Can investors benefit from the performance of alternative UCITS funds?

UCITS (Undertakings for Collective Investments in Transferable Securities) is a coordinated pan-European regulatory framework for mutual funds with a strong focus on investor protection that facilitates cross-border marketing and distribution of funds inside the European Union.

Mutual funds with hedge fund-like strategies are also available in the United States, but under a slightly different regulatory framework. We do not analyze these funds in our study, but provide references to related studies whenever appropriate because they are also mutual funds that apply hedge fund-like strategies. For example, Agarwal et al. (2009) study the performance of hedged mutual funds.

Blake and Timmermann (1998) analyze UK mutual funds, Dahlquist et al. (2000) Swedish mutual funds, and Bessler et al. (2009) German equity mutual funds.

As of June 2014, European mutual funds regulated by the UCITS framework managed a total of 7464 billion Euros, according to the European Fund and Asset Management Association (2014). Of this, 32.5% were managed by funds domiciled in Luxembourg and 15.5% by funds domiciled in Ireland. In our sample, funds from Luxembourg and Ireland account for 70% of the sample.

Investments in physical commodities as well as hedge fund, private equity, and real estate holdings are generally not allowed. However, some jurisdictions allow investment in financial indices representative of non-eligible assets.

This is in contrast to US Act 40 mutual funds, which are not permitted to charge performance fees.

For multiple share classes with the same inception date, we use either (1) the share class whose net asset value was calculated using the fund’s base currency, or (2) if there were still multiple share classes left, the one with the lowest management fee. If a share class had a 0% management fee, we use the one with the next lowest published management fee. When it closed but other classes remained active, we select a different share class according to the procedure described, and merge the return histories to extend the time series. We count 48 cases where the oldest share class closed while others remained active.

To test whether currency effects have an impact on our findings, we convert all funds and benchmark portfolio returns into Euro. The results (not shown) remain qualitatively similar.

We use 1-month LIBOR rates for Euro-, USD-, GBP-, JPY-, and CHF-denominated funds, and 1-month STIBOR, CIBOR, and NIBOR rates for SEK-, DKK-, and NOK-denominated funds, respectively. A caveat is that this approach does not lead to full hedging unless the fund return is known in advance.

For example, see Brown and Goetzmann (1995), Malkiel (1995), Elton et al. (1996), Ackerman et al. (1999), Brown et al. (1999), Fung and Hsieh (2000), and Malkiel and Saha (2005).

We require only 52 weeks of return history, while many studies on hedge fund performance and performance persistence require at least 24 months of data (Ackerman et al. 1999; Fung and Hsieh 2000; Kosowski et al. 2007; Boyson 2008). Using a shorter minimum return history should further mitigate any potential bias.

An exception is the fixed income category, within which all funds are combined. Busack et al. (2014) find that different fixed income strategies exhibit similar exposure to systematic risk factors.

See https://​doi.​org/​mba.​tuck.​dartmouth.​edu/​pages/​faculty/​ken.​french/​data_​library.​html.

See https://​doi.​org/​research.​stlouisfed.​org/​fred2/​.

Agarwal et al. (2009) analyze only 52 alternative mutual funds with equity hedge strategies (27 with and 25 without offshore experience). Dewaele et al. (2013) use 182 funds with hedge fund experience in their sample, of which 63 had direct hedge equivalents. Moreover, due to short fund histories, Dewaele et al. (2013) aggregate funds into portfolios and mainly test differences over time.

A prominent example is the DB Traxis Global Equity Macro UCITS fund, whose strategy was implemented in the Traxis Fund LP, managed by former Morgan Stanley Chief Strategist Barton Biggs. The UCITS fund was shut down 3 days after Mr. Biggs died on July 14, 2012.

When testing for risk persistence, winning (losing) funds are defined as funds with above (below) median risk for a given measure.

Based on the work in Christensen (1990), \(\sigma _{\mathrm{ln}({\mathrm{CPR}})} =\sqrt{\frac{1}{WW}+\frac{1}{WL}+\frac{1}{LW}+\frac{1}{LL}}\) is the standard error of the natural logarithm of the CPR (see also Agarwal and Naik 2000 using hedge fund data).

However, this approach can lead to another problem: as funds cease to exist during the evaluation period, there is not always a sufficient number of weekly returns in the evaluation period to calculate performance measures. For example, when a fund does not last more than 8 weeks during an evaluation period, we would not be able to calculate a value for the seven-factor alpha. To solve this problem, we follow Malkiel and Saha (2005) and consider funds that fail to last over the entire evaluation period as losers (L). As shown above, defunct UCITS funds significantly underperform surviving funds, which justifies this procedure.

This procedure is also employed in studies of mutual and hedge fund performance persistence (see, e.g., Brown and Goetzmann 1995; Gruber 1996; Carhart 1997; Otten and Bams 2002; Kosowski et al. 2007; Boyson 2008; Jagannathan et al. 2010).

As previously mentioned, fund returns are net of fees. Transaction-based fees such as subscription or redemption fees are not included.

We address possible backfill or incubation bias in the robustness section.

