Analyzing the contribution of each predictor variable in forecasting models is a critical task.To address this, Lundberg and Lee (2017) proposed the SHAP (SHapley Additive exPlanations) as a method to explain the contribution of individual predictors in machine learning forecasts.However, in time series forecastings that require consideration of lagged variables, the SHAP requires heavy computation and makes it difficult to evaluate the contribution of predictor vectors composed of lagged variables.To overcome these limitations, Choi et al. (2024) introduced the vector SHAP, which measures the contribution of each predictor vector composed of variable lags.The vector SHAP not only reduces computation compared to the SHAP but also provides insights into the contributions of predictors that include lagged variables.In this study, we construct realized volatility forecasting models for the S&P 500 using realized volatility, implied volatility, the search volume index, and their lagged variables as predictors, and apply the vector SHAP to evaluate their contributions.We compare the predictor vector contributions in two models: the ADL model, a statistical time series framework, and the LSTM model, a machine learning approach.The results show that past realized volatility has the most significant impact on forecasting realized volatility, while the search volume index contributes only marginally.In the ADL model, multicollinearity often obscures the interpretation of predictor contributions, whereas this limitation is considerably reduced in the LSTM model.Furthermore, unlike in the ADL model, the LSTM model reveals nonlinear contributions of the search volume index.

Retrieval-Augmented Generation (RAG) is a method of generating text based on retrieved information, which seeks to overcome the limitations of large language models' pre-trained knowledge by retrieving external knowledge and utilizing it in responses.In RAG-based systems, retrieval performance closely affects the overall quality of responses, so various enhancement techniques to improve retrieval accuracy have emerged as an important challenge.If more precise document retrieval is possible for user queries, it is expected that accurate and reliable answers can be generated.In this paper, we propose a method to compare and analyze retrieval performance and classification performance in four domains by combining various embedding models and advanced RAG techniques to improve retrieval accuracy.To this end, we apply multiple embedding models to Korean text data collected by each domain, and quantitatively analyze the changes in retrieval performance and classification performance using various retrieval enhancement techniques such as HyDE, Multi Query, and Reranker.Finally, the performance evaluation results show that more accurate retrieval and higher classification performance can be achieved by applying advanced RAG techniques than simple embedding-based retrieval.

This study discusses the theoretical background and empirical analysis of the Difference-in-Differences (DID) method and the Doubly Robust DID (DRDID) estimator in causal inference using panel data.In causal inference, treatment effects cannot be directly measured, necessitating appropriate assumptions, with the Stable Unit Treatment Value Assumption (SUTVA) being fundamental.SUTVA assumes no interference between units and that the same treatment always produces the same outcome.However, in panel data, temporal dependencies and treatment interference may violate this assumption.The DID method, based on the Parallel Trends Assumption, considers both temporal differences and group-level differences simultaneously, making it a powerful tool in causal inference.However, the DID estimate obtained from a mixed effects model may significantly deviate from the true value, necessitating an alternative approach such as the DRDID estimator.DRDID has the advantage of providing a consistent estimator as long as either the outcome regression model or the propensity score model is correctly specified.Verifying the consistency and efficiency of the DRDID method through simulation, this study conducts an empirical analysis using the Korean Aging Study Panel Data to estimate the DRDID estimator for treatment effects within the DID framework.

Differential privacy has emerged as a key method for safeguarding data, offering a quantifiable measure of risk exposure.This paper focuses on its application in the realm of statistical analysis, particularly in addressing the challenge of model selection in generalized linear models.Our approach involves the integration of differential privacy techniques, coupled with a more flexible consideration of parameter assumptions via cross-validation.This is particularly beneficial in practical data analysis scenarios.The effectiveness and practicality of our proposed methods are demonstrated through both simulation studies and real-world applications, specifically using data from the National Health and Nutrition Examination Survey (NHANES).

In survival analysis, the use of deep neural networks has become popular. It requires the mini-batch type stochastic gradient descent (SGD) algorithm. However, the existence of risk set in the partial likelihood can be problematic, which can be addressed by many previous works. In this paper, we proposed an advanced algorithm compared to the conventional SGD by applying an online mirror descent algorithm. It can be used for any convex optimization problem where the given tasks are closely related to online learning. A re-parameterization trick and bi-level optimization are used to construct the algorithm. The experiments on various setups reveal the superiority of the proposed algorithm. It can contribute to making an efficient mini-batch-based algorithm over the convex optimization and semi-parametric survival models.

Over the past decade, the remarkable advancements in deep neural networks have paralleled the development and evolution of language models.Initially, language models were developed in the form of Encoder-Decoder models using early RNNs.However, with the introduction of Attention in 2015 and the emergence of the Transformer in 2017, the field saw revolutionary growth.This study briefly reviews the development process of language models and examines in detail the working mechanism and technical elements of the Transformer.Additionally, it explores statistical models and methodologies related to language models and the Transformer.