國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
4.3 Regression Results and Findings
Table 5 to Table 8 summarize the associated t-statistics for brokerage commission, transaction fee, custodian fee, turnover dummy and lagged fund performance for investor’s buy and sell over January 2010 to December 2014.
Comparing the two models, we find consistent evidences for the predictive power of brokerage commission, transaction fee, turnover dummy and past fund performance while custodian fee is arguably inconsistent. Another disputable variable is the management fee but based on the fund industry practice, we would support the results of panel fixed effect model. We show our empirical regression results below and look into each independent variable by analysis, survey, and interview.
Brokerage Commission
For our hypotheses H1-1.1 and H1-1.2, our result shows positive relation between brokerage fee and fund flows, either fund inflows or outflows, with p<0.000 as exhibited in Table 5 to Table 8. One simple explanation is related to the trading frequency of buying and selling more securities. Funds incur commissions as the results of inflow and outflow to them (liquidity trades). This supports the results of Rabarison (2015) that commissions paid by mutual fund managers are strongly influenced by investor flows. In the same vein, Lückoff (2011) shows daily gross inflows and gross outflows can only balance and offset each on the same day without incurring more costs. This is in line with the results of Chordia (1996) that fund investors trade-off diversified equity positions by investing mutual funds but absorb these direct costs for liquidity. Therefore, funds with more volatile daily flows tend to experience higher brokerage fees unless the fund size is big enough to meet clientele’s liquidity.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
In practice for Taiwan domiciled funds, some fund companies design
‘Short-Term Trading rules’ stating the required holding period to discourage and punish excessive or short-term trading and to charge investor’s extra fees for such cases. This is also to avoid that the fees are born equally by all fund investors. In comparison, most offshore funds charge the short-term trading fees in the same way;
however, offshore funds also use ‘Anti Dilution Levy’ or ‘Swinging Single Pricing’ in face of huge inflows or outflows on the day. But this practice is seldom taken by Taiwan domiciled funds.
Transaction Fee
The hypotheses tested (H1-2.1 and H1-2.2) based on predictions is that transaction fee is negative to investor’s Buy but positive to investor’s Sell. For our empirical results in response to it, investor’s buy is sensitively negative to transaction related fees (p<0.000) in Table 5 and Table 7 while investor’s sell is not significant in face of transaction fees by the results of multiple regression (p = 0.081) in Table 6 but it is significantly positive by the results of the panel fixed effect model (p<0.000) in Table 8. The related study is done by Bergstresser and Poterba (2002), who find that gross inflow is sensitive to tax burden. We therefore think that transaction related fees on mutual funds would finally reflect on and affect the fund price and then investors buy and sell. Lang (2014) highlights the main focus should lie on the tax rate charged to a fund company, irrespective of any transaction costs, since in face of higher tax costs, fund companies are likely to increase the charge of the fees on funds so that fund companies can keep their profit margin. This suggests a positive relation between fees and taxes. Based on this, we argue transaction fee and investor’s Buy are expected to be related negative while transaction fee and investor’s Sell are expected to be
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
positive.
Management Fee
In response to our hypotheses H1-3.1 and H1-3.2, there’s a strongly positive relation between management fee and investor’s Buy, with p<0.000 in Table 5 and Table 7. As for investor’s sell, management fee shows seemingly conflicting results by the multiple regression (positive) in Table 6 and the panel fixed effect model (negative) in Table 8. However, we cast doubts on the positive relation between management fee and investor’s Sell from the multiple regression result. We would conclude the positive relation between management fee and investor’s buy while negative relation between management fee and investor’s sell as following explanations. Management fee represents one of the main revenues for fund companies, which accounts for roughly 56% of the total expense in the mutual fund based on our sample data. Our results point to three possible explanations for why mutual funds with significantly higher management fees are positive to investor’s inflows: fund product signal, revenue-sharing mechanism, and performance-related compensation.
Fund product signal reflects significant research costs, more monitoring efforts, better mutual fund managers’ stock-picking abilities, or superior fund strategy. For this reason, we find the evidences from some funds with higher management fee expenses disclosed in the fund’s prospectus/factsheet fee table. For examples, Nomura Middle East & Africa Fund charges 2% management fee annually and explains that the political climate in some countries is unstable and research/ monitoring is not easily achievable. Hence, more research resources and greater monitoring efforts are required. JPMorgan (Taiwan) Japan Brilliance Fund also charges 2% management fee annually, highlighting the absolute return strategy, even when share markets face
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
volatile, flat or falling situations. Still, Allianz Global Investors Global Resources Trends claims the better bottom-up skills with 2% management fee annually.
