The Predictive Performance of Swedish Premium Pension Fund Ratings

Transcription

1 The Predictive Performance of Swedish Premium Pension Fund Ratings Author: Yanjun Wang Abstract Rating is a well-known tool to identify the performance of funds. Swedish Pension Agency import Standard & Poor s rating and Wassum rating as new references for fund selection. This study focuses on studying whether the predictive performance of premium pension funds ratings exists. The analysis shows we can barely conclude any predictive performance for ratings given by both companies. Master Paper in Economics Supervisor: Bo Larson Department of Economics September 2011

2 I. Introduction The Swedish pension reforms in 1998 introduced a partly-funded pension plan for accumulating available assets to cover long-term social liabilities. Individuals are taking activities in their own pension savings, i.e. the development of chosen pension funds is directly in relation to the future retirement pension. Most of the time, people are blind and naive on financial investmenting. The study by Capon, Fitzsimons and Prince (1996) investigates that for investors, fund performance is seen as the most important fund selection criterion, but a large group of investors appear to be highly ignorant about their investments. Sirri and Tufano (1998) point out those mutual fund consumers investing in funds with the highest recent returns. In fact, many factors should be taken into the consideration while choosing funds. This leads to investors seeking for aid to identify the best performing funds. Year 2007, the Swedish Pension Agency imports a new tool - fund ratings, as an additional assistance for fund selecting. Many academic studies indicate that fund ratings are important in explaining fund performance and may significantly influence fund flow (Patel, Zeckhauser and Hendricks (1994), Del Guercio and Tkac (2002)), and there also exits increasing number of paper that analyze the predictive performance of fund ratings (e.g. Black and Morey (2000), Kräussl and Sandelowsky (2007)). The purpose of the paper is to test the predictive performance of premium fund ratings provided by Standard & Poor s Fund Management Rating and Wassum Fund Rating. The results indicate that it hard to see any predictive ability with either rating systems by the empirical experiments. The paper is going to be organized as follows: first, we introduce the background of premium pension funds catalogs, information of Standard & Poor s Fund Management Rating and Wassum Fund Rating and information about Morningstar Ratings.; second, we formally describe our data and build up the empirical model; third, we test our model and explain the results, then we give some case studies for individual phenomenon; finally, we give our conclusion. 1

3 II. Background Recall the structure of the public pension in the following triangle, Figure 1. Fully funded (2.5% of earnings) Premium Pension Pay-as-you-go (16% of earnings) Income Pension Guarantee pension Basic Pension The public pension consists of income pension (pay-as-you-go part) and premium pension (funded part). It is financed by contributions of 18.5% of total earnings every year. Out of the total contribution of 18.5%, 16 % points allocates to pay as you go earnings related system. The remaining part 2.5% will finance the fully funded system, called the premium pension. The basic pension is provided as guarantee pension for those individuals who have no or low income. Premium pension accounts for 2.5% of the total public pension. The investment risk of the pension assets is managed by employing premium pension. This part of the public pension encourages people to have more initiative and enthusiasm to manage their pension assets. The pensioner will obtain benefits by investing into funds from various markets. The size of the premium pension depends on whole life-time earnings and how those chosen funds develop. Therefore, even though the premium pension takes a small part of retirements, it is still very important and the decisions of choosing funds are important as well. The Swedish Pension Agency publishes a fond category every March each year. It contains all the information of existing premium pension funds, namely fund types, fund numbers, fund names, fees, risks and so on. It also explains to the pensioners how to choose their funds and the helping tools of choosing right funds to meet individual s requirements. 2

