Exchange-traded funds (ETFs) can be traded in a variety of ways. At first glance, striking a deal in an ETF at the fund's end-of-day net asset value (NAV) seems quite different from trading at an intraday "risk" price with a market maker. However, the opposite is true and the two methods of execution are in fact closely connected. This is because an intraday risk price directly reflects the mechanism of trading the ETF at its net asset value in the primary market. Normally, if a market maker has sold ETF shares intraday, they will create the shares in the primary market in order to deliver them to the buyer. The market maker is therefore bridging the gap between the investor's wish to trade intraday and the opportunity to trade with the fund in the primary market. The primary market set-up therefore has a direct impact on the liquidity in the secondary market.
In this article we will investigate the different primary market models for ETFs and the way in which these models influence the secondary market. We will compare ETFs with different creation/redemption models and see how these models affect their liquidity. In addition, we will take a look at the extent to which tracking error is related to the primary market model of an ETF.
Let's begin with a quick reminder of the basics of the creation/redemption process for an exchange-traded fund. Only an Authorised Participant ("AP") is able to create/redeem ETFs directly with the ETF provider, either by paying cash or by delivering the underlying securities of the ETF, depending on what the ETF provider allows. An order placed by an AP with the issuer could be the result of client order flow or the desire of the AP to exploit an arbitrage opportunity. The creation/redemption process is the key for an efficient market, since market makers will arbitrage away any inefficiencies in the pricing of the ETF. They are able to do that because they can always create or redeem additional shares.
The ETF creation/redemption process, which takes place overnight, enables market makers to provide liquidity intraday. But since it is not possible to create ETF units during the day, the market maker will require a risk premium to cover the risk of holding a position in an ETF. When a trade is done intraday a market maker will immediately hedge his exposure, and this hedge will be unwound at the end of the day when the NAV trade takes place. After this, whatever remains of the risk premium will be the reward left for the market maker. If the market maker makes a mistake in calculating the risk premium, this could result in a loss on the trade.
An advantage for investors of NAV trading is transparency: the investor will know exactly how much he pays on top of the NAV (typically in the form of a basis point spread, which varies from fund to fund). The NAV always corresponds with the value of the underlying stocks, while this might not be the case with the price quoted for the ETF during the day.
In order to describe funds' creation/redemption models we first need to differentiate between the two most commonly used ETF structures: physically replicated ETFs and swap-based ETFs.
- Physically replicated ETFs hold the securities in the index underlying the ETF. This does not necessarily mean that all constituents are held in the fund; sometimes an optimisation approach is used, where not all securities are purchased but only an optimised subset of them. For example, some emerging markets have holding limits on stocks or bonds that prohibit a fund from buying those instruments. Another example is where the ETF issuer decides to make a trade-off between tracking error and tradeability of the fund. The more constituents of a fund, the more expensive it can be to create/redeem.
- Swap-based ETFs can hold a form of collateral that might be unrelated to the ETF basket. Through the swap, the performance of the index (minus a fee) tracked by the ETF is exchanged for the performance of the basket.
There are two main forms of ETF creation/redemption:
- Cash creations. A swap-based ETF can only be created when cash is delivered to the ETF provider, which in its turn will deliver the ETF units to the market maker.
- In specie creations involve the actual delivery of a basket of financial instruments, as specified by the ETF issuer. Physically replicated ETFs can normally be created both in specie and via an exchange for cash, depending on what the investor prefers. In most cases, in specie creations are cheaper for the issuer than cash creations. However, for the market maker the costs of creating in specie are not necessarily lower as a result of the need to trade in the underlying securities.
There are several factors that need to be considered when deciding to do a creation or redemption:
- The (minimum) creation size. Pretty much all ETFs have a minimum creation size which can be a minimum value or a minimum block size (the number of ETF units that have to be created).
