• Communications for Statistical Applications and Methods
• /
• v.30 no.3
• /
• pp.273-289
• /
• 2023

In this paper, we develop a new time series model for predicting IPO (initial public offering) data with non-negative integer value. The proposed model is based on integer-valued autoregressive (INAR) model with a Poisson thinning operator. Just as the heterogeneous autoregressive (HAR) model with daily, weekly and monthly averages in a form of cascade, the integer-valued heterogeneous autoregressive (INHAR) model is considered to reflect efficiently the long memory. The parameters of the INHAR model are estimated using the conditional least squares estimate and Yule-Walker estimate. Through simulations, bias and standard error are calculated to compare the performance of the estimates. Effects of model fitting to the Korea's IPO are evaluated using performance measures such as mean square error (MAE), root mean square error (RMSE), mean absolute percentage error (MAPE) etc. The results show that INHAR model provides better performance than traditional INAR model. The empirical analysis of the Korea's IPO indicates that our proposed model is efficient in forecasting monthly IPO volumes.

• Communications for Statistical Applications and Methods
• /
• v.31 no.1
• /
• pp.105-127
• /
• 2024

We propose recurrent neural network (RNN) methods for forecasting realized volatility (RV). The data are RVs of ten major stock price indices, four from the US, and six from the EU. Forecasts are made for relative ratio of adjacent RVs instead of the RV itself in order to avoid the out-of-scale issue. Forecasts of RV ratios distribution are first constructed from which those of RVs are computed which are shown to be better than forecasts constructed directly from RV. The apparent asymmetry of RV ratio is addressed by the Piecewise Min-max (PM) normalization. The serial dependence of the ratio data renders us to consider two architectures, long short-term memory (LSTM) and gated recurrent unit (GRU). The hyperparameters of LSTM and GRU are tuned by the nested cross validation. The RNN forecast with the PM normalization and ratio transformation is shown to outperform other forecasts by other RNN models and by benchmarking models of the AR model, the support vector machine (SVM), the deep neural network (DNN), and the convolutional neural network (CNN).

• The Korean Journal of Applied Statistics
• /
• v.31 no.1
• /
• pp.139-153
• /
• 2018

This paper investigates solar power forecasting based on several time series models. First, we consider weather variables that influence forecasting procedures as well as compare forecasting accuracies between time series models such as ARIMAX, Holt-Winters and Artificial Neural Network (ANN) models. The results show that ten models forecasting 24hour data have better performance than single models for 24 hours.

• The Korean Journal of Applied Statistics
• /
• v.28 no.6
• /
• pp.1235-1243
• /
• 2015

Increased data volume in the ICT area has increased the importance of forecasting accuracy for internet traffic. Forecasting results may have paper plans for traffic management and control. In this paper, we propose combined forecasts based on several time series models such as Seasonal ARIMA and Taylor's adjusted Holt-Winters and Fractional ARIMA(FARIMA). In combined forecasting methods, we use simple-combined method, MSE based method (Armstrong, 2001), Ordinary Least Squares (OLS) method and Equality Restricted Least Squares (ERLS) method. The results show that the Seasonal ARIMA model outperforms in 3 hours ahead forecasts and that combined forecasts outperform in longer periods.

• Journal of Korean Institute of Industrial Engineers
• /
• v.28 no.1
• /
• pp.63-75
• /
• 2002

To improve the efficiency of the electric power generation, monthly maximum electric power consumptions for a next one year should be forecasted in advance and used as the fundamental input to the yearly electric power-generating master plan, which has a greatly influence upon relevant sub-plans successively. In this paper, we analyze the past 22-year hourly maximum electric load data available from KEPCO(Korea Electric Power Corporation) and select necessary data from the raw data for our model in order to reflect more recent trends and seasonal components, which hopefully result in a better forecasting model in terms of forecasted errors. After analyzing the selected data, we recommend to KEPCO the Winters' multiplicative model with decomposition and exponential smoothing technique among many candidate forecasting models and provide forecasts for the electric power consumptions and their 95% confidence intervals up to December of 1999. It turns out that the relative errors of our forecasts over the twelve actual load data are ranged between 0.1% and 6.6% and that the average relative error is only 3.3%. These results indicate that our model, which was accepted as the first statistical forecasting model for monthly maximum power consumption, is very suitable to KEPCO.

