Liquidity in ETFs: What really matters

Transcription

1

2 Liquidity in ETFs: What really matters Laurent DEVILLE, Affiliate Professor, EDHEC Business School This research has been carried out with the support of Amundi ETF

3 ETFs and liquidity ETF markets are designed so as to ensure stable and efficient provision of liquidity creation/redemption process specific market-making rules The industry often claims that the real liquidity of an ETF is the liquidity of its underlying stock basket liquidity of blue-chips ETFs is generally higher than the liquidity of ETFs replicating small-cap indexes. Empirical observations raise reasonable doubts about this view ETFs tracking a given index display different levels of liquidity liquidity of ETFs is generally impacted by the introduction of competitors

4 ETFs and liquidity (cont d) The European equity ETFs markets present features deserving specific investigation no consistent report of OTC trades over all markets and/or ETFs how to measure ETF liquidity? high degree of market fragmentation does the ETF microstructure impacts ETF liquidity? widespread use of synthetic replication techniques even for highly liquid indexes is synthetic replication associated with higher/lower liquidity? does liquidity provision follow the same model for synthetic and physical ETFs?

5 Motivation Given the large development of ETFs market (in terms of volume, benchmark type and uses), it is crucial to understand and estimate what drives liquidity in ETF markets Existing studies on ETFs liquidity essentially focus on US ETFs liquidity of ETFs is higher than the liquidity of the underlying stocks portfolio (Hedge and Mc Dermott, 2004) new ETFs tend to track narrower and less liquid indexes than mutual funds (Guedj and Huang, 2009) ETFs may serve as a channel for the propagation of exogenous liquidity shocks to underlying stocks (Ben David et al., 2011) We provide empirical evidence of the determinants of ETF spreads for a large sample of European equity ETF markets from their inception in 2000 to the end of 2011

6 Main results The liquidity of an ETF does build on the liquidity of its benchmark index Index liquidity is not the only significant factor the ETF AUM and trading volume are also important market microstructure plays a role through competition and fragmentation Physical and synthetic ETFs exhibit different behaviors in terms of liquidity synthetic spreads are poorly explained by our model the liquidity of synthetic ETFs is not significantly affected by the liquidity and size of their benchmark the crisis affected spreads in opposite ways for the two types of ETFs

7 OUTLINE 1. Liquidity provision in ETF markets 2. Data and summary statistics 3. The determinants of ETF spreads 4. Does the replication technique matter? 5. Conclusions

8 1. LIQUIDITY PROVISION IN ETF MARKETS

9 Market making in ETFs One of the ETFs specificities is their open-ended structure shares can be created or redeemed against the basket at any time market makers can use this feature when facing large orders and/or inventories In the case of in-kind creations and redemptions ETFs Authorized Participants are likely to reflect in their spreads the costs and risks borne when trading the basket shares illiquidity on the stock market is likely to smear over the ETF liquidity

10 Synthetic ETFs liquidity In the case of synthetic ETFs the fund is invested in a stock basket that may be very different from the benchmark index replication is achieved through the use of a total return swap Creations and redemptions are generally done againts cash or futures instead of exchange of physical this may insulate the liquidity of ETFs from that of their benchmark index this may allow to provide liquidity even in the case of low liquidity benchmarks

11 Measuring index liquidity Index liquidity is calculated as the average of measures built at the stock level We work with low frequency liquidity measures calculated on a monthly basis with daily data (Goyenko and al., 2009) our first level of analysis is given by transaction costs proxies the relative quoted spread at close Roll (1984) effective spread estimator as liquidity providers may compete on prices but not on volumes we also compute a price impact proxy Amihud (2002) price impact measure

12 Measuring ETF liquidity Liquidity measures based on the asset returns are irrelevant in the case of ETFs ETF returns mainly reflect those of the benchmark index Roll estimator and Amihud ratio are indirect ways of capturing the tracking error of an ETF in particular, the Roll estimator of the effective spread mainly builds upon the benchmark index autocorrelation of returns (rather than on the ETF s own liquidity characteristics) Liquidity in ETFs is thus computed as the monthly average relative quoted spread observed on the primary listing market

13 2. DATA AND SUMMARY STATISTICS

14 Sample We use a sample of 346 equity ETFs listed on 19 European markets totalizing 1,122 listings (different exchanges and/or currency issued by 27 issuers replicating 204 indices for which we know the composition (thus excluding MSCI indexes) Without MSCI indexes, our sample is representative of the European market for equity ETFs 43% of ETFs listed in Europe, 64% of AuM proportions of physical and synthetic ETFs are similar in our sample and in the whole market Both replication types seem to offer some advantages most issuers, though specialized, offer both types of ETFs synthetic ETFs dominate in emerging market or sector indices physical ETFs dominate in developed/regional indices

