy in Korea.

MethodologyAnalyzed cash-dividend data of Korean firms from 2001 to 2017, split by debt type.

ResultsPrivate-debt-only firms raise dividends with temporary earnings (dampened by investment opportunities); public-debt firms do not — supporting precautionary-savings and signaling hypotheses.

Effects of Incentive Alignment on Debt Maturity StructureJ. of Insurance and Finance, 23(3), 2012

TopicWhether executive stock-linked pay (incentive alignment) shapes corporate debt maturity.

MethodologyAnalyzed interactions of managerial ownership with liquidity risk and investment opportunities.

ResultsOwnership shortens maturity when liquidity risk is low but lengthens it when high; high-growth firms prefer short-term debt — consistent with the under-investment hypothesis.

Negative ESG News and Firm ValueKorean J. of Broadcasting & Telecommunication Studies, 38(4), 2024

TopicThe effect of negative ESG news coverage on firm value (market-to-book ratio).

MethodologyWeb-scraping plus the ChatGPT-4.0 API to classify 30,098 negative ESG articles for 1,845 listed firms (2019–2022); panel regression.

ResultsNegative ESG news significantly lowers MTB, driven mainly by Governance issues, and strongest in healthcare/pharma, chemicals, and finance.

Tax-Evasion Controversy and Firm Value via Big-DataJ. of Internet Computing and Services, 22(6), 2021

TopicWhether tax-evasion controversy in portal news erodes firm value.

MethodologyCrawled portal news to build per-firm keyword-frequency time series; panel regression and impulse-response analysis.

ResultsHigher controversy frequency lowers firm value, with the effect decaying gradually over about 12 months.

Determinants of Corporate-Bond Rating DisagreementJ. of Korean Economic Studies, 33(2), 2015

TopicWhy credit-rating agencies disagree on the same corporate bond.

MethodologyEmpirical analysis relating split ratings to issuer financial characteristics and information asymmetry.

ResultsIdentifies the factors that drive rating disagreement, with implications for bond pricing and rating quality.

Financial & Economic Forecasting4

A Simultaneous Test of Unit Root and Level ChangeJournal of Forecasting (Wiley), 29, 301–312, 2010

TopicDistinguishing whether a time series has a unit root, a level change, or both — usually tested separately.

MethodologyDerived a likelihood-ratio test from a state-space model, with null distributions built by simulation.

ResultsThe joint test identifies the true underlying structure; validated on simulations and two Korean macroeconomic series.

Distribution-Based Level Change Detection in a Random-Level Forecasting ModelJ. of the Korean Institute of Industrial Engineers, 43(4), 2017

TopicUnexpected level changes cause persistent forecasting errors in macro and financial series.

MethodologyDerived the correct statistical distribution of the level-change statistic and adapted the forecasting equation accordingly.

ResultsImproved change detection and forecast accuracy, validated on simulations and two empirical cases.

Long-Term Mean Reversion of Stock Prices Based on Fractional IntegrationInternational Journal of Management Science, 17(2), 2011

TopicThe very-long-horizon behavior of stock returns versus random-walk models.

MethodologyIntroduced a fractionally integrated process into the nonstationary component of a stock-price model and tested NYSE returns.

ResultsNegative autocorrelations persist up to ~10 years, implying market inefficiency can last far longer than random-walk models predict.

Pairs-Trading Algorithm with Structural-Change DetectionJ. of the Korean OR/MS Society, 42(3), 2017

TopicImproving pairs-trading performance by handling structural breaks in cointegration.

MethodologyEmbedded a structural-change detection procedure (cointegration breaks, unit-root tests) into the algorithm, tested by simulation and empirics.

ResultsOutperforms the baseline on risk-adjusted return, with a steady upward cumulative P/L at low variance.