Massey University – School of Economics and Finance
Massey University – Department of Economics and Finance
October 14, 2015
We provide a comprehensive analysis of ETF liquidity based on tick and daily data for in excess of 800 ETFs over the 1996-2014 period. We make several contributions. First, we show there is a strong relation between ETF and underlying stock liquidity. ETF liquidity is influenced by and influences underlying stock liquidity. Second, both supply- and demand-side factors have a strong bearing on ETF liquidity. Third, the end-of-day spread, high-low, and Amihud proxies best measure actual transaction costs. Fourth, there are intraday and daily patterns in spreads. Finally, liquidity risk is priced in ETF returns.
ETF Liquidity – Introduction
Liquidity has received a lot of attention from researchers due to the critical role it plays in financial markets. Papers have documented for stocks, bonds, commodities, and currencies a) the level of transaction costs, b) how best to measure liquidity, c) variation in liquidity across different time periods, and d) the link between liquidity and returns.
However, there has been relatively little focus on exchange traded fund (hereafter ETF) liquidity despite the surge in ETF investment, the increased focus on ETFs from researchers, and the unique characteristics of ETFs. We provide a comprehensive examination of ETF liquidity using tick and daily data for over 800 ETFs.
There has been a sharp increase in the number, variety, and asset value of ETFs. While the original ETFs were backed by physical assets, many current ETFs now provide asset class exposure via derivative contracts. Early ETFs provided long-only equity market exposure but it is now possible to find ETFs with short and leveraged exposure to equity sectors, styles, bonds, commodities, currencies, real estate, and other “exotic” asset classes. The assets of U.S. listed ETFs are estimated to have grown to over US1.7 trillion by the beginning of 2014, with ETFs typically representing between 25% to 40% of the total dollar volume of trading on U.S. exchanges (e.g. Hill, Nadig, and Houhan, 2015).
There are several reasons why ETF liquidity may differ from the liquidity of other instruments such as stocks. First, Kyle (1985) and Admati and Pfleiderer (1988) show there can be a reluctance to supply liquidity when there are concerns that other market participants have superior information. Adverse selection costs are therefore an important aspect of liquidity and liquidity proxies (e.g. Glosten, 1987). However, Subrahmanyam (1991) posits that a basket of securities (such as an ETF) should have higher liquidity than the underlying stocks due to lower adverse selection costs. Second, Bessembinder, Carrion, Tuttle, and Venkataraman (2015) find that the liquidity of ETFs holding futures contracts is affected by the rolling of contracts. Spreads are lower and depth is higher on roll dates.
While it is well known that the first ETFs on large equity indices have extremely low bid-ask spreads (e.g. Hill, Nadig, and Hougan, 2015), the relatively large transaction costs of more exotic ETFs are also receiving attention.2 Our first contribution is providing comprehensive measures of transaction costs across ETFs and through time. These will likely to be of interest to a number of parties. The first group are investors and researchers. Lesmond, Schill, and Zhou (2004) show transaction costs are a crucial input into analysis to determine whether active investment strategies add value, while De Roon, Nijman, and Werker (2001) find apparent diversification advantages often disappear when transaction costs are accounted for. The second group is risk managers. Elton, Gruber, Comer, and Li (2002) find ETFs are popular hedging tools and Perold and Schulman (1988) point out that transaction costs are a key input in judging hedging effectiveness. Transaction costs are also very relevant for exchanges, as Harris (2003) suggests they are an important determinant of the level of business they attract. Finally, Chordia, Roll, and Subrahmanyam (2008) show that there is a direct link between transaction costs and market efficiency so transaction costs affect all market participants.
Our sample includes US, international, market, style, and sector equity ETFs, and bond, commodity, currency, real estate, and “other” ETFs, which include those providing more exotic exposure, such as long – short positions. Our results show average effective spreads over the 1996 – 2014 period range from 0.187% for commodities to 0.572% for international equities. Some of the variation in spread across asset classes is accounted for by the different start dates and the trend of lower transaction costs through time. However, this is not the entire explanation. While there are many very liquid international equity ETFs, there are also a large number that are relatively illiquid. Overall, ETF transaction costs compare very favorably to the average developed and emerging market effective spreads of 2.2% and 2.7% documented by Fong, Holden, and Trzcinka (2010) for the 1996 – 2007 period.
Our second contribution is comparing the transaction costs of the most liquid Dow Jones Industrial Average (DIA) ETF to the transaction costs of the underlying stocks over time. Hedge and McDermott (2004) provide empirical evidence that transaction costs are lower for this ETF than for the basket of underlying securities over the fifty days following the launch of the ETF, and we add to their work by measuring transaction costs from inception in 1998 to 2014. The ETF effective spreads declined from 0.075% in the 1998 – 2005 period to 0.021% in the 2006 – 2014 period, while the underlying stock transaction costs declined from 0.083% to 0.023% over the same two periods. The fact that ETF transaction costs are lower in both periods and in the overall sample confirms Subrahmanyam’s (1991) suggestion that ETFs should have higher liquidity than the underlying stocks due to lower adverse selection costs. Movements in underlying stock liquidity influence future ETF liquidity, and while there is evidence of ETF liquidity also influencing underlying asset liquidity, this relation is weaker.
See full PDF below.