european sovereign bond etfs – tracking errors and sovereign debt crisis

42
EUROPEAN SOVEREIGN BOND ETFs – TRACKING ERRORS AND SOVEREIGN DEBT CRISIS Branko Urošević, Faculty of Economics, University of Belgrade and National Bank of Serbia First Moscow Finance Conference November 2011

Upload: alayna

Post on 11-Jan-2016

53 views

Category:

Documents


1 download

DESCRIPTION

EUROPEAN SOVEREIGN BOND ETFs – TRACKING ERRORS AND SOVEREIGN DEBT CRISIS Branko Urošević , Faculty of Economics, University of Belgrade and National Bank of Serbia First Moscow Finance Conference November 2011. Outline. Background, motivation, contributions Literature and hypotheses - PowerPoint PPT Presentation

TRANSCRIPT

EUROPEAN SOVEREIGN BOND ETFs – TRACKING ERRORS AND SOVEREIGN DEBT

CRISIS

Branko Urošević, Faculty of Economics, University of Belgrade and National Bank of

Serbia

First Moscow Finance ConferenceNovember 2011

Outline

• Background, motivation, contributions• Literature and hypotheses• Data and methodology• Results• Conclusions

Motivation

• Market for sovereign bonds of the Euro zone in the forefront of interest of investors, politicians, …

• Fundamental shift in perception: from virtually riskless to much more akin to corporates (some of them with a junk bond status)

• Diverging performance across countries• EFTs: new, liquid and relatively transparent way to

get exposure to that market• While many asset classes are severely impacted by

the crisis, ETFs steadily grow in importance

Literature and motivation

• Amenc and Golz (2009) – EU Survey - ETFs play an increasingly important role in providing exposure to various asset classes

• Rompotis (2008; German equity index ETFs); Milionas and Rompotis (2006; Swiss equity index ETFs); Gallagher and Segara (2005; Australian equity index ETFs); Blitz et al. (2010; EU equity index ETFs);

• Drenovak and Urosevic (2010), Houweling (2011) analyse corporate (European and US) and sovereign (US) bond ETFs. Performance measured using correlation-based tracking error.

• Alexander (1999); Alexander and Dimitriu (2004):– Two time series can be cointegrated even if static correlations between

them are low

• Important to examine tracking performance based on both correlation and cointegration, especially for passive investment strategies.

Literature and motivation• Norden and Weber (2009)

– CDS market reflects information earlier than bond market

– Bond spreads adjust to CDS spreads.• The dynamics of spread volatility are driven by those of

spread levels (Nomura, 2011). They may influence tracking performance after the crisis commenced.

• Composition of ETFs may also play an important role now (and did not play much of a role before)

Contributions

• Study tracking performance of the most important families of ETF funds that track indexes of euro zone sovereign bonds using a novel data set

• Compare performance using 4 types of tracking errors: active return (TE1), two short-term correlation based measures (TE2 and TE3) and a cointegration-based tracking error measure (TE4)

• Synthetic versus physical replication• Preliminary study of determinants of tracking errors

including how crisis influenced them

Hypothesis

• H1: Euro zone sovereign ETFs underperform their underlying bond indexes, regardless of the replication method.

• H2: ETFs’ tracking errors are positively associated with the volatility of target indices.

• H3: ETFs’ tracking errors are positively associated with the maturity of underlying indexes.

Hypotheses

• H4: Tracking performance of EU sovereign debt ETFs deteriorated during the post Lehman period.

• H5: CDS spreads for index constituents are an important determinant of tracking performance of EU sovereign debt ETFs in a post-Lehman period

• H6: Synthetic ETFs perform better measured by TE1 and TE4, and worse measured by correlation based measures (TE2 and TE3), compared to full replication physically-based ETFs.