4 Year 2007, the Swedish Pension Agency imported a new aid tool rating, as the measurement to control the quality of the funds. Fund ratings display the possibility of a fund outperforming the average return in the fund category it belongs. The ratings are based on evaluation of how fund companies managing chosen funds and also the fund returns and risks. Pension agency selected Standard & Poor s Fund Management Rating and Wassum Fund Rating to provide fund ratings as an additional assistance for decision making. Pensioner can compare funds in the same category with the help of fund rating. The rating of funds is voluntarily and it is up to fund companies own accord. Fund companies are responsible for the cost of their own fund ratings. Therefore fund ratings are not tailored to pension agency s fund supply, but are the rating companies own products. There are about 300 out of 800 pension funds have ratings. Standard & Poor s requires the fund to establish for at least two years in order to get rated. They rate only those funds with best returns and only 20% funds with best performance are included in the rating process in every category. They do not use the same fund categories as pension agency does but use their own way of choosing them. Meanwhile, Wassum Fund Rating requires the fund to establish for at least three years in order to get rated. This rating company does not have any selection criteria and rates funds with both good and less good returns. Like Standard & poor s, Wassum also chooses those fund categories that they consider to be relevant. Standard & Poor s Fund Management Rating uses the following symbol for rating: AAA: Best opportunity to outperform the average returns in its category. AA: Very good opportunity to outperform the average returns in its category. A: Good opportunity to outperform the average returns in its category. Wassum Fund Rating uses the following symbol for rating: WWWWW: Opportunity to outperform the average returns in its category. WWWW: Opportunity to outperform or perform on par with the average returns in its category. WWW: May perform on par with the average returns in its category or slightly below it. WW: May perform on par with the average returns in its category or slightly below it. W: May perform on par with the average returns in its category or slightly below it. 3

5 Another rating system that could be used by Pension Agency is the Morningstar rating system. Morningstar provides ratings for all funds with more than three years of history and uses a concept of Morningstar Risk-Adjust Return as basis to assign the star ratings within the respective category. However, Standard &Poor s rating or Wassum rating is built on other method with major elements of a quality assessment. Pension Agency considers itself as a regulator that should do nothing but control the formal requirements and the fund supply should be an open system for all. The decisions of whether to get ratings and rating fees to the rating companies are determined by fund companies themselves. This is the reason why Pension Agency chose Standard & Poor s and Wassum as rating companies for premium pension funds. The purpose of the paper is to test the predictive performance of premium pension fund ratings, i.e. Standard & Poor s ratings and Wassum ratings. We will also import the Morningstar ratings to do the same test. Since Standard & Poor s and Wassum use some kind unknown fundamental analyses and there is financial relationship between fund companies and these two rating companies. Morningstar ratings are calculated base on a technical analysis (This will be explained in details in the coming part) and Morningstar does not get paid by fund companies. The costs have already deduced when calculating fund s unit price. Morningstar ratings would be seen as a kind of control for the others since they are relatively unbiased. Morningstar assigns fund ratings by Morningstar Risk Adjust Return (MRAR). The measurement of MRAR is base on Expected Utility Theory, which is investors prefer expected low returns better than a unexpected high returns and are willing to give up part of expected return in exchange for benefits with strong certainty. In order to calculate MRAR, we first define the fund s monthly total return as: (1) where is monthly total return, is net asset value at the beginning of the month, is the net asset value at the end of the month, is the dividend at time i, is reinvestment net asset value at time i, is the number of dividend. The cumulative total return is: (2) 4

6 where is the cumulative value, is total return for month t, T is the number of months in the period. A fund s load-adjusted return is calculated as follows: (3) Where is the cumulative value, adjusted for loads, is cumulative value, is the front load, is redemption fees, is deferred load, is net asset value at beginning of the period, is net asset value at the end of the period. The monthly adjustment factor for loads, a, is defined: (4) where T is the number of months in the period. The total monthly return is adjusted by this factor: Where is adjusted total return for month t. The MRAR is defined as: (5) (6) where is the geometric excess return in month t expressed as: where is the return on a risk-free asset in month t and is a parameter that describes the (7) degree of risk aversion. Morningstar uses Morningstar 2007). equal to 2 to calculate the star ratings. (See Morningstar category is basis for star ratings. Morningstar distributes funds with similar investment focus in the same category and there are more than 170 categories in Europe. (See Morningstar 2010). Whether or not compare funds within the same category will directly influence the effectiveness of the rating. The most popular categories in Sweden are, for example, Sweden big company, Global & Sweden, Europe mix company, Russia, Industry fund medicine, risk free fund SEK money market and so on. When set ratings for funds, we first divide funds into categories, then calculate the MRAR for each fund and sort them in the descending order. The first 10% are rated with 5 stars, the next 5

