Premiums and discounts of exchanged-traded funds

Abstract

The objective of this study is to determine whether the spread in underlying exchange-traded fund (ETF) investments is a significant cause of the premium/discount of the ETF. Spreads of underlying investment portfolios are alternatively calculated using weighted bid-ask and bid-close spreads for a sample of ETFs listed on the Johannesburg Stock Exchange (JSE) in South Africa from 2010 to 2014. Results show that spreads of underlying investment portfolios are positively associated with larger premiums/discounts of ETFs as a whole. However, stratified results show that this relationship exists only for premiums; underlying spreads are not significantly associated with discounts. In addition, the findings show that expense ratios offer a significant explanation for premiums/discounts of ETFs. This paper contributes to the existing literature by offering an explanation for the size of premiums of ETFs at reporting date. Its findings imply that relative illiquidity in the underlying portfolio of the ETF means that a premium will likely persist. A deeper understanding in this regard assists investors in determining whether an ETF premium is worth paying for. In addition, this paper reveals that premiums and discounts of ETFs do not always arise from the same causes and should be investigated as separate phenomena in future research.

Keywords:

• Exchange-traded fund
• ETF
• premiums
• discounts
• bid-ask
• bid-close

Acknowledgements

The author would like to thank Marna de Klerk, Petri Ferreira, Elmar Venter and three anonymous reviewers for helpful comments and suggestions.

ORCiD

Wessel M. Badenhorst http://orcid.org/0000-0001-6159-9158

Notes

1. Dividend tax is not a consideration for this study. While Blitz et al. (2012) show that dividend tax has an impact on tracking errors of ETFs, it relates to tax withheld on dividends paid to the ETF. The characteristics of the ETFs included in the sample of this paper imply that only South African dividend tax is applicable, which is only withheld upon payment of a dividend by the ETF.

2. The data for this study is only updated annually by the sample ETFs. Spreads are therefore the most appropriate measure of liquidity, as they can be calculated at a specific point in time, whereas other measures tend to reflect liquidity over a period of time.

3. The sample size is comparable to that of similar prior research (e.g. Badenhorst, 2015; Barnhart & Rosenstein, 2010; Chu, 2011; Milonas & Rompotis, 2006). The sample size is therefore inherent to the nature of the study and not specific to the sample country or sample period.

4. Both Gow, Ormazabal and Taylor (2010) and Thompson (2011) find that robust standard errors clustered in two dimensions are the most appropriate method to correct for serial and cross-sectional correlation within smaller samples (although also appropriate for larger samples).

5. Another method of dealing without outliers requires the winsorisation of variables. However, winsorisation is less effective within a small sample size, as the outlying observation is retained within the sample (for this reason deletion has been preferred for the purposes of this paper). If variables are winsorised at the 1% level, inferences for TER remain qualitatively unchanged, while all other variables are insignificant.