Sample and data

• Sample: 31 Euro sovereign bond indices ETFs, January 2007-December 2010 (captures more than 90% of the market): – iShares (track Barclays term, Markit iBoxx Liquid Capped,

eb.rexx German Government). Physical replication. – db x-trackers (track Markit iBoxx Sov). Synthetic

replication– Lyxor (track EuroMTS). Synthetic replication

• All indices tracked by sample ETFs are total return indices (all interest payments are reinvested); all maturities considered

• Daily data on: Net Asset Values (NAV), Weights, CDS, and Bid-ask. Data on NAV from Frankfurt (for consistency)

•  Sources: Bloomberg, index providers, Morningstar

Aggregate country exposure of government bond indices tracked by sample ETFs, stratified by ETF providers

G I Fr Sp B Nd Gr P A Ir FiniSharesBarclays Term 1-3 32.6 42.1 11.1 14.2Barclays Term 3-5 32.2 23.7 30.8 13.3Barclays Term 5-7 48.0 25.0 12.7 14.2Barclays Term 7-10 56.6 7.9 29.9 5.6Barclays Term 10-15 9.3 41.4 30.3 7.0 12.0Barclays Term 15-30 29.4 34.8 21.7 9.7 4.4iBoxx€ Liq Sov Cap 1.5-2.5 20.3 20.0 20.2 8.0 4.4 9.5 13.9 3.8iBoxx€ Liq Sov Cap 2.5-5.5 20.3 20.0 20.5 20.0 13.0 6.1

iBoxx€ Liq Sov Cap 5.5-10.5 20.5 20.3 20.3 19.9 13.1 5.9

iBoxx€ Liq Sov Cap 10.5+ 20.6 19.8 20.1 19.3 14.5 5.7

iBoxx€ Liq Sov Cap 1.5-10.5 20.4 20.2 20.3 19.8 11.9 7.4

eb.rexx 1.5-2.5 100.0eb.rexx 2.5-5.5 100.0eb.rexx 5.5-10.5 100.0eb.rexx 10.5+ 100.0eb.rexx DE 100.0Db x-trackersShort iBoxx € Sov 21.6 23.7 20.9 9.4 5.9 5.4 3.9 2.2 3.7 2.0 1.1iBoxx € Sov 21.6 23.7 20.9 9.4 5.9 5.4 3.9 2.2 3.7 2.0 1.1iBoxx € Sov 1–3 25.1 24.4 19.8 11.1 5.7 5.5 3.8 1.8 1.2 1.0 0.7iBoxx € Sov 3–5 24.4 17.2 19.8 11.0 7.2 5.4 4.2 2.3 4.3 2.0iBoxx € Sov 5–7 21.1 20.1 22.4 6.8 8.6 6.0 3.9 2.5 5.7 1.7 1.2iBoxx € Sov 7–10 19.3 25.3 21.2 8.3 4.3 5.8 4.5 2.5 4.1 3.1 1.7iBoxx € Sov 10–15 4.3 26.0 23.4 7.2 5.8 6.6 5.8 4.8 7.2 7.6 1.3iBoxx € Sov 15+ 23.9 30.1 21.3 9.4 4.6 3.7 2.4 0.9 3.4 0.5iBoxx € Sov 25+ 22.0 20.7 31.0 11.6 4.9 4.4 2.3 3.0LyxorEuroMTS 1-3Y 23.7 23.4 21.8 11.5 6.2 4.3 4.6 0.9 1.3 1.84 0.7EuroMTS 3-5Y 24.1 18.0 21.2 9.0 7.6 6.0 3.7 2.6 4.7 1.34 1.8EuroMTS 5-7Y 20.8 16.9 23.6 6.1 8.9 6.2 5.5 2.8 6.2 1.91 1.2EuroMTS 7-10Y 18.6 24.2 19.6 9.1 4.1 5.6 5.3 3.3 3.9 4.36 2.1EuroMTS 10-15Y 2.1 31.0 23.2 5.2 6.8 8.5 5.3 4.6 8.4 5.0EuroMTS 15Y+ 22.3 29.0 21.3 11.0 5.3 3.7 2.3 0.9 3.6 0.8

Summary statistic (average values)AUM €(000)

TER (bps) Dur (yrs) Vol (%) Corr M-star index

iShares -Barclays

189,594 20 6.77 4.42 0.99 4.25

iShares -iBoxx

114,943 16 5.80 4.63 0.97 3.40

iShares- eb.rexx

689,901 15 5.70 4.55 0.98 4.80

db x-trackers

321,861 15 6.79 5.56 0.98 3.40

Lyxor 543,808 16.5 6.77 5.12 0.51 3.83

Methodology – Standard methods

TE2

P2

B2 2

P

B

P,B

Methodology - OLS (TE3 = standard error of the regression)

Methodology - Cointegration framework (TE4 = autocorrelaton of the residual from the

cointegrating regression)

Here, NAVp and NAVb are daily log NAV values for ETFs and underlying indices, z is the cointegrating vector NAVpt - αNAVbt, Δ is the first difference operator, and the lags, lengths and coefficients are determined by OLS regression.