- The creation fee. The fee can be quoted in cents or in basis points (bps). Fees can be fixed or variable and can include a fixed admin fee. The fees can decrease with size for some ETFs. In particular, fixed income ETFs can have variable fees and it is also not uncommon for them to have negative fees (many fixed income indices are computed using the bid prices of the bonds, while bond ETF prices trade somewhere between the average bid and average offer prices on all the bonds in the underlying index, with this difference causing the negative fees). An important component of ETF creation fees relates to stamp duty or other transaction-based taxes in the underlying securities (such taxes are payable in the UK, France, and at the time of writing this article, soon Germany, Italy, and Spain as well).
- The cut-off time. For many ETFs cut-off time to create/redeem is in the afternoon. After this time no creations or redemptions are accepted by the issuer. The later the cut-off time, the better for the market maker. If the cut-off is early and a trade is done after that, the position on the book has to be kept for another day. An interesting feature in the creation/redemption process of gold ETFs and ETCs is that NAV trades can take place at the AM and PM gold fixings, effectively meaning that there are two cut-off times during the day. The PM fixing is the most common, but with some ETFs it is possible to create with the AM gold fixing as well.
- The number of APs. Some products can only be created through one AP; others can have four or five or even more. When there are more APs it is easier for investors to have access to creations and redemptions and this will therefore increase liquidity in the fund.
LIQUIDITY IN THE SECONDARY MARKET
The question here is whether the aforementioned factors influence the spread on screen and/or the quoted size of the ETF. As a measure to describe the influence of market maker behaviour on ETFs, two benchmarks will be used:
- Tracking error measures the volatility of daily tracking differences between an ETF and its benchmark.
- Tracking difference is the absolute difference in returns between an ETF and its benchmark.
Both tracking error and tracking difference can be caused by multiple factors, for example:
- An incomplete basket is used instead of full replication.
- Management and swap fees.
- An execution of the basket before the close where it is difficult to trade exactly at the NAV price. The execution can lead to a better or worse result compared with the real fair value of the ETF, which is commonly referred to as slippage. How these costs are passed on to investors depends on whether the NAV is guaranteed (the NAV is not always guaranteed for every ETF; in particular non-guaranteed NAVs may be found in emerging markets). When the NAV is guaranteed and the executed price of the basket is higher than NAV, this cost will be incorporated in the fund, which will hurt all existing holders of the fund. If the NAV is not guaranteed the extra cost will only apply to the investor that is doing the creation.
What is the influence of the minimum creation size? When the minimum basket size is relatively large it will increase the on-screen bid-offer spread of the ETF in the secondary market. This is because the market maker will have to keep a trading position on the book for a longer time if the position is not large enough to trigger a creation or redemption in the ETF, and as a consequence the risk premium and therefore spread will increase. Market makers prefer not to have large positions due to the funding and management fees they have to pay on their inventory, and the correlation risk they are exposed to via the hedge. A high minimum basket size also potentially increases the short-term tracking error of ETFs because it is more difficult to arbitrage away deviations from the NAV. So smaller basket sizes will increase liquidity and decrease tracking error but they are less efficient for the fund. This is mainly the case for physically replicated funds since they will have to buy the underlying securities after a creation.
From a market maker's perspective other factors also have to be taken into consideration. The spread quoted by the market maker in the ETF secondary market, for example on exchanges' order books, represents the cost of the hedge plus the cost of creating/redeeming, the cost of holding the position, trading cost and a risk premium for the market maker. The risk premium will depend on the risk appetite of the market maker and in some cases on correlation risk. If a trade is done when some parts of the underlying market are not open, the market maker will have to deploy a proxy hedge, which will increase the risk premium required.
An example of an ETF where a proxy hedge is necessary is the db x-trackers FTSE Vietnam, the only Vietnam tracker in Europe. Limited foreign investment is possible in Vietnam, which makes it hard to acquire the underlying stocks; therefore from the moment a trade is done intraday till the time of the NAV trade a proxy hedge has to be used. The market maker has the risk that this hedge has a different performance to the ETF and therefore an additional risk premium is visible on top of the creation/redemption cost implicit in the spread. In the case of the db Vietnam, the cost for creating/redeeming is 35 bps for end-clients. As can be seen in Figure 2, the secondary market bid-offer spread in the ETF is wider than this to make up for the correlation risk of the hedge.
