Before we examine our hypothesis with multiple regression, we need to present numbers, means, minimum, maximum and standard deviation for each variable (Table 1).
Besides, we also examine Industry tenure, Dummy of Market competition, fund size, and fund age by Pearson Correlation Coefficients to see if these variables have highly correlation with each other and influence the outcome of the regression (Table 2). Thus we can trace each variable and to see if all can work well in our multiple model.
In Table1, H(s,it)here is fund level herding, it is influenced by indicator I which designates out that if the fund is close to crowd or against the crowd. All the variables is set quarterly and we use independent variables in quarter (t-1) to examine herding fund level measure in quarter t. Especially manager tenure here is the original value by counting the difference between manager’s fist date when he is under management in the industry and each fund’s first record date. We set young manager and old manager variable are dummy, when manager falls in the tenure rank of the first 10%, he/she is taken as young manger;
while the manager falls in the tenure rank of the last 10%, he/ she is taken as old manager.
We further examine how managers interacts with herding measure if they possess career concerns.
However, data of manager’s industry tenure is apparently less than herding measure sample.
The reason is that original data of manager management reporting date is not available on each fund from CRSP database. Further, while processing manager’s industry tenure, there is a selection bias: we intentionally select industry tenure from fund industry, so we neglect other possibilities in other industries. Therefore we can only took these data are missing value. This might influenced our testing outcome, since there is less than over 80,000 missing value in manager’s industry tenure. Other control variables are proven to be
27
significant variables which influence herding measure in some past literature review. Here we control these variables in the model in case of influencing our main independent variables.
In Table 2, the correlation coefficients between each variable is not highly correlated.
The correlation coefficient for fund size and fund age falls between 0.25769, which is positively weakly correlated. Therefore, we can further examine variables
4-1 Multiple regression: Industry tenure and fund tenure without career concern
In this section, we examine Industry tenure and Fund tenure separately for each fund into multiple regression while not considering career concern. We expect to see even without career concern, young managers still tend to herd. In next section, we compare industry tenure and fund tenure without career concern to those with career concern to see if career concern works strongly for young managers.
Table 3-a model 1 isindustry tenure and market competition predicting herding behavior;
while model 2 is fund tenure and market competition predicting herding behavior. Both model don’t consider career concern intersection. Meanwhile, both model put the same market competition variable and other control variables as stated on formula (10). Industry tenure counts each fund manager’s tenure under his first starting date in the management industry. We trace this date on which manager starts to manage his first fund in the industry.
If there are co-workers in the same fund, we will separate and count each fund manager’s industry tenure and depends on one whose tenure is longest. If the fund is managed by a team, we count the fund’s entire reporting survival period as its independent manager’s tenure.
On the other hand, fund tenure is calculated under each fund and each manager. If fund changes new manager, it means next manager is young to this fund. Still, if the fund is
co-28
managed by more than one manager, we calculate each manager’s total tenure in the same fund and pick the longest one as this fund’s manager tenure. When we set up these two model, we expect to see young manager without career concern still tend to take herd and when market competition becomes fiercer, funds under management would rather herd as well.
In model 1, we find out career concern-dominated market has negative relationship with fund level herding. It shows career concern-dominated market would rather not to herd, which is against to our hypothesis. However, young industry tenure manager dummy variable has significantly positive relationship with fund level herding. It shows young industry tenure managers without career concern intersected have 0.000557 effect on fund level herding. It means young industry tenure managers would more likely to take herd.
However the estimate is not significant. On the other hand, when we compare it to young fund tenure managers, it shows young fund tenure managers without career concern intersected have -0.00026992 effect on fund level herding. It’s negatively correlated. We preliminarily think it’s because of managers with fund tenure are too short to determine the herding effect. Besides, managers with shorter fund tenure maybe possess longer industry tenure. We need to further examine: does shorter industry tenure managers affects herding behavior much more than shorter fund tenure managers. Therefore, what it show in model 2 is not what we expect to see.
When it comes to fund’s market competition, the estimate is 0.00436 and at 1%
significant level. It means when fund’s market competition rise 1, it would rather take herd on 0.00436, which is support our hypothesis. The idea of fund’s market completion is resulted from Roberts (1999), who provided the concept that product innovation can ensure sustained high profitability. That’s why we cited this idea and examine the same idea in mutual fund industry. Roberts (1999) mentioned that when time passes, the profitability
29
from product innovation will decrease eventually because other competitors will take the same strategy as the first mover and take first mover’s market share away. The measure in our research is market share in current period, comparing to its maximum market share in the past.
However, this research doesn’t tell if this fund’s maximum market share presents the fund is the first mover. According to informational cascade research, herding behavior easily happens when there is an apparent successful example to imitate. Everyone wants to obtain information quickly and immediately before the profitability of this information runs away. If research later on can deal with market competition with detail to determine which fund is actually the first mover, the outcome here would be more convincing.
Other control variables such as fund size, fund flow fund return, fund volatility, fund turnover rate and 12b-1 fee are significantly negative related with herding measure respectively, while other variable is controlled.
30
4-2 Multiple regression: Industry tenure and fund tenure with career concern
This section, we will show industry tenure with career concern and fund tenure with career concern to see if career concern works strongly for young managers.
Table 3-b also presents two multiple regression models. Model 1 is industry tenure and market competition predicting herding behavior; while model 2 is fund tenure and market competition predicting herding behavior. Both model consider career concern intersection.
These two model represents that when young industry tenure managers with career concerns, they tend to herd; while young fund tenure managers with career concerns are negatively with fund level herding. It shows young industry tenure managers with career concern intersected have 0.00237 effect on fund level herding. On the other hand, when we compare it to young fund tenure managers, it shows young fund tenure managers without career concern intersected have -0.00271 effect on fund level herding. Both of beta are important in 5% significant level. To young industry tenure managers, when they face bear market, their career concern arise. To avoid being laid off, they would tend to herd. It is accorded with our hypothesis 1. However, to young fund tenure managers, it’s against our hypothesis 1. The reason we try to put together and figure out is that young fund tenure managers might possess long industry tenure already. This should be considered while in later research. Furthermore, because of missing value in tenure is too much to be ignored, it might be a selection bias. The multiple regression model we use here should be revised by Heckman selection model. It’s a two stage model. We have to find out the selection bias and set first stage to revise the probability of missing value. On the second stage, a new model comes out with this probability and run a regression.
When it comes to fund’s market competition, the estimate in both model is 0.00436 and at 1% significant level. It means when fund’s market competition rise 1, it would rather take herd on 0.00436, which is still support our hypothesis 2. Therefore, we can tell, the
31
market competition is a good and new idea to examine the fund’s herding effect. However, this variable still has a flaw: it doesn’t determine which fund is a first mover. We suggest research later on should revise this problem and make the outcome more convincing.
32