7 22.5% are rated with 4 stars, the middle 35% are rated with 3 stars, the coming 22.5% are rated with 2 stars and the last 10% are rated with 1 star. (See Morningstar 2007). Finally we count the Morningstar overall rating as following (See Morningstar 2007): Table 1 Historical returns Weighted Overall Rating months 100% three year rating months 60% five year rating 40% three year rating 120 months or more 50% ten year rating 30% five year rating 20% three year rating III. Data and methodology The datasets used for analysis in this paper are pension funds unit price sheets, dividend sheets, premium fund catalogs, and finally Morningstar ratings. Unit price sheets were downloaded from the website of Pension Agency. For each fund, it contains data on fund company name, fund number, fund name, date, exchange rate, unit price, and unit price in SEK, on a daily basis. Dividend sheet was also downloaded from Pension Agency s website. It contains fund names, fund numbers, dividend dates and dividend ratios. Since the premium fund rating started from year 2007, there were four Premium fund catalogs ordered from Pension Agency from year 2007 to 2010 and it contains the Standard & Poor s and Wassum ratings in each catalog for that year. Dataset of Morningstar ratings for pension fund was directly obtained from Morningstar. The premium fund catalog is published in March; we therefore required the overall star ratings in March from year 2007 to year Moreover, fund company name, fund number, fund name and inception date are supplied for each rated fund. As we could see from each year s premium fund catalog, approximately one third of the population; more than 300 funds have the rating record. Within all the rated funds, Standard & Poor s provide around 80 funds and the rest of the ratings belong to Wassum. Morningstar 6

8 on the other hand rate all the funds in the market with presence of more than 3 years. We select all funds that have a rating in March from 2007 to We formulate the four corresponding groups for each rating type in Table 1. These funds were divided into three sub-groups according to Standard & Poor s rating and five sub-groups according to Wassum and Morningstar. Table 1 shows the characteristics for the samples used to calculate the predictive performance of each rating system. It provides information on sample size and rating distribution. The concepts of testing the predictive performance of ratings given by different rating company are the same. We apply the following OLS-estimated equations using one year, two year, three year and four year out of sample fund returns and theirs corresponding ratings in the regression. Standard & Poor s: (8) Wassum: (9) Morningstar: (10) Table 2. Sample information S&P Sample 1A 2A 3A % 48.10% 13.92% N= % 44.58% 15.67% N= % 36.84% 14.47% N= % 37.18% 12.82% N= Wassum Sample 1W 2W 3W 4W 5W % 20.25% 43.39% 18.60% 10.33% N= % 17.86% 40.08% 24.6% 8.73% N= % 17.94% 41.22% 23.28% 9.92% N= % 23.05% 34.94% 24.16% 11.52% N=

9 MorningstarSample 1S 2S 3S 4S 5S % 24.74% 55.41% 17.01% 0 N= % 18.51% 36.83% 29.20% 12.6% N= % 18.63% 39.6% 26.76% 10.31% N= % 18.62% 36.58% 30.20% 10.23% N= These equations are expressed in the same manner. We choose equation (9) as an example to explain the meanings of all variables and coefficients. is a return for fund at time ; is the constant term indicating the performance of a 5w fund at time 0; is a binary dummy variable indicating whether fund is a 1w fund at 0; this is same for, and, which are all binary dummy variables indicating whether the fund is a 2w fund, 3w fund or 4w fund respectively, at time 0; finally, is the error term. Intuitively, if a fund is a 5w fund, all the dummy variables should be equal to zero. The constant term is similar to the return of 5w fund group. If a fund is a 4w fund, then, and should equal to zero and should equal to 1. The return of 4w fund group is therefore, i.e.. Likewise, the returns of 3w fund group, 2w fund group and 1w fund group are ( ), i.e.,, i.e. and, i.e. respectively. Assuming Wassum rating system has great predictive performance abilities, the following inequality should hold: This implies: As the rating system is perfect in predicting future performance, a 1w fund should not outperform a 2w fund, a 2w fund should not outperform a 3w fund and so forth. A 5w fund should have the best performance with no fund can surpass. This implication is identical for Standard & Poor s ratings and Morningstar ratings, i.e.,, and are similar to returns with the highest rating fund groups. This study can be very valuable for investors to understand the predictive 8