Tracking error and replication methods

• Previous studies did not examine tracking errors in the context of different replication methods adopted by ETFs.

• Replication methods should be taken into consideration when selecting an appropriate measure of tracking errors.

• A full physical replication of indexes with strict inclusion criteria would, for example, lead to a very high correlation of ETF and tracking index returns and, therefore, low TE2 and TE3.

• This can come at the expense of long term co-movement (TE4, for example).

Ranking Based on TE1• Consistent with our hypothesis 1, sample ETFs underperformed their

respective indexes during the sample period• The average underperformance for the total sample is similar to the

underperformance reported in previous studies on bond ETFs (Drenovak and Urosevic, 2010; Houweling, 2011).

• The average annual TE1, however, varies significantly from 0.53 (LyxorMTS+15) bps to 27.38 bps (iShares-BarclaysTerm10-15).

• Overall, the iShares–BarclaysTerm family exhibits largest whilst Lyxor ETFs exhibit the smallest average TE1s.

• Lyxor ETFs actually over-performed relevant indices in 2007 and 2009. • db x-trackers exhibited the most consistent performance during the

sample period. • While iShares-Barclays family continues to exhibit similar TE1s, the rest of

the ETF families exhibited a sharp drop in the TE1 levels in 2009. • In 2010, iShares-eb.rexx and Lyxor’s were the only ETFs with a sharp

increase in average TE1.

Performance and sample TE1This table represents ETFs annual returns (%) and TE1 presented in basis points [in brackets]. Positive basis points indicate undeperformance of the

ETFs.  

 

  Year iShares db x-track LyxorBarclays T. iBoxx

L.S.C.eb.rexx

Mean

2007 1.10 [23] 1.35 [8] 1.51 [5] 5.00 [16] 2.29 [-7]2008 9.99 [28] 8.99 [24] 12.35 [32] 8.17 [17] 9.13 [17]2009 4.39 [28] 3.64 [9] 1.91 [7] 2.79 [15] 4.46 [-3]2010 12.45 [20] 7.88 [16] 14.58 [68] 7.80 [15] 8.37 [33]

2007-10 6.58 [27] 4.95 [13] 6.10 [19] 5.37 [16] 5.86 [10]

Median

2007 2.15 [23] 1.89 [7] 2.66 [9] 5.03 [16] 2.97 [10]2008 10.37 [29] 9.77 [24] 11.52 [36] 8.97 [17] 9.43 [17]2009 4.21[27] 3.73 [8] 2.78 [10] 4.19 [15] 4.81 [0]2010 14.08 [21] 6.53 [14] 12.03 [63] 9.37 [16] 8.12 [31]

2007-10 6.51 [26] 5.22[13] 6.29[16] 6.62 [16] 5.91[12]

Min

2007 -3.42 [19] -3.29 [3] -3.59 [-11] 3.42 [15] -0.2 [99]2008 6.42 [24] 6.48 [18] 7.13 [15] -3.54 [12] 6.48 [15]2009 2.23 [21] 1.19 [6] -1.59 [-12] -3.21 [13] 2.13 [-7]2010 4.14 [13] 1.87 [10] 5.42 [29] -7.77[12] 3.36 [4]

Max

2007 3.52 [26] 3.42 [12] 3.51 [13] 6.44 [16] 3.55 [20]2008 12.8 [31] 11.53 [27] 18.44 [47] 13.81

[19]10.76

[18]2009 7.39 [39] 5.34 [11] 3.47 [15] 5.29 [16] 5.40 [17]2010 16.38 [25] 17.21 [30] 28.02

[127]14.37

[18]13.51

[63]

Ranking of funds based on TE2• All sample ETFs have statistically significant average (mean) tracking errors at the

1% level of significance. • This result is robust to the use of weekly or monthly instead of daily price series.• Overall, iShares funds which replicate Barclays Term indices exhibit the smallest

while Lyxor ETFs exhibit the largest values of TE2. • TE2 (based on monthly returns) range from 1.1 basis points (iShares-

BarclaysTerm1-3) to 43.83 (LyxorMTS10-15). • The results also suggest that ETFs tracking higher maturity indices have typically

higher levels of tracking error. This is the case for all sample ETFs except for db iBoxx Sov 5-7.

