A Comparative Study of the Temporal Characteristics of Closed-End Fund Discounts/Premiums

Abstract

The primary objective of this study is to compare the difference in temporal characteristics between municipal bond funds and domestic equity funds. The secondary objective is to identify the characteristics of the temporal behavior of closed-end fund discounts/premiums from a class of linear stochastic autoregressive-integrated-moving average (ARIMA (p, d, q)) models popularized by Box-Jenkins (1994). The tertiary objective is to examine the ability of the individually identified models to forecast out-of-sample closed-end fund discounts/premiums. Comparisons of the forecasting abilities will be made between the individually identified ARIMA (p, d, q) and random walk models using mean absolute percent error metric (MAPE) as criterion. For the weekly series our results show: out of 27 municipal bond closed-end funds, there are five following random walk model, thirteen following ARIMA(1,0,0) model, five following ARIMA(0,1,1) model, two following ARIMA(2,1,0) and one following each of ARIMA(0,1,2), and ARIMA(1,0,1) models; for the domestic equity (core) funds, out of twenty funds one follows the random walk model, six follow ARIMA(1,0,0) model, eight follow ARIMA(0,1,1), two follow ARIMA(2,1,0) and one follows each of ARIMA(0,0,1), ARIMA(0,0,2) and ARIMA(0,1,3) model . Except for one week out of the four weeks examined the forecast results for the weekly municipal bond closed-end funds do not show significant difference between the fund-specific ARIMA (p, d, q) models and the random walk model. Likewise, the forecast results for the weekly core funds do not show significant difference between the fund-specific ARIMA (p, d, q) models and the random walk model which is not consistent with the findings in Woan and Kline (2008). Our results show clearly that the ARIMA (p, d, q) models provide superior forecasts for the core funds with less than seventeen percent mean and median APEs to municipal bond funds with over fifty percent mean and median APEs in terms mean and median APEs. These results could be because the current data are from the rare persistent recessionary period of 2011 and 2012. Further research is needed to explain this phenomenon.