10 performance of a given fund rating system. We could use the findings of the test to check if predictive performance of a fund rating is valid in the certain period and also see the shift in performance as time period increases from one year to four year. The results are evident for proving whether fund rating is a practical tool for pensioner while choosing their premium pension funds. For instance, if a fund has a Wassum rating of 5w, will it outperforms rest of the funds with lower Wassum ratings in a coming period? IV. Empirical Analysis This part of the study will analyze the predictive performance of Standard & poor s and Wassum s rating system. As a control, the analysis of the predictive performance for Morningstar s rating system will also be included. Table 3 shows the results of predictive performance of premium pension funds with one year S&P ratings. It lists different base year (2007 to 2010) of the sample for one year out of sample period, the aggregated sample from all base year and also number of observations in each sample. This table presents the regression results on the samples in which all the funds are rated by Standard & Poor s Fund Management Rating. The predictive performance is calculated by a period of one year. Intuitively, the results tell the pensioners that, if one chooses a fund with a Standard and Poor s highest rating 3A in March year 2007, will this fund still be as good as it was at the same time in year This intuition is the same when testing predictive performance by two, three and four year out of sample period. Table 3. Predictive Performance of premium pension funds with one year S&P ratings S&P 2007(n=63) ** ** ** 2008(n=78) *** (n=73) *** (n=75) Aggregated(n=289)

11 The sample size for Standard and Poor s is not big for each year. The numbers of observations in regressions become even smaller when we calculate the one year future returns (compare with Table 2). This is because premium fund supply is not unchanged for each year. Pension Agency may add or exclude certain funds during the year, which lead to a given rated fund at time t may no longer exit at time t+1. It also explains the shrinkage in number of observations for the rest of the regressions. However, this shortage could somehow be balancing out by Standard & Poors fewer independent variables (see equation 8). Table 3 indicates that we can hardly tell the predicting ability in one year period. Among these regressions, all s carry the correct sign, of which only one of them is significantly different from zero at the 5% level. This implies that, with Standard & Poor s rating, it is hardly to tell 2A funds and 3A funds perform differently. The situation is worse with s, since one of them carry the wrong sign and only one of them is significant. Additionally, of all the one year predictive performance tests, the regression for year 2007 is significant at 5% level and both coefficients of the regression are significant. This means the regression s results make sense, but 2A funds slightly outperform 3A funds. Table 4. Predictive Performance of premium pension funds with one year aggregated Morningstar ratings Star Agg. (161) *** Morningstar does not provide all ratings for those Standard & Poor s rated funds. We lose some observations when testing predictive performance Standard & Poor s rated fund with Morningstar ratings. This can be problematic when regression has fewer observations but more independent variables (equation 10). We therefore use only the aggregated data to avoid defect of small sample size. Table 4 clearly shows that the regression results do not make any sense. There is no significant difference among all rated funds. It is similar to Standard & Poor s aggregated regression s results. We cannot conclude any existence of predictive performance for aggregated Standard & Poor s rated fund using Morningstar ratings. We apply equation 9 while testing the predictive performance of Wassum rating system. Table 5 presents the regression results on the samples in which all the funds are rated by 10