• There are also some differences in the way TE2s changed during the sample period. iBoxx Liquidity and eb.rexx, for example, exhibited the highest TE2 in 2009.

• This is consistent with high weightings for Greece (in iBoxx liquidity) and Germany (in eb.rexx), the two countries with extremely volatile interest rates during 2009.

• Barclays, db, and Lyxor indices, however, exhibited the highest TE2 during 2008.• Lyxor funds have particularly high TE2.

Sample TE2 This table presents results for average (mean) 3 month TE2 for respective ETFs, based on monthly and daily NAV series. N is the number of bonds in the portfolio – calculated by authors based on the data from ETFs’ prospectuses as in May 2010. P-values for one sample T-test for mean=0 vs. mean#0 in brackets. Unreported results for one sample Wilcoxon test for median=0 vs. median#0 are economically and statistically consistent with the reported results for the T-test.

AVERAGE 3-MONTH TE2 (IN BPS)N TE2 (monthly) TE2 (daily)

    2007-10 2007-10 2007 2008 2009 2010

iShares              

Barclays Term 1-3 10 1.23(0.000) 0.77(0.000)

0.69 1.37 0.44 0.2

Barclays Term 3-5 15 1.96(0.000) 1.53(0.000)

1.28 3.02 0.59 0.54

Barclays Term 5-7 9 2.38(0.000) 2.29(0.000)

 -  - 3.48 0.83

Barclays Term 7-10

13 2.80(0.000) 2.33(0.000)

2.08 5.08 0.36 0.76

Barclays Term 10-15

14 3.32(0.000) 2.95(0.000)

 -  - 4.43 1.12

Barclays Term 15-30

30 4.69(0.000) 4.08(0.000)

4.14 8.77 0.62 0.68

Average for Barclays

15 

2.73(0.000) 2.32(0.000)

2.05(0.000)

4.56(0.000)

1.65(0.000)

0.69(0.000)

AVERAGE 3-MONTH TE2 (IN BPS)N TE2

(monthly)TE2 (daily)

    2007-10 2007-10 2007 2008 2009 2010iBoxx€ Liq Sov Cap 1.5-2.5

15 1.97(0.000) 1.87(0.000) 1.84 2.28 1.55 1.72

iBoxx€ Liq Sov Cap 2.5-5.5

15 2.63(0.000) 2.52(0.000) 1.81 2.81 2.93 2.11

iBoxx€ Liq Sov Cap 5.5-10.5

15 7.05(0.000) 5.30(0.000) 7.53 4.26 6.02 1.75

iBoxx€ Liq Sov Cap 10.5+

15 17.46(0.000) 12.80(0.000)

19.65 7.9 16.61 2.1

iBoxx€ Liq Sov Cap 1.5-10.5

25 6.03(0.000) 4.65(0.000) 6.20 3.96 5.29 1.72

Average for iBoxx

 17 7.03(0.000) 5.43(0.000)

7.40(0.000)

4.24(0.000)

6.48(0.000)

1.88(0.000)

eb.rexx 1.5-2.5 6 1.77(0.000) 1.89(0.000) 1.55 2.33 1.92 1.33eb.rexx 2.5-5.5 12 3.15(0.000) 2.57(0.000) 1.74 2.63 3.56 1.47eb.rexx 5.5-10.5 10 6.54(0.000) 4.84(0.000) 1.68 4.11 9.18 1.62eb.rexx 10.5+ 10 15.66(0.000) 11.78(0.00

0)2.23 14.23 20.09 2.38

eb.rexx DE 25 2.77(0.000) 2.90(0.000) 1.54 3.09 4.28 1.44Average for

eb.rexx13 5.98(0.000) 4.79(0.000

)1.75(0.0

00)5.28(0.0

00)7.8(0.00

0)1.65(0.0

00)

AVERAGE 3-MONTH TE2 (IN BPS)

TE2 (monthly) TE2 (daily)