During the intensifying Eurozone debt crisis of summer 2011 the iShares Corporate Bond (IBCX LN) also gave a good example of the interaction between creation costs and the secondary market bid-offer spread in the ETF. The spreads of the underlying bonds widened significantly, which caused the creation and redemption fees to rise (see Figure 3). Higher ETF creation fees do not necessarily impact funds' secondary market spreads dramatically but during the crisis the volumes traded in corporate bond ETFs also decreased. This meant that market makers were unable to adjust their inventory positions via exchange-based trading in the ETF and had to use the primary market, incurring creation and redemption fees (these fees were then passed on to investors via the secondary market quotes in the ETF). When buy and sell orders in the market are balanced, a market maker does not need to go through the creation process. In an unbalanced market APs have to use the primary market. Market makers would have been able to quote tighter ETF spreads in more active two-way markets.
There are two types of cost that can influence the creation/redemption fees charged by the issuer:
- Commissions and settlement costs. Paid on buying/selling the instruments in the basket.
- Taxes and/or stamp duties. Paid on instruments from countries like the UK, France, Hong Kong, Taiwan, Korea, etc.
Whether an ETF is physically replicated or swap based matters when taxes on buying certain foreign stocks are involved. The stamp duty in the UK, for example, causes a big difference between the creation costs of physical and swap-based ETFs. The iShares FTSE 100 ETF (ISF LN) has a creation cost of 51 bps due to the stamp duty of 50 bps that is incurred by the ETF issuer when buying the UK-listed shares underlying the index, while the Lyxor FTSE 100 (L100 LN) has a 1bp creation cost. In Figure 5 the deviations in the closing prices, by reference to the funds' net asset values, are plotted for both ETFs. It can be seen that the effect of stamp duty is to push the iShares FTSE 100 to trade at a premium to NAV, typically in a range between -10bps and +50 bps. The Lyxor FTSE 100 trades closer to NAV since it is swap-based and the issuer does not therefore need to transact directly in the underlying stocks of the index, thereby avoiding stamp duty.
However, despite being the most expensive ETF of the two, ISF LN has the lowest spread and highest average value traded when compared to its competitors. Figure 4 shows that it has an average spread of 4 bps while the ETFs of Lyxor and db x-trackers have a spread around 10 bps. Furthermore the average daily value traded is more than ten times higher for the iShares ETF.
This example shows that in the end a higher tax does not necessarily matter for the on-screen liquidity of the ETF when there is high volume. As a consequence of high volume the ETF's bid-offer spread will be relatively tight; market makers have a greater incentive to quote tighter spreads because of the higher volume. On the other hand, an ETF affected by stamp duty will trade in a wider range around NAV and as an investor you have to watch out that you don't buy ISF at the top of its range.
Another example of how stamp duty has a significant impact is given by ETFs tracking the MSCI Europe index. Again we use iShares as the physical ETF (IMEU LN), which is compared here with the Lyxor MSCI Europe ETF (MEU FP), which is swap-based. The iShares ETF is more expensive to create since it has to purchase stocks in the UK to replicate the relevant portion of the underlying index, thereby incurring stamp duty.
In Figure 6 it can be seen that the iShares MSCI Europe trades at a premium of around 10 bps by comparison with the Lyxor MSCI Europe. Also, the average spread of IMEU is higher than that of similar MSCI Europe ETFs. We found an average spread of 15 bps, while the swap-based trackers of Source, Lyxor and db x-trackers have slightly tighter spreads. Nevertheless, the iShares fund has the highest assets under management of all MSCI Europe ETFs.
An example of a less liquid physical ETF whose secondary market quotation reflects stamp duty is the Credit Suisse UK Small Cap ETF (CUKS LN). The creation costs are slightly above 50 bps due to the stamp duty. Figures 8 and 9 show that the secondary market bid-offer spread is around 60 bps and the premium moves between -0.1% and 0.6%. It should be noted that the premium/discount moves are caused by the way in which physical ETFs invest in shares and are across all issuers, not just the issuers mentioned in these examples.
