Bias

6.2.1. Management fees

It could be assumed that the size of the management fees might influence the outcome of the performance analysis. This is probably due to the screening process, which is assumed to be more extensive for a fund with an ethical focus in comparison to a conventional fund, since the screening for ethical stocks is more time consuming. However, Bauer et al. (2005) examined the impact of the management fees on the fund performance and stated that the difference in return between ethical and conventional mutual funds still is statistically insignificant after it has been adjusted for management fees. Furthermore, Renneboog et al. (2007) studied the difference in management fees between ethical and conventional mutual funds. The results demonstrated no significant difference in management fees. Management fees are not included in the calculation of NAV, since previous findings imply that a significant difference between conventional and ethical funds does not exist after adjusting for management fees.

6.2.2. Survivorship bias

Brown et al. identified the survivorship bias in their study in 1992. The study examined dead mutual funds’ performance historically and the found a tendency that those funds had performed badly. Moreover, poor performance tends to be the reason why the funds are closed down. The effect of excluding dead funds in performance studies was observed. The findings showed that the average performance is overestimated if dead funds are disregarded.

6 A mutual fund with a minimum of 75% of its holdings invested in equities. (Finansportalen, 2016)

Also, Bauer et al. (2005) tested which influence it had on the returns to exclude non-surviving funds from the analysis over the period 1990 - 2001. The authors found that “restricting our sample to surviving mutual funds leads to a substantial overestimation of average returns, namely by 0.14% (Germany), 0.17% (United Kingdom) and 0.31 % (United States) per year.”

Our dataset of ethical funds consists of funds that have survived through the entire investigation period up until February 2016. Consequently, the dataset is subject to survivorship bias. Based on findings from previous studies, the returns are likely to be overestimated when excluding dead funds, due to the historical track record of bad performance among non-surviving funds.

There are several solutions to correct for this probable bias. The most obvious one is to include dead funds in the dataset. Elton et al. (1996) presented another possible solution to the problem of survivorship bias. A fund that disappears is most often merged into another fund, rather than dissolved. The authors computed the risk-adjusted return of a merged fund by observing return for the original fund prior to the merger then computed return in the month of the merger, and lastly calculated risk-adjusted return for the combined fund after the merge. The sample was free of survivorship bias and the original fund’s performance was estimated by using data from the new merged fund. In that way, was the merged fund accounted for during the entire period. The effect of the survivorship bias was estimated by comparing the performance of a sample of funds subject to survivorship bias and the survivor ship bias free fund. The result showed that survivorship bias does exist. However, an identification of all ethical funds over the estimation period have to be done and it has been proven to be impossible to identify dead ethical funds through Morningstar, since Morningstar delete historical information about dead funds. However, Datastream does obtain data of non-surviving funds. This makes it possible to collect the financial data on dead funds for the period when the fund existed and include that in the sample.

Unfortunately, a complete list over dead ethical funds has been proven too difficult and too time consuming to identify.

Another solution to avoid survivorship bias is to obtain the US ethical fund data from the CSRP (Center for research in Security prices) survivor-bias free mutual fund database. However, this was not an option since the access to the database is restricted, and Copenhagen Business School has not got any access to this database.

As previous mentioned are the resources of documentation of ethical funds in general limited.

Furthermore, it should be noted that our observation period covers the global financial crisis where the financial markets dramatically changed. There was a net outflow from actively managed

funds during 2008⁷ and the number of funds, which were liquidated or merged with other funds are likely to have increased during the period. It has been proved impossible to be able to track all non-surviving funds over the whole investigation period. Hence, our estimation of fund returns suffers from survivorship bias.

6.2.3. Heteroscedasticity

The problematic issue of heteroscedasticity is previously discussed in Chapter 5. The data is adjusted for heteroscedasticity, and heteroscedasticity-robust standard statistics after estimation by OLS will be used.

The package sandwich in the statistical computing program R is applied in order to adjust for heteroscedasticity.

6.2.4. Multicollinearity

Multicollinearity occurs when two, or more, independent variables are highly correlated.

Multicollinearity is further described in Chapter 5. In order to test for multicollinearity the correlations between pairs of the individual independent variables are examined. The highest correlation observed are between the market premium factor and the local factor in the Scandinavian area. The correlation values between those two factors range from 0.4 to 0.65 for Scandinavian mutual funds, investing both regionally and globally. However, all correlation values were smaller than 0.65, which should be considered as a moderate level of correlation.

Next, a further examination was performed in order to assess multicollinearity. The Variance Inflation Factor (VIF) measures the effect of multicollinearity among the variables in a regression.

The VIF value range starts at 1. If VIF =1 no multicollinearity exists, although multicollinearity occurs if VIF > 1. However, the presence of multicollinearity might not be a problem. A problem arises when the independent variables are highly correlated with each other and will have an impact on the estimation of 𝛽_m. The critical VIF value, when multicollinearity is considered as problematic is hard to determine and could be arbitrary (Brooks, 2008) .

Authorities differ on how high an acceptable value of VIF could be, and the acceptable values differ from source to source. Brooks (2008) suggest that a critical value of 10 is most commonly

7http://www.wsj.com/articles/morningstar-says-actively-managed-mutual-funds-saw-outflows-in-2015-1452701873

used. However, a more conservative value has been set as the critical value for the VIF test of the regression models in this study.

A VIF value greater than 2.5 is considered to be problematic.⁸

The VIF tests in this study show no indications of multicollinearity in the regression models. All values are smaller than the critical value and are closer to 1, i.e. a VIF value near 1 suggests that multicollinearity is not a problem for that independent variable (Wooldridge, 2013) .

6.2.5. Presence of outliers in the data set

The presence of outliers might cause a non-identical distribution. Outliers are anomalous values in the data that might affect the fitted coefficients, which will result in a poor fit in relation to the data observations. The presence of outliers in the data set was checked for through a histogram and no anomalous values were detected.