    2007-10 2007-10 2007 2008 2009

db x-trackersshort iBoxx 25

24.54(0.000) 4.38(0.000

)-  6.81 4.81 0.56

iBoxx € Sov 252

3.49(0.000) 3.32(0.000)

3.69 4.85 2.72 0.48

iBoxx € Sov 1–3 60 1.57(0.000) 1.53(0.000)

2.52 2.05 1.01 0.6

iBoxx € Sov 3–5 48 3.03(0.000) 3.07(0.000)

4.06 4.67 2.03 0.65

iBoxx € Sov 5–7 33 5.34(0.000) 5.03(0.000)

5.26 9.03 2.71 0.39

iBoxx € Sov 7–10 47 4.10(0.000) 3.81(0.000)

3.89 6.2 2.63 0.55

iBoxx € Sov 10–15 24 4.55(0.000) 4.34(0.000)

3.88 6.74 3.55 0.51

iBoxx € Sov 15+ 44 6.33(0.000) 6.40(0.000)

5.45 7.69 7.56 0.79

iBoxx € Sov 25+ 18 11.74(0.000) 9.42(0.000)

5.90 13.86 9.2 1.3

Av. for db iBoxx  86 4.97(0.000) 4.59(0.000)

4.33(0.000)

6.88(0.000)

4.02(0.000)

0.65(0.000)

LyxorEuroMTS 1-3 20 10.08(0.000) 9.54(0.000

)6.97 12.76 9.07 7.37

EuroMTS 3-5 22 20.33(0.000) 19.34(0.000)

13.68 25.15 20.07 13.27

EuroMTS 5-7 20 28.61(0.000) 26.94(0.000)

20.60 32.61 28.39 19.82

EuroMTS 7-10 22 34.78(0.000) 32.65(0.000)

26.20 38.86 34.22 24.02

EuroMTS 10-15 18 44.40(0.000) 40.72(0.000)

31.52 49.46 43.12 29.6

EuroMTS 15+ 36 15.77(0.000) 12.87(0.000)

34.85 15.88 5.68 8.09

Average for Lyxor

23 25.66(0.000) 23.68(0.000)

22.30(0.000)

29.12(0.00)

23.42(0.00)

17.03(0.00)

Figure 2 Evolution patterns of sample TE2This figure presents the patterns of average three month TE2 (in bps) for the

sample ETFs targeting 10+ year maturities. TE2 based on daily returns.

0

10

20

30

40

50

60

70

1/1/2008 6/1/2008 11/1/2008 4/1/2009 9/1/2009 2/1/2010 7/1/2010 12/1/2010

iShares Bar 10-15 iShares eb.rexx 10.5+ iShares iBoxx 10.5+

Lyxor EuroMTS10-15 db iBoxx 10-15

Ranking based on TE3• The results presented in the following table suggest that only

3 out of 31 sample ETFs have alphas statistically different from zero.

• The estimations for all beta coefficients are statistically different from 1 for all but 3 sample funds.

• Lyxor family is a clear outlier with an average beta of 0.53 confirming that ETFs from this family depart from full replicating strategy.

• The coefficient beta is below 1 for all but two sample funds, indicating that the sample’s ETFs may be more conservative than their respective benchmarks.

Ranking based on TE3

• Regarding R2, the Lyxor family is again a clear outlier with an average of only 30%.

• For all other sample ETFs, ranges between 93% and 99%, indicating a very good regression fit.

• The iShares family exhibits the highest average of 98%.

• Reported values for TE3, measured by standard deviation of residuals, indicate the same ranking of sample funds as is it was case when we employed TE2 measure.

OLS Regression model for ETFs’ returns and TE3This table presents results of the OLS model for ETFs daily returns as dependent variable and daily returns of the underlying indices as explanatory variable. TE3 is standard error of regression (i.e. standard deviation of residuals) in basic points.