15 Data ETF level data characteristics: benchmark, replication technique, type of exposure, issuer, primary market and listing date, other listings (BlackRock) market data: daily prices, spreads, volumes, NAV and shares outstanding (Bloomberg) Index level data daily composition, weights and closing values (Bloomberg) Stock level data market data: daily prices, spreads and volumes

16 Sample description Heterogeneous market ETFs track a wide range of indices in terms of size, liquidity and number of constituent stocks Competition in equity ETFs an average of but 3.8 ETFs replicate the same benchmark some indices attract up to 21 competing ETFs some benchmarks are in competition with other indexes Cross-listing most ETFs are cross-listed (78% for stocks and 68% for swaps) on average, an ETF is listed on 3 different exchanges some ETFs count up to 12 listings the primary listing is generally dominant cross-listed ETFs present a significant degree of fragmentation

17 Evolution of liquidity measures

18 Univariate results The link between ETF liquidity and index liquidity shows up when we split the sample into categories a priori reflective of the difficulty of replicating the index spreads are significantly tighter for developed markets ETFs spreads are significantly wider for emerging markets and global ETFs Physical ETFs exhibit lower average closing spreads than synthetic ETFs even though the closing spreads of their underlying baskets are higher than those of synthetic ETFs baskets which could be explained by significantly higher trading volumes

19 3. GENERAL DETERMINANTS OF THE SPREAD

20 Multivariate analysis In a multivariate setting, we control for heterogeneity in index selection trading environment and global market conditions Explanatory variables Benchmark index liquidity measure Basket closing spread Conventional determinants of spreads in equity markets ETF trading volume Index volatility Variables specific to ETFs Replication technique Sector index

21 Multivariate analysis (cont d) Other explanatory variables Microstructure variables fragmentation of order flow (Herfindhal) number of listings number of ETFs competing on the same index benchmark Other factors likely to affect the provision of liquidity in ETFs and control variables index size asynchronicity of trading between ETF shares and the benchmark portfolio currency risk availability of a futures contract on benchmark index crisis

22 Determinants of ETF spreads Model 1 Model 2 Model 3 Model 4 Intercept Basket quoted spread ETF volume Index volatility Replication SectorETF Fragmentation Cross listing Competition Index.MV Stock avg.mv Crisis Asynchronicity Currency risk Futures Adj. R % 16.80% 17.61% 18.42%

23 Determinants of ETF spreads To summarize our results the liquidity of the benchmark index is always significant spreads are affected by ETF trading volume and benchmark index volatility competition and fragmentation significantly impact ETF spreads crisis impacted spreads beyond the effect of volatility surprisingly, other risk factors do not impact ETF spreads index style and replication type show up in regressions The fit remains modest possibly due to high heterogeneity in the sample we investigate differences between physical and synthetic ETFs in a sub-sample analysis

24 4. DOES THE REPLICATION TECHNIQUE MATTER?

25 Physical vs synthetic ETFs Stocks Swaps Coeff. t-stat Coeff. t-stat Intercept Basket quoted spread ETF volume Index volatility SectorETF Fragmentation Cross listing Competition Index.MV Stock avg.mv Crisis Asynchronicity Currency Futures Adj. R² 47.07% 13.90%

26 Physical vs synthetic ETFs The relationship between the ETF liquidity and the benchmark liquidity seems to loosen for synthetic ETFs swap-based ETFs spreads are not significantly affected by the benchmark s size and liquidity the quality of fit is poor for swap-based ETFs Financial crisis is associated with tightened spreads for physical ETFs a flight from synthetic ETFs to physical ETFs may explain this effect Fragmentation is associated with opposite effects for physical ETFs moderate fragmentation is associated with higher liquidity for synthetic ETFs the higher degree of fragmentation seems to deteriorate market liquidity

27 5. CONCLUSION

28 Conclusion The liquidity of the ETF underlying stock baskets is significantly and positively correlated to ETF spreads only explains a modest 13% of the variance of ETF spreads Other determinants of ETF spreads include conventional determinants of spreads in equity markets (trading volume, volatility) variables specific to ETFs (index style, replication type) variables related to the microstructure of the ETF market (competition, fragmentation)

29 Conclusion Synthetic ETFs exhibit tighter spreads in univariate analysis but the replication type loses significance in multivariate analysis when competition is accounted for it is not the replication style per se but rather the increased competition that drives spreads Limits closing spreads are a robust but noisy measure of liquidity intraday measures of liquidity specifically built for ETFs still need to be developed the impacts of the different types of ETFs on index constituent stocks need to be investigated