12 Wassum Fund Rating. The predictive performance is calculated by a period of one year. It indicates that we can hardly tell the predicting ability in one year period with these regression results. Of all the 20 coefficients, 16 of them carry the correct sign, of which none of them is significantly different from zero. These empirical findings are very frustrating. Wassum ratings are unable to predict any performance of rated funds in one year periods. Table 5. Predictive Performance of premium pension funds with one year Wassum ratings Wassum 2007(n=203) *** (n=233) *** (n=243) *** (n= 250) ** Agg.(n=929) Table 6 presents the regression results on the samples with Wassum rated funds but Morningstar s ratings. The predictive performance is calculated by a period of one year. When look at the coefficients, the situation does not improve. Of all the s, 2 are significant at the 1% level, but none of them has the correct sign. This means in base year 2009, one star rated funds significantly outperform five star rated funds. Out of 5 s, one is significant at the 1% level, but it has the wrong sign. This means in base year 2008, two star rated funds significantly outperform five star rated funds. It is failed to use Morningstar rating as the control in this case. The results cannot indicate much predictive performance by Morningstar ratings in one year periods. Table 6. Predictive Performance of premium pension funds with one year Morningstar ratings Star 2007(161) *** (190) *** *** ** ** 2009(208) *** *** ** 2010(228) * Agg.(787) *** *** 11

13 Table 7 and table 8 presents the regression results on the samples in which all the funds are rated by Standard & Poor s and these funds with Morningstar rating. The predictive performances are calculated by the period of two years. Most of the coefficients for Standard & Poor s regression and all coefficients in Morningstar regression have the correct signs; however they are all insignificantly different from zero. These results show no significant difference among funds with different ratings. Table 7. Predictive Performance of premium pension funds with two year S&P ratings S&P 2007(n=63) ** (n=74) (n=70) *** Aggregated(n=201) Table 8. Predictive Performance of premium pension funds with two year aggregated Morningstar ratings Star Aggregated ** Table 9 presents the regression results on the samples with Wassum rated funds. The predictive performance is calculated by a period of two year. When looking at s, one is significant at the 10% level, but it has the wrong sign. This means in base year 2007, 1W rated funds significantly outperform 5W rated funds. Out of 5 s, three are significant at the 5% level, but one has the wrong sign (year 2007) and two have the correct signs (year 2008 and 2009). This means in base year 2007, 1W rated funds significantly outperform 2W rated funds, 2W rated funds significantly outperform 5W rated funds. In base year 2008 and 2009, 5W rated funds outperform 2W rated funds, which is logical. For all s and s, none of them is significant, which implies that these is no significant difference between 3W rated funds and 5W rated funds and between 4W rated funds and 5W rated funds. 12

14 Table 9. Predictive Performance of premium pension funds with two year Wassum ratings Wassum 2007(n=207) *** * ** * 2008(n=221) *** ** * 2009(n=226) *** ** Agg.(n=654) * Table 10 presents the regression results on the samples with Wassum rated funds but Morningstar s ratings. The predictive performance is calculated by a period of two year. Of all the s, one is significant at the 1% level and two are significant at 5% level, but none of them has the correct sign. This means in base year 2007, 2009 and three samples aggregated, one star rated funds significantly outperform five star rated funds. Out of 4 s, one is significant at the 5% level, but it has also the wrong sign. This means in base year 2007, two star rated funds significantly outperform five star rated funds. For all s and s, none of them is significant, which implies that these is no significant difference between 3W rated funds and 5W rated funds and between 4 star rated funds and 5 star rated funds. Table 10. Predictive Performance of premium pension funds with two year Morningstar ratings Star 2007(161) *** ** ** *** 2008(190) *** (208) *** ** * Agg. (559) *** ** Table 11 presents the regression results on the samples in which all the funds are rated by Standard & Poor s Fund Management Rating on a period of three years. The coefficients are significant in base year 2007 and they also bear the correct sign, which means funds with highest scores outperform the rest of the funds after three year, however one scores funds outperform two scores funds. There is no significant difference among the funds of year 2008 and aggregated funds. With results estimated from aggregated Standard & Poor s funds with Morningstar ratings (Table 12), all S are bearing the right sign, in which only is significant. 13