Alpha

T-test Beta T-test R2 TE3 N

(1) (2) (3) (4) (5) (6) (7) (8)iSharesBarclays Term 1-3 0.000

0-2.56** 1.00 -1.15 0.99 1.01 1037

Barclays Term 3-5 0.0000

-0.93 1.00 -0.55 0.982.32

1037

Barclays Term 5-7 0.0000

-0.22 0.98 -2.33** 0.964.23

439

Barclays Term 7-10 0.0000

-0.54 0.99 -2.24** 0.984.07

1037

Barclays Term 10-15 0.0000

-0.35 0.99 -1.33 0.984.84

439

Barclays Term 15-30 0.0000

-0.41 0.99 -2.37** 0.995.78

1037

Average 0.0000

-12.70***

0.99 -3.18** 0.98

Alpha T-test Beta T-test R2 TE3 N(1) (2) (3) (4) (5) (6) (7) (8)

iBoxxiBoxx€ Liq Sov Cap 1.5-2.5 0.000

00.20 0.96 -9.82*** 0.98

1.991037

iBoxx€ Liq Sov Cap 2.5-5.5 0.0000

-0.13 0.98 -4.80*** 0.972.91

1037

iBoxx€ Liq Sov Cap 5.5-10.5

0.0000

0.22 0.94 -8.83*** 0.956.64

1037

iBoxx€ Liq Sov Cap 10.5+ 0.0000

0.17 0.89 -11.50*** 0.8917.52

1037

iBoxx€ Liq Sov Cap 1.5-10.5

0.0000

0.19 0.94 -8.89** 0.965.58

1037

Average 0.0000 1.97 0.94 -4.10** 0.95Ebeb.rexx 1.5-2.5 0.000

00.10 0.96 -7.00*** 0.96

1.801037

eb.rexx 2.5-5.5 0.0000

0.00 0.96 -9.26*** 0.982.55

1037

eb.rexx 5.5-10.5 0.0000

0.10 0.96 -8.19*** 0.975.47

1037

eb.rexx 10.5+ 0.0000

0.17 0.94 -7.68*** 0.9316.88

1037

eb.rexx DE 0.0000

-0.03 0.98 -5.54*** 0.983.06

1037

Average 0.0000 1.29 0.96 -7.20*** 0.96

Alpha

T-test Beta T-test R2 TE3 N

(1) (2) (3) (4) (5) (6) (7) (8)db x-trackersshort iBoxx 0.00

00-0.27 0.98 -2.85*** 0.96

5.43687

iBoxx € Sov 0.0000

-0.24 0.99 -2.22** 0.974.64

937

iBoxx € Sov 1–3 0.0000

-0.45 0.98 -2.57** 0.962.04

934

iBoxx € Sov 3–5 0.0000

-0.18 0.98 -2.32** 0.954.22

934

iBoxx € Sov 5–7 0.0000

0.00 0.97 -3.24*** 0.936.79

931

iBoxx € Sov 7–10 0.0000

-0.22 0.99 -1.97** 0.975.49

931

iBoxx € Sov 10–15 0.0000

-0.21 0.99 -2.13** 0.976.35

929

iBoxx € Sov 15+ 0.0000

-0.10 0.99 -2.50** 0.979.87

929

iBoxx € Sov 25+ 0.0000

-0.07 0.98 -2.35** 0.9613.75

927

Average 0.0000 -6.37*** 0.98 -9.21*** 0.96LyxorEuroMTS 1-3 0.00

012.65** 0.44 -

18.98***

0.17

9.21

1037

EuroMTS 3-5 0.0001

1.72* 0.44 -18.95**

*

0.18

17.34

1037

EuroMTS 5-7 0.0001

1.36 0.44 -19.21**

*

0.18

23.76

1023

EuroMTS 7-10 0.0001

1.01 0.44 -19.34**

*

0.18

28.23

1023

EuroMTS 10-15 0.0001

0.62 0.43 -19.51**

*

0.18

35.58

1037

EuroMTS 15+ 0.0000

0.13 0.97 -3.75*** 0.9316.91

911

Average 0.0001 5.23*** 0.53 -5.38*** 0.30

Cointegration-based measure TE4

• Based on the results of the JT test for cointegration, we strongly reject the null hypothesis of no cointegration in all but five ETFs.

• Four out of the five ETFs are from the iShares–Barclays Term family (1-3, 3-5, 7-10, and 15-30). In contrast, the log likelihoods, JT test statistics, and DF test statistics indicate a very good model fit for iShares - iBoxx€LiqSovCap and Lyxor ETFs, followed by the iShares - eb.rexx and the db x-trackers funds.

• Based on TE4, Lyxor ETFs are the best and iShares – Barclays Term the worst performers. Overall, swap-based ETFs perform better than physically-based ETFs measured by TE4.