The revenue-sharing mechanism is related to the distribution channels. The most common form is trailing commissions, which is the embedded compensation for the distribution channels and it is from the management fees. Trailing commission is one of the incentives for financial consultants, who do efforts to evaluate various fund products and assist clients in selecting the funds that meet clients’ needs. By the revenue-sharing, fund companies pay the distribution channels for marketing and promoting support, training and development, consulting & advisory services, and selling funds. Revenue-sharing is based on the total sales of the total client assets in the fund products and this incentive is calculated by asset under management (AuM) or fund inflows. Therefore, the more asset under management and/or fund inflows come, the more revenue-sharing generates.
The performance-related compensation is usually linked to the fund size. The rationale for the performance-related compensation is that it aligns the interests of both fund managers and fund investors, and that it plays an incentive for fund managers to generate positive returns to attract more fund inflows. As long as the fund size reaches the threshold, the fund manager’s compensation would be marked-up, which works like progressive mechanism reflecting the actual fund inflows generated from exceeding a fund size (or asset under management). This kind of compensation usually means that investment funds are managed or consulted by third-party investment advisers. Hence, the bigger a fund size is, the more performance-related compensation wins. Our empirical evidence is also in line with the result of Berk and Green (2004), which demonstrates that mutual fund manager costs are an increasing
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
function of the amount of funds under management.
Based on the above three explanations – fund product signal, revenue-sharing mechanism, and performance-related compensation, while considering fund's individual-specific effect and time-specific effect, we argue that ongoing charges of management fee is positive to investor’s buy but negative to investor’s sell.
Custodian Fee
Our hypotheses tested (H1-2.1 and H1-2.2) seem to be arguably inconsistent as shown in Table 5 to Table 8. The independent custodian aims at the double mission of safekeeping the fund assets and making sure fund portfolios in compliance with the law. Ideally, this indirect cost should not have a direct effect on investor’s buy and sell as we hypothesize. However, the multiple regression shows the results that custodian fee is significantly negative to both investor’s buy and sell with p<0.000. On the flip side, the panel fixed effect model tells that custodian fee should be irrelevant to influence investor’s Buy (p=0.703) and Sell (p=0.708). Considering the practices in Taiwan fund market, we argue that, from the multiple regression, the evidence on custodian fee and investor’s buy and sell presents a somewhat nonstandard and unexpected portrait. Overall, our empirical analysis does not find consistent evidence that custodian fee matters for mutual fund investors.
Turnover Dummy
Our alternative hypotheses (H2-1 and H2-2) with p<0.000 document that turnover ratios of mutual funds have both negative effects on investor’s buy and sell in Taiwan domiciled equity fund market as evidenced in Table 5 to Table 8. This corresponds to the fact that investors prefer to purchase the funds with buy-and-hold
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
strategy with lower turnover ratios.
If we look into the turnover ratio, something interesting could be found that the yearly cumulative top 10 selling turnover ratios consist of the most mutual funds investing in Greater China or Taiwan markets as exhibited in Table 3. One explanation is familiarity effect or home bias. This is maybe because of language, culture, and distance so that fund managers prefer to trade familiar companies, which echoes the results of Grinblatt and Keloharju (2001) in Finnish markets and Huberman (2001) in the United States markets.
In order to have further checking for the turnover ratio on investor’s buy and sell, we extract the data with turnover ratio dummy marked ‘1’ and run the simple regression below for the analysis:
Performancet =α+ Selling Turnovert + e
The results show the t-statistics is -9.094 with p<0.000. This documents the evidence that, for the funds with turnover ratios above the market average, mutual funds have not been able to generate enough returns to compensate their costs.
However, despite the above negative results of funds with selling turnover affecting performance returns, there exists the possibility that a few funds may be skilled enough to consistently perform well so their turnover costs are compensated as highlighted by Rubio (1995).
Past Fund Performance
The alternative hypotheses (H3-1 and H3-2) state that mutual fund past performance does stimulate mutual fund investor's buy and sell. Table 5 to Table 8
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
illustrate the t-statistics associated with the examination of mutual fund past performance explaining the investor’s subscription and redemption, which indicates a significant result with p<0.000.
This can be explained by behavioral finance: one is representative heuristic, another is disposition effect, and still another is prospect theory proposed by Kahneman and Tversky (1979), which proposes that investors choose a reference point to evaluate the performance of their investments and exhibit different investment behaviors when facing investment gains and losses.