15 This implies 3 star rated funds outperform 5 star rated funds and no significant difference among the rests. Table 11. Predictive Performance of premium pension funds with three year S&P ratings S&P 2007(n=60) * * (n=73) Aggregated(n=133) Table 12. Predictive Performance of premium pension funds with three year Morningstar ratings Star Aggregated(115) ** * Table 13 presents the regression results on the samples with Wassum rated funds by a period of three year. When looking at all the Ws, and are significant with the correct signs in the second regression. This means in base year 2008, 5W rated funds significantly outperform 1W rated funds and 2W rated funds, 2W rated funds significantly outperform 1W rated funds, meanwhile these is no significant difference among 3W, 4W and 5W rated funds. For base year 2007 and aggregated sample, no significant difference among rated funds exists. Table 13. Predictive Performance of premium pension funds with three year Wassum ratings Wassum 2007(n=192) *** (n=206) *** * ** Agg.(n=398) With results estimated from regression on samples with Wassum rated funds but Morningstar ratings, as shown in Table 14, nine of twelve S are bearing the right sign, in which none is significant. This implies there is no significant difference among the rated funds by Morningstar ratings. 14

16 Table 14. Predictive Performance of premium pension funds with three year morningstar ratings Star 2007(161) *** (190) *** * Agg.(351) Table 15 presents the regression results on the samples in which all the funds are rated by Standard & Poor s Fund Management Rating. The predictive performance is calculated by a period of four years. Both coefficients carry the right sign and are significant, however is larger than, which implies the predictive performance of 1A rated funds in year 2007 outperform 2A rated funds; and 3A rated funds outperform all after four years. Table 15. Predictive Performance of premium pension funds with four year S&P ratings S&P 2007(n=59) ** ** * Table 16 and Table 17 presents the regression results on the samples in which all the funds are rated by Wassum and these funds with Morningstar rating. The predictive performances are calculated by the period of four years. Half of the coefficients for Wassum regression and all coefficients in Morningstar regression have the correct signs; however they are all insignificantly different from zero. These results show no significant difference among funds with different ratings. Table 16. Predictive Performance of premium pension funds with four year Wassum ratings Wassum 2007(n=180)

17 Table 17. Predictive Performance of premium pension funds with four year Morningstar ratings Star 2007(161) According to premium pension fund catalog, all the funds have divided in different types, which are risk free asset funds, balanced funds, generation funds and stock asset funds. This helps the pensioner with distinguishing and finding funds in a simple way. With the following tests, we can analyze the predictive abilities for different rating system within a certain type. If the results have statistical interpretations, it is more reasonable to choose funds within the same fund type. The following tables show the regression results based on premium pension funds with Wassum ratings and same rated funds with Morningstar ratings of Swedish stock asset funds type, regional fund type, and country fund type. For having enough observations in the regression, we use the aggregated of all the one year samples and all two year samples to estimate. Results show accordingly the overall predictive performance in different type, instead of basing on a certain base year. Table 18 to table 25 lists the regression results, of all 82 coefficients, only 7 are significant, but two have the correct sign. The remaining 91.46% show no significant difference. These empirical findings have great difficulties in achieving our expectation. Table 18. Wassum Predictive Performance of Swedish Stock Asset Funds SwedishMutualFunds 1 year agg.(n=155) * year agg.(n=110) 0.409*** ** Table 19. Morningstar Predictive Performance of Swedish Stock Asset Funds SwedishMutualFunds 1 year agg.(143) * year agg.(100)

18 Table 20. S&P Predictive Performance of Regional Fund Regional Funds 1 year Aggregated(n=175) year Aggregated(n=98) year Aggregated(n=75) ** **. Table 21. Morningstar Predictive Performance of Regional Funds RegionalFunds 1 year agg.(151) ** year agg.(105) *** ** 3 year agg.(64) NA Table 22. Wassum Predictive Performance of Regional Funds Wassum 1y ag.(634) y ag.(445) y ag.(268) y ag.(120) * Table 23. Morningstar Predictive Performance of Regional Funds Star 1 y ag.(524) *** ** 2 y ag.(374) *** * 3 y ag.(235) y ag.(181) ** Table 24. Wassum Predictive Performance of Country Funds Country Fund 1 year agg(130) year agg(91)