NL JT trace

Normalised coefficients

DF TE4. N

iShares ETF Index

Barclays Term 1-3 4 5.79 - - - - 1037

Barclays Term 3-5 4 12.41 - - - - 1037

Barclays Term 5-7 4 24.46** 1 -0.9609*** -30.033 0.47 439

Barclays Term 7-10 4 13.67 - - - - 1037

Barclays Term 10-15 4 16.23** 1 -0.9718*** -34.067 0.62 439

Barclays Term 15-30 4 14.94 - - - - 1037

Average (mean) 14.583 -32.050

0.55

NL JT trace

Normalised coefficients

DF TE4 N

iBoxxiBoxx€ Liq Sov Cap 1.5-2.5

4 19.82** 1 -0.9755*** -25.255 0.790 1037

iBoxx€ Liq Sov Cap 2.5-5.5

4 31.24** 1 -0.9740*** -20.314 0.700 1037

iBoxx€ Liq Sov Cap 5.5-10.5

4 125.03**

1 -0.9793*** -30.453 0.580 1037

iBoxx€ Liq Sov Cap 10.5+ 4 151.25**

1 -0.9820*** -29.650 0.550 1037

iBoxx€ Liq Sov Cap 1.5-10.5

4 151.22**

1 -1.0574*** -29.606 0.58 1037

Average (mean) 95.712 -27.056 0.64

Ebeb.rexx 1.5-2.5 4 17.88** 1 -0.9655*** -29.716 0.790 1037eb.rexx 2.5-5.5 4 27.67** 1 -0.9626*** -32.244 0.710 1037eb.rexx 5.5-10.5 4 127.01*

*1 -0.9660*** -24.649 0.350 1037

eb.rexx 10.5+ 4 172.03**

1 -0.9765*** -29.972 0.390 1037

eb.rexx DE 4 29.41** 1 -0.9774*** -29.979 0.67 1037Average (mean) 74.800 -29.312 0.58

NL JT trace Normalised coefficients

DF TE4 N

db x-trackersshort iBoxx 4 46.07** 1 -1.0273*** -30.982 0.420. 687iBoxx € Sov 4 35.27** 1 -0.9776*** -32.723 0.54 937iBoxx € Sov 1–3 4 9.42 - - - 934iBoxx € Sov 3–5 4 25.56** 1 -0.9781*** -35.604 0.610 934iBoxx € Sov 5–7 4 56.85** 1 -0.9802*** -36.015 0.390 931iBoxx € Sov 7–10 4 36.93** 1 -0.9796*** -33.610 0.510 931iBoxx € Sov 10–15 4 38.78** 1 -0.9782*** -31.539 0.500 929iBoxx € Sov 15+ 4 80.99** 1 -0.9809*** -27.250 0.310 929iBoxx € Sov 25+ 4 103.73*

*1 -0.9808*** -24.073 0.24 927

Average (mean) 48.178 -30.497 0.44

LyxorEuroMTS 1-3 4 75.79** 1 -0.9664*** -30.409 0.210 1037EuroMTS 3-5 4 169.30*

*1 0.9731*** -31.854 0.09 1037

EuroMTS 5-7 3 283.29**

1 -0.9786*** -29.626 0.09 1023

EuroMTS 7-10 3 278.16**

1 -0.9774*** -29.696 0.09 1023

EuroMTS 10-15 3 261.71**

1 -0.9723*** -29.634 0.11 1037

EuroMTS 15+ 1 360.42**

1 -0.9792*** -22.996 0.36 911

Average (mean) 238.12 -28.761 0.16

Overall ranking

TE1 TE2 TE3 TE4 Total score Tracking

performance rank

iShares Barclays 5 1 1 5 12 3

iShares iBoxx€LiqSovCap 4 4 4 4 16 4

iShares eb.rexx 3 3 2 3 11 2

db x-trackers 2 2 3 2 9 1

Lyxor 1 5 5 1 12 3

Determinants of TE1 and TE2• STDEV3: rolling 3 month standard deviation of underlying indexes. • LNDURATION: log of index duration. Proxy for maturity of bonds

and their sensitivity to changes in interest rates. • PLEHMAN: 1 for observations post 1st September 2008, and zero

otherwise. • SYNTHETIC*PLEHMAN: examine differences in the tracking

performance of swap and physically-based ETFs during the credit crisis.