LIQUID VERSUS ILLIQUID ETFS
The impact of the creation/redemption terms on the liquidity of ETFs also depends on the existing liquidity of those ETFs. When an ETF is very liquid because of natural client flow it will not be necessary to create or redeem every position at NAV, since the market makers can trade out of their positions on screen during the day.
This is the case especially in the US, where in many cases the ETF is more liquid than the constituents of the underlying index. An example of this effect can be seen with IWM US, the iShares Russell 2000. The ETF trades at a 1 penny spread, which is equivalent to a 1.28 bps average spread in the underlying index stocks. However, it would be impossible to acquire all 2,000 stocks for such a tight spread. The reason this ETF can trade at such a tight spread is because the natural client flow is so large that the cost of creations is no longer important and for most orders there is enough flow to trade against. In other words, the market absorbs most orders naturally and only when there is a strong trend will there be a slight premium or discount.
Another example shows that for ETFs with a lot of client flow, the ability to create or redeem easily does not necessarily make a big difference for the secondary market spread of the fund. The ZKB Gold ETF, one of the largest gold trackers in Europe in terms of assets under management, is expensive and difficult to create yet still very liquid on screen, as can be seen in Figure 10.
The main listing, ZGLD on SWX, has an average spread of 8 bps even though it is easier to create PHAU and GBS, which have spreads of 11 and 5 bps, respectively. Where the high fees have an effect is in the higher tracking error, since it is harder to arbitrage away deviations from fair value when a relatively large order comes in. In Figure 11 and 12 the deviations in bps from the gold spot price can be seen for ZGLD and PHAU. By looking at the deviations from the spot price in a week in June this year it is evident that ZGLD trades in a wider range around NAV than PHAU LN. It can be argued therefore that PHAU is more liquid because of its more efficient creation/redemption process and lower fees compared to ZGLD, even though the average spread is slightly higher for PHAU.
When Creations Are Halted
There have been some instances in the past few years where issuers were forced to halt creations for specific funds. In 2011, for instance, this was the case for the Market Vectors Egypt ETF due to the turmoil in Egypt at that time. Market Vectors had to close the fund for creations because the underlying market was closed. It was however still possible to trade the ETF in the secondary market. In the month that the Egyptian stock exchange was closed, the underlying index hardly moved (only some ADRs in the underlying baskets were still trading). The ETF, on the other hand, started trading at a premium since there was much buying activity in the secondary market causing the tracking error that can be seen in Figure 13.
Another example where creations were halted this year, although this happened in an exchange-traded note rather than fund, was in the VelocityShares Daily 2x Vix Short-Term ETN. Credit Suisse, the ETN issuer, decided to halt creations due to internal risk limits being reached. The demand for the ETN stayed high, which caused it to trade at a very large premium. Figure 14 shows that the premium eventually ceased at the end of March, right before creations were permitted again. These two examples show the importance of the creation/redemption process. It could lead to a large(r) tracking error when the process is disturbed.
We have seen that the creation/redemption process has a significant impact on the liquidity of ETFs. Costs and the overall efficiency of the process have less impact on the spread when ETFs have natural client flow, i.e. there is sufficient volume in the secondary market for market makers to trade out of a position.
Furthermore, we saw that the creation/redemption process of an ETF (in-specie or cash) does make a significant difference when costs are high. Physical ETFs will typically trade in a wider range around NAV by comparison with swap-based ETFs, but physical ETFs can still have a tighter spread on screen and higher value traded. When an ETF is illiquid the creation/redemption process will be more visible in the spread on screen, since a market maker will have to use the primary market to trade out of the position.
Finally, whilst high creation/redemption costs are definitely a potential concern for investors, not being able to create at all can have even more significant effects on the liquidity and tracking error of ETFs. In those cases, ETFs could potentially trade at a significant premium or discount far from their fair value. This shows that having the option to create/redeem is crucial.