19 Table 25. Morningstar Predictive Performance of Country Funds Country Fund 1 year agg.(110) year agg.(77) There are also different fund categories introduced for each fund type. For risk free asset funds, the categories are Swedish short, Swedish long, Sweden real interest, Europe and euro countries and foreign countries. For balanced funds, the categories are Swedish stocks and interests, Swedish and foreign stocks with Swedish interests, foreign stocks and interests. Categories for generation funds are pension less than 10 years, pension less than 20 years and pension more than 20 year. For stock asset funds, we explain in details in the following: Swedish stock funds: Sweden Global Europa North American and USA Asia and the far east New markets Country funds: Japan China Russia Other countries Industry funds: Biology IT and communication Medicine Other industries Table 26. Wassum Predictive Performance of Categories in Different Fund Types Category Sample Swedish stock asset funds Sweden 1 y agg.(121) NA y agg. (85) NA Regional funds Global 1 y agg.(150) y agg.(106) Europa 1 y agg.(132) y agg.(98) * ** North America 1 y agg.(93) and USA 2 y agg(67) Asia and Far East 1 y agg.(74) New Market 1 y agg.(71) ** **

20 Country funds Japan 1 y agg.(69) Similar to previous tests, we can analyze the predictive abilities for different rating system within different category. If the results have statistical interpretations, then we should choose funds within the same fund category. This is actually the method of selecting funds that is introduced in the premium pension fund catalogs. The following tables show the regression results based on premium pension funds with Wassum ratings and same rated funds with Morningstar ratings. Funds with Wassum ratings are mainly assembled in Stock asset fund type. We test the predictive performance therein seven categories with enough observations.. Since we use aggregated data, results show accordingly the overall predictive performance in different category. Table 26 presents the regression results, of all 40 coefficients, 4 are significant, and two have the correct sign. NA means there is no fund with such rating 1W in category Sweden. The remaining 90% show no significant difference. With Morningstar ratings, see table 27, we can test only three categories due to the lack of observations. The tests results are as insignificant as listed in table 26. The fact that hardly any coefficients are significant different from zero would imply that dividing funds into categories makes no sense with either Wassum ratings or Morningstar ratings. Table 27. Morningstar Predictive Performance of Categories in Different Fund Types Category Sample Swedish mutual funds Sweden 1 y agg.(115) y agg.(80) Regional funds Global 1 y agg.(126) y agg.(91) Europa 1 y agg.(151) y agg.(88)

21 V Case study In each year s fund catalog, we can see the existence of funds having both Standard & Poor s and Wassum ratings. It is nothing weird that funds are rated by both companies. However the phenomenon of having completely opposite ratings for the same fund is worth to study. In year 2007, there are 30 funds with both ratings, of which two funds are totally different. NA means for fund , its unit price is not provided in the dataset, we can therefore not calculate the returns. We evaluate the validity of rating by comparing the return with average return. If the return exceeds the average, the fund rating given by Standard &Poor s should not less than two and fund rating given by Wassum should not less than three, and vice versa. Take fund as an example, its one year return , which is bigger than one year sample mean of Standard &Poor s rated funds The rating for this fund should be not less than two. The given rating one is therefore inappropriate. On the other hand, Wassum provide a relatively correct rating of four, since the one year return excesses sample mean of Wassum rated funds, We record a little w in the parenthesis to inform that Wassum gives better rating than Standard &Poor s. Table 28 shows Wassum provides more predictive rating for funds. Table 28. Year year return 2year return 3year return 4 year return Fund S&P W S&P mean W mean S&P mean W mean S&P mean W mean S&P mean W mean Number NA NA NA NA (w) (w) (w) (w) In year 2008, out of 37 funds with both ratings, seven are opposite. We use the same method to determine which rating is relatively more correct. For funds without parenthesis behind means both ratings make sense comparing with their own sample mean. For instance, one year return for fund is , which is smaller than the Standard & Poor s sample mean , the rating should be less than two. The table shows that Standard & Poor s rating for this fund is one, which is reasonable. Meanwhile, the Wassum sample mean is , which is larger than one year return So Wassum rating of four is also reasonable in this case. Out of these seven funds, two of them show the better predictive 20