• TER: portion of management fees deducted on a daily basis. • BID-ASK: daily bid-ask spread for ETFs. • LNSIZE: is a natural logarithm of daily AUM values.

Determinants of TE1 and TE2• LNWCDS: log of of weighted averages indexes’ CDS.

Calculate that using daily weights and CDS spreads for the index constituents.

• WPCDS: weekly percentage change of the weighted average CDS spread.

Determinants of TE – Panel data analysis

1 2 3

4 5

6 7 8

*it it it it

it it

it it it it

TE BID ASK LDURATION PLEHMAN

SYNTHETIC SYNTHETIC PLEHMAN

STDEV TER LNSIZE

1 2 3 4

5 6 7 8 9

10 *

it it it it it

it it it it it

it it

TE BID ASK LDURATION LYXOR PLEHMAN

STDEV TER LNSIZE LNWCDS WPCDS

LNWCDS WPCDS

Findings from the panel regressions

• The results for model 1 indicate a positive and statistically significant association of tracking errors with STDEV3 and LNDURATION (hypotheses 2 and 3).

• The coefficient for LNSIZE negative and statistically significant suggesting that larger funds tend to exhibit better tracking performance.

• Tracking errors have increased since the sovereign crisis, resulting in a positive and statistically significant coefficient for PLEHMAN, as predicted by our hypothesis 4.

Findings from the panel regressions

• Positive and statistically significant variable SYNTHETIC suggest worse overall performance of sample synthetic ETFs. This is consistent with Hypothesis 6 for TE2 but not for TE1.

• To study TE1 further examine possible differences between sample synthetic funds we introduce LYXOR and DBX categorical variables. The results highlight the differences between Lyxor and db x-trackers.

• For example, the coefficient for LYXOR is negative while the coefficient for DBX is positive and highly statistically significant.

• The results for SYNTHETIC were, therefore, driven by the performance of db x-trackers.

Findings from the panel regression• Coefficient for SYNTHETIC*PLEHMAN significant. This

confirms a different tracking performance of synthetic and physically-based ETFs during the sovereign debt crisis.

• The results suggest a better performance of synthetic funds during the crisis measured by TE2 and worse performance measured by TE1.

• First of these results in part related to higher correlation of returns for different asset classes during the crisis. The increase in correlation was particularly evident for corporate and government bonds (Nomura, 2011).

Findings from the panel regression• Proxies related to the sovereign credit risk show importance

of CDS spreads for the tracking performance lending support to hypothesis 5.

• The coefficients for credit spread levels (LNWCDS) are, however, positive and statistically significant only in models for TE2.

• Similarly, the coefficients for volatility of spreads (WPCDS) and interaction variables (LNWCDS*WPCDS) are statistically significant only in models for TE2 and not for TE1.

• Overall, the results confirm that variations in credit risk of index constituents tend to be more relevant for TE2 than for TE1.

Conclusions

• ETFs underperform their respective benchmarks. Large discrepancy across types of tracking errors and funds.

• Deterioration of the sample ETFs’ TE2 tracking performance during the crisis period. Credit risk considerations are increasingly important. Not so much for TE1.

• Evidence for the importance of volatility of underlying indices, duration, the replication method, bid-ask spreads, management fees and ETFs’ size to the tracking performance.

• To compare across families of ETFs adopting different replication methods caution is needed.

• When TE2 and TE3 are used, physically-based ETFs show superior performance

• Opposite is true if TE1 and, especially, TE4 are used. • Lyxor ETFs tend to improve the performance during the more

volatile market conditions, measured by TE2.

Conclusions• Edhec, 2009 survey suggests that European investors prefer

to use ETFs passively and favour ETFs that adopt a full replication.

• At the same time, more than 70% of the investors use TE1 as the measure of the tracking performance. Half of them see ETFs tracking quality as fairly good

• Our results suggest that a great deal of the investors’ dissatisfaction with the ETFs tracking performance could be due to a mismatch between the tracking performance measures used by investors and the characteristics of selected ETFs.

• According to our findings, each ETF family should be assessed on its own merits and the choice of the error metric should be based on characteristics of ETFs, investment strategies and the underlying indexes they seek to replicate.