Is the best generalized autoregressive conditional heteroskedasticity(p,q) value-at-risk estimate also the best in reality? An evidence from Australian interconnected power markets

Rangga Handika, Sigit Triandaru

Research output: Contribution to journalArticlepeer-review

Abstract

This paper investigates whether the best value-at-risk (VaR) estimate will also perform the best in empirical performance. The study explores the linkage between statistical world and reality. This paper uses VaR generalized autoregressive conditional heteroskedasticity (GARCH)(p,q) estimates and performs the back testing from both generator (buyer) and retailer (seller) sides, at different confidence levels, and at different out-of-sample periods in the four regions of Australian interconnected power markets. Using VaR approach, we find that the best GARCH(p,q) model tends to generate best empirical performance. Our findings are consistent for both generator (buyer) and retailer (seller) sides, at different confidence levels and at different out-of-sample periods. However, our strong results are only in the daily series. Therefore, our study has two important practical implications in Australian power markets. First, generator and retailer can continue choosing the best GARCH(p,q) model based on statistical criteria. Second, the users of GARCH(p,q) model should be aware that the model tends to be appropriate for estimating the daily series only.

Original languageEnglish
Pages (from-to)814-821
Number of pages8
JournalInternational Journal of Energy Economics and Policy
Volume6
Issue number4
Publication statusPublished - 1 Jan 2016

Keywords

  • Generalized autoregressive conditional heteroskedasticity
  • Power markets
  • Value-at-risk

Fingerprint

Dive into the research topics of 'Is the best generalized autoregressive conditional heteroskedasticity(p,q) value-at-risk estimate also the best in reality? An evidence from Australian interconnected power markets'. Together they form a unique fingerprint.

Cite this