The representative heuristic leads mutual fund investors to buy past winners with high fund performance returns. When making decision, fund investors don't always have the time or resources to compare and research all the information before buy and sell a mutual fund, so investors could use heuristics to grab and reach decisions quickly. This is significantly evident when we use lag fund performance (t-1) to predict fund investor’s the monthly buy and sell (t). Disposition effect could also determine investors’ decisions when they have tendency to hold losers too long and sell winners too soon. Barberis and Xiong (2009) explain if an investor is holding a stock with gains in value since purchase, he might think of trading the stock as a gain.
If he is risk-averse over gains, he may then be inclined to sell the stock (sooner).
Similarly, if he is risk-seeking over losses, he may be inclined to hold on to it when the value going down (longer). In our empirical analysis, we could deduce the similar investor’s behavior in face of past fund performance returns. The similar findings are from Shu, Yeh, Chiu, and Chen (2005), which shows that Taiwanese investors exhibit the stronger disposition effect (the ratio of realizing gains to losses is 2.5 times) than US investors (the ratio is 1.5 times) (Odean, 1998).
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Another explanation is the flow-performance relationship documented by Ellison and Chevalier (1997) and Sirri and Tufano (1998). It highlights the asymmetric fund flow-performance relationship in nature since fund investors rush into funds with lagged high performance, while tend to stay in funds that has performed poorly. Similarly, Clifford, Fulkerson, Jordan and Waldman (2011) examining U.S. actively managed equity mutual funds do find both retail and institutional investor inflows and outflows strongly chase past total returns over the period 1996 to 2009. In light of our flow performance, these patterns perhaps are not surprising since investors chase performance when purchasing funds. It's therefore evident for Taiwan domiciled equity fund flows sensitive to past fund performance over the period 2010 to 2014. Besides, our findings are consistent with several empirical studies (Patel, Zeckhauser, and Hendricks, 1994; Gruber, 1996; Sirri and Tufano, 1998) in the U.S. market that fund flows into and out of U.S. domestic mutual funds are related to past performance.
Together with these studies, our results conclude that, from fund investor’s perspective, past performance might be an indicative of future performance for investor’s buy and sell as noted.
Control Variables
In the multiple regression models, the analysis introduces five determinative macro factors as control variables affecting investor’s buy and sell. As stated in the section ‘3.2.2 Control Variables,’ these include M1B, CPI change rate (yoy), commercial paper interest rate, Taiwan 10-year bond yield, and TWSE. These five macro-economic factors in the models aim to capture the effects of business cycle in
‧
observe that most of the factors are significant but some of them are not logical. We believe these control variables still take some real effects so the inclusion does not cause inefficiency. However, on the other hand, some explanations can be caused by other reasons. This is why our findings support the results from panel fixed-effect model since it consider the both individual and time specific effects.Table 5 The results of the multiple regression model
Dependent Variable = Buy (corresponding to Model A)
Variables Coefficient Std Error T-statistics P-value VIF
Intercept -0.151 0.076 -1.994 0.046
Brokerage Commission 33.887 1.802 18.804 0.000 2.420
Transaction Fee -13.245 1.644 -8.056 0.000 2.548
‧
Table 6 The results of the multiple regression model
Dependent Variable = Sell (corresponding to Model C)
Variables Coefficient Std Error T-statistics P-value VIF
Intercept -0.972 0.067 -14.468 0.000
Brokerage Commission 41.199 1.596 25.811 0.000 2.420
Transaction Fee 2.542 1.456 1.746 0.081 2.548
Table 7 The results of the two-way fixed effect model
Dependent Variable = Buy (corresponding to Model B)
Variables Coefficient Std Error T-statistics P-value
Intercept -0.006 0.014 -0.43 0.669
Brokerage Commission 43.021 2.192 19.63 0.000
Transaction Fee -30.716 2.053 -14.96 0.000
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Table 8 The results of the two-way fixed effect model
Dependent Variable = Sell (corresponding to Model D)
Variables Coefficient Std Error T-statistics P-value
Intercept 0.064 0.011 5.55 0.000
Brokerage Commission 33.021 1.818 18.17 0.000
Transaction Fee 8.095 1.703 4.75 0.000
Management Fee -12.266 3.188 -3.85 0.000
Custodian Fee 9.620 25.690 0.37 0.708
Turnover Dummy -0.023 0.001 -21.44 0.000
Lag Performance 0.010 0.004 24.83 0.000
Adjusted R2 35.95%
F Prob > F
17.84 0.000
‧
國立 政 治 大 學