Abstract | PDF (765 kilobytes)

No abstract is available for this article.

Abstract | PDF (5061 kilobytes)

This study aims to decompose recent employment fluctuations in Korea into
structural and cyclical components and forecast future employment trends. To
achieve this, we develop and estimate a trend-cycle hidden factor model that
incorporates Korea's macroeconomic environment and demographic structure.
Within the model, employment is modeled as the sum of a unit root process
(trend) and a stationary process (cycle). The trend and cycle are each designed to
have a dynamic correlation with key macroeconomic variables and demographic
structure. The main results are threefold. First, the cyclical component of
employment for all ages as of the fourth quarter of 2023 is estimated at 150,000
and 130,000 for those under 60. Second, the trend
of employment for all ages and those under 60 is mainly determined by the
population growth rate and aging rather than the potential growth rate. On the
other hand, the cyclical component is closely related to the GDP gap and economic
sentiment index. Finally, employment for all ages
is expected to increase by 228,000 by the end of 2024, of which 74,000 is cyclical.
On the other hand, employment for those under 60 is expected to decrease by
115,000, but the cyclical component is 56,000, indicating that employment is
expected to exceed the trend.

Abstract | PDF (4067 kilobytes)

In this study, we explore the use of generative models for time series prediction and construction of prediction intervals, addressing the challenge of quantifying uncertainty in deep learning models. Specifically, we employ a Variational Autoencoder (VAE), a form of a Bayesian neural network also a part of generative AI models, to model and generate latent factors for the exchange rates of ten currencies. These latent factors enable the approximate reconstruction of the exchange rate series through the decoder part of VAE. By generating a thousand sets of latent factors and reconstructing exchange rates, we create prediction intervals through a Multi-Layer Perceptron (MLP) applied to the reconstructed series. This approach provides valuable insights into evaluation of the uncertainty associated with time series predictions using neural networks.

Abstract | PDF (344 kilobytes)

This study develops a KPSS (Kwiatkowski et al., 1992)-type cointegration test utilizing residuals from integrated and modified ordinary least squares (IMOLS) estimation. The test statistic, denoted by $KPSS^{Fb}$ has a pivotal null limit distribution under fixed-$b$ assumption. The proposed test demonstrates reasonable performance in terms of size and power when the Andrews' AR(1) plug-in data-dependent ($DD$) bandwidth is employed and fixed-$b$ critical values are used. Additionally, two modified IMOLS residuals are proposed to obtain alternative data-dependent bandwidths. In the simulation experiment, these bandwidths deliver improved power properties for the proposed test.

Abstract | PDF (2916 kilobytes)

This study employs machine learning to analyze news article sentiment, developing a stock market sentiment index (SSI) based on this analysis. By examining the textual data from news articles, which constitute unstructured data, we aimed to capture the prevailing sentiments among market participants across the financial market. Specifically, this study utilizes The BERT model to decipher the psychological sentiment embedded in the articles through its contextualized understanding of the tone and language patterns. The variables tested included the risk aversion estimate, calculated using the VKOSPI and Bekaert’s method for assessing risk aversion. The empirical analysis involving the SSI, VKOSPI, and risk aversion reveals a significant negative impact of SSI on VKOSPI and risk aversion. We further find that the news sentiment index (NSI) and SSI simultaneously exhibit a converging trend.