22 ability with Wassum ratings and rest five shows the better predictive ability with Standard & Poor s. Table 29. Year year return 2year return 3year return Fund S&P W S&P mean W mean S&P mean W mean S&P mean W mean Number (w) (sp) (sp) (sp) (sp) (sp) (sp) (w) (w) (w) (sp) (sp) (sp) (sp) In year 2009, out of 44 funds with both ratings, 12 are quite different. We compare sample mean and fund return to check the correctness of both ratings as mentioned before. For funds with an x written in parenthesis means neither ratings make sense comparing with their own sample mean. For fund , the two year return is equal to , which is slightly larger than two years Standard & Poor s sample mean , the rating should be therefore at least two. However then given rating is actually one. Meanwhile, this two years return is smaller than the Wassum sample mean , which should lead to a rating less than three. This is also not true since the given Wassum rating is four. From table 30, we conclude that despite the five funds with missing returns, three of them show the better predictive ability with Wassum ratings and rest four shows the better predictive ability with Standard & Poor s. Table 30. Year year return 2year return Fund S&P W S&P mean W mean S&P mean W mean Number (sp) (sp) NA NA (sp) (sp) NA (sp) (w) NA NA NA NA NA (sp) (x) (w) (w) 21

23 (w) (w) NA NA NA NA In year 2010, out of 41 funds with both ratings, 11 are rated differently. We keep using the same method to do the comparison. Table 31 shows either both ratings are reasonable or both ratings are unreasonable. We cannot conclude which company provides the better rating by studying these double rated funds. Table 31. Year year return Fund S&P W S&P mean W mean Number NA (x) VI. Conclusion Pension Agency imports Standard & Poor s and Wassum ratings as reference for pensioner when selecting their pension funds. The goal of this study is to answer the question of whether there ratings display qualifies information. Base on four years data from 2007 to 2010, we calculate the one year to four years out-of-sample future returns. The results given by different empirical tests are highly insignificant for both ratings, and even for the control Morningstar rating. In general, we cannot conclude there is any predictive performance for different rating systems base on whole sample, rating systems base on fund types and rating system base on categories. For few of those significant test results, the predictive performances are mostly incorrect. We can either see the shift in performance as time period 22

24 changes. In the case study of funds with both ratings, Standard & Poor s rating and Wassum rating behave differently through years. One speculation about why these rating behave so badly under the test period could due to the recent financial crisis. Global economic recession affects the normal operation of the funds and we can see the dramatic declining of funds returns. This may also have negative effect on the rating system. After all we are predicting future performance base on the historical information. When this historical pattern has been dramatically changed, how can the future performance be predictive? 23

25 Acknowledgements I would like first thank my supervisor Bo Larsson for his warm and patient guidance. I would also like to thank Morningstar Sweden for providing the support on Morningstar rating dataset. I am particular grateful for the help from Josefine Lanze during the requesting process and Karl Marthon for delivering the dataset in short notice. Reference Blake, C. R. and M. R., Morey, 2000, Morningstar Ratings and Mutual Fund Performance, Journal of Financial and Quantitative Analysis, Vol. 35, Capon, Noel, Gavan J. Fitzsimons and Russ Alan Prince, 1996, An individual level analysis of the mutual fund investment decision, Journal of Financial Services Research, Volume 10, Number 1, 59-82, DOI: /BF Del Guercio, Diane and Paula A. Tkac, 2002, The Determinants of the Flow of Funds of Managed Portfolios: Mutual Funds vs. Pension Funds, Journal of Financial and Quantitative Analysis 37, Kräussl, Roman and Ralph Sandelowsky, 2007, The Predictive Performance of Morningstar s Mutual Fund Ratings, Working Paper, VU University Amsterdam. Morningstar, 2007, The Morningstar Rating Methodology, Morningstar Methodology Paper, July 26, Morningstar, 2010, Category Definitions, Europe and Asia Morningstar Methodology Paper, November Patel, J., R. Zeckhauser and D. Hendricks, 1994, Investment flows and performance: Evidence from mutual funds, cross-border investments and new issues, Japan, Europe and the International Financial Markets: Analytical and Empirical Perspectives, Chapter 4, Cambridge University Press, 1996 Sirri, Erik R., and Peter Tufano, 1998, Costly search and mutual fund flows, The Journal of Finance, Vol. LIII, No.5, October