BOOTSTRAP BASED MULTI-STEP AHEAD JOINT FORECAST DENSITIES FOR FINANCIAL INTERVAL-VALUED TIME SERIES
COMMUNICATIONS FACULTY OF SCIENCES UNIVERSITY OF ANKARA-SERIES A1 MATHEMATICS AND STATISTICS, vol.70, no.1, pp.156-179, 2021 (ESCI, TRDizin)
- Publication Type: Article / Article
- Volume: 70 Issue: 1
- Publication Date: 2021
- Doi Number: 10.31801/cfsuasmas.534711
- Journal Name: COMMUNICATIONS FACULTY OF SCIENCES UNIVERSITY OF ANKARA-SERIES A1 MATHEMATICS AND STATISTICS
- Journal Indexes: Emerging Sources Citation Index (ESCI), TR DİZİN (ULAKBİM)
- Page Numbers: pp.156-179
- Keywords: Multivariate forecast, resampling methods, interval-valued time series, vector autoregressive model
- Dokuz Eylül University Affiliated: No
Abstract
This study presents two interval-valued time series approaches to construct multivariate multi-step ahead joint forecast regions based on two bootstrap algorithms. The first approach is based on fitting a dynamic bivariate system via a VAR process for minimum and maximum of the interval while the second approach applies for mid-points and half-ranges of interval-valued time series. As a novel perspective, we adopt two bootstrap techniques into the proposed interval-valued time series approaches to obtain joint forecast regions of the lower/upper bounds of the intervals. The forecasting performances of the proposed approaches are evaluated by extensive Monte Carlo simulations and two real-world examples: (i) monthly S&P 500 stock indices; (ii) monthly USD/SEK exchange rates. Our results demonstrate that the proposed approaches are capable of producing valid multivariate forecast regions for interval-valued time series.