• Journal of Environmental Science International
• /
• v.22 no.8
• /
• pp.953-963
• /
• 2013

Reliable long-term streamflow forecasting is invaluable for water resource planning and management which allocates water supply according to the demand of water users. Forecasting of seasonal inflow to Andong dam is performed and assessed using statistical methods based on hydrometeorological data. Predictors which is used to forecast seasonal inflow to Andong dam are selected from southern oscillation index, sea surface temperature, and 500 hPa geopotential height data in northern hemisphere. Predictors are selected by the following procedure. Primary predictors sets are obtained, and then final predictors are determined from the sets. The primary predictor sets for each season are identified using cross correlation and mutual information. The final predictors are identified using partial cross correlation and partial mutual information. In each season, there are three selected predictors. The values are determined using bootstrapping technique considering a specific significance level for predictor selection. Seasonal inflow forecasting is performed by multiple linear regression analysis using the selected predictors for each season, and the results of forecast using cross validation are assessed. Multiple linear regression analysis is performed using SAS. The results of multiple linear regression analysis are assessed by mean squared error and mean absolute error. And contingency table is established and assessed by Heidke skill score. The assessment reveals that the forecasts by multiple linear regression analysis are better than the reference forecasts.

• Journal of Digital Contents Society
• /
• v.13 no.3
• /
• pp.325-334
• /
• 2012

Production of drama programs has been granted as a creative work. Thus Systematic approaches to improving quality of drama programs have hardly been tried. This research is for producing a statistical computing model that is capable of forecasting on popularity rating of drama programs produced by KBS, especially forecasting prior to the broadcasting. For the work, we traced various factors affecting on drama popularity ratings, found the relationships among the factors with a regression analysis work, and created the forecasting model on drama popularity ratings. The research result could be applied for finding proper scales of input factors necessary to drama program productions.

• The Korean Journal of Applied Statistics
• /
• v.26 no.6
• /
• pp.1053-1061
• /
• 2013

In this paper, we propose the time series forecasting models for internet traffic with long memory and heteroscedasticity. To control and forecast traffic volume, we first introduce the traffic forecasting models which are determined by the volatility and heteroscedasticity of the traffic. We then analyze and predict the heteroscedasticity and the long memory properties for forecasting traffic volume. Depending on the characteristics of the traffic, Fractional ARIMA model, Fractional ARIMA-GARCH model are applied and compared with the MAPE(Mean Absolute Percentage Error) Criterion.

• Journal of the Institute of Convergence Signal Processing
• /
• v.2 no.3
• /
• pp.73-78
• /
• 2001

BP neural network model and multiple-regression model were composed for forecasting the special-days load. Special-days load was forecasted using that neural network model made use of pattern conversion ratio and multiple-regression made use of weekday-change ratio. This methods identified the suitable as that special-days load of short and long term was forecasted with the weekly average percentage error of 1∼2[%] in the weekly peak load forecasting model using pattern conversion ratio. But this methods were hard with special-days load forecasting of summertime. therefore it was forecasted with the multiple-regression models. This models were used to the weekday-change ratio, and the temperature-humidity and discomfort-index as explanatory variable. This methods identified the suitable as that compared forecasting result of weekday load with forecasting result of special-days load because months average percentage error was alike. And, the fit of the presented forecast models using statistical tests had been proved. Big difficult problem of peak load forecasting had been solved that because identified the fit of the methods of special-days load forecasting in the paper presented.

• Atmosphere
• /
• v.19 no.3
• /
• pp.269-287
• /
• 2009

CAT (clear-air turbulence) forecasting algorithm, the Graphical Turbulence Guidance (GTG) system developed at NCAR (national center for atmospheric research), is evaluated with available observations (e.g., pilot reports; PIREPs) reported in South Korea during the recent 5 years (2003-2008, excluding 2005). The GTG system includes several steps. First, 44 CAT indices are calculated in the domain of the Regional Data Assimilation and Prediction System (RDAPS) analysis data with 30 km horizontal grid spacing provided by KMA (Korean Meteorological Administration). Second, 10 indices that performed ten best forecasting scores are selected. Finally, 10 indices are combined by measuring the score based on the probability of detection, which is calculated using PIREPs exclusively of moderate-or-greater intensity. In order to investigate the best performance of the GTG system in Korea, various statistical examinations and sensitivity tests of the GTG system are performed by yearly and seasonally classified PIREPs. Performances of the GTG system based on yearly distributed PIREPs have annual variations because the compositions of indices are different from each year. Seasonal forecasting is generally better than yearly forecasting, because selected CAT indices in each season represent meteorological condition much more properly than applying the selected CAT indices to all seasons. Wintertime forecasting is the best among the four seasonal forecastings. This is likely due to that the GTG system consists of many CAT indices related to the jet stream, and turbulence associated with the jet stream can be activated mostly in wintertime under strong jet magnitude. On the other hand, summertime forecasting skill is much less than other seasons. Compared with current operational CAT prediction system (KITFA; Korean Integrated Turbulence Forecasting System), overall performance of the GTG system is better when CAT indices are selected seasonally.