Cogneau and Hübner (2009) provide a categorization of more than 100 performance measures. Considering that the nature of their risk dimension is “absolute” (see Exhibit 2 in Cogneau and Hübner 2009), we denote the Sharpe ratio and variants thereof as absolute performance measures. For measures like the appraisal ratio and related measures, they note that the underlying idea is to obtain a performance relative to a benchmark. However, these authors call measures like alpha “incremental return vs. benchmark” measures (see their Exhibit 3). Although alpha and the appraisal ratio are in different categories according to Cogneau and Hübner (2009), we chose to call them “relative” performance measures because a benchmark (factor) model is needed to compute each of these measures.

Herzberg and Mozes (2003) further document that low-risk hedge funds outperform high-risk hedge funds; thus persistence in risk might be of value to investors when choosing funds.

More technically, these measures are the higher and lower partial moments of order one (with zero threshold).

In the presence of negative excess returns, ranking funds on performance measures that are increasing in excess returns and decreasing in risk can lead to rankings that appear counterintuitive. For example, consider two funds, one with a negative excess return of −5% and a standard deviation of 10%, and another with a negative excess return of –10% and a standard deviation of 30%. Without the adjustment, the first fund would have a Sharpe ratio of −0.5, while the second fund would have a Sharpe ratio of −0.3. To correct for negative values of the numerator in a performance measure, Israelsen (2005) adds a correction term to the performance measure’s denominator. In this example, the Sharpe ratio would take the following form: \(SR_i =r_i /\hat{\sigma }_\imath {({r_i /| {r_i } |})}\), where \(r_i \) is the fund’s excess return, and \(\hat{\sigma }_{\imath }\) the fund’s estimated standard deviation. The two Sharpe ratios would then be \(-5\% \times 10\% =-0.005\) and \(-10\% \times 30\% =-0.03\). Assuming that investors prefer higher returns, in this case losing less, and lower risks, this correction leads to the correct ranking order \((-0.005>-0.03)\). In the case of positive excess returns, the correction would yield the standard Sharpe ratio.

Alternatively, we could report the absolute number of repeat winners and winner–loser funds. However, as the table is already large, we chose to present percentages. The percentage of winner–loser funds can easily be calculated as \((0.5 - WW).\) As contingency tables divide the sample at the median, the percentages for LL and LW are equal, except for ranking periods with an uneven number of funds.

We require that the sample for our initial ranking period contains at least 20 funds to ensure meaningful results. Brown et al. (1999) have 43 funds in their initial 1989–1990 period, and Malkiel and Saha (2005) begin their sample in 1996 with only 18 winner funds.

An example illustrates the methodology for ranking funds over 52 weeks and evaluating performance over the following 26 weeks. At the beginning of January 2005, we sort funds into equally weighted quartile portfolios according to their prior 52-week performance. Portfolio weights are rebalanced and remain equally weighted during the evaluation period, and the first period for which portfolio returns are calculated ends in June 2005. Subsequently, funds are sorted according to their performance from July 2004 until June 2005, and new equally weighted portfolios are formed. This procedure is repeated until the end of the sample period.

Hendricks et al. (1993), Brown and Goetzmann (1995), Carhart (1997), Bollen and Busse (2004), Kosowski et al. (2006, 2007), and Boyson (2008) use risk-adjusted performance to sort funds into portfolios.

Although funds from the top quartile outperform bottom quartile funds, the pattern is not always monotonic and depends on the length of the period considered as well as the measure used to select funds. For example, quartile returns monotonically decline for a ranking period of 52 weeks with a holdout period of 26 weeks when one uses absolute performance measures, like the Sharpe, Omega, or Calmar ratio, to select funds. Using relative performance measures, we find that funds that were placed in the third quartile exhibit the worst performance during the evaluation period. If one extends the evaluation period to 52 weeks, there are no longer any clear patterns, except that the top quartile funds achieve positive returns, while the bottom quartile funds do not.

If the evaluation period is extended to 52 weeks, the performance of the top quartile portfolio declines to 0.7% (Sharpe ratio) and 0.8% (Calmar ratio), while the bottom quartile portfolios lose about 2%.

As another robustness check, we rerun ranked portfolio tests for the full sample and the equity long/short subsample, but now let the first ranking period begin in July 2005 instead of January 2005. It is possible that our results for the 52 (104) week ranking and 52 (104) week evaluation combinations might be influenced by constantly ranking funds from January–December. The results (not tabulated) remain qualitatively unchanged.

Note that the unconditional survival probabilities for the evaluation period are separately calculated for winner and loser funds. This survival probability is conditional upon being a winner or a loser during the ranking period (Ter Horst et al. 2001; Baquero et al. 2005). Therefore, to calculate the unconditional survival probability during the evaluation period for a winner fund, we divide the number of surviving winners by the number of all winners at the end of the ranking period.

The detailed results are available upon request.

In robustness tests (not shown), we construct differential return series of top quartile funds with sister hedge funds and top quartile funds without. As in our main analysis, top quartile funds with a sister hedge fund subsequently achieve higher performance than top quartile funds without sister hedge funds, albeit this result is limited to ranking and evaluation periods of 52 weeks.

We use data on fund assets from Morningstar Direct. To fill gaps and obtain as much asset data as possible, we complement missing data with data from Bloomberg.

We use a pooled panel model instead of a fund fixed effects model because some variables (e.g., fund domicile, performance fees, and offshore experience) are time invariant. To account for fund-specific heterogeneity, we use standard errors that are clustered at the fund level. In addition, we re-estimate our regression as a random effects probit model and find very similar results.