Dmitrii V. Kandaurov Influence of return and risk indicators on open-end mutual fund flows. 1) Introduction As often mutual fund manager income depends on the value of assets under management, manager primarily interested in fund growth. In order to attract individual investors and their money managers try to offer them the most wanted product (portfolio with optimal balance of risk and return). There are two main directions in contemporary researches dedicated to the problem of portfolio optimization: portfolio optimization based on the advanced modern portfolio theory (e.g. Campbell, Huisman, Koedijk, 2001) or direct maximization of the utility function (e.g. Lan, Wang, Yang, 2013). Thus, different risk measures, from variance to spectral and coherent measures, are used in studies of both directions. Many researches confirm the relatively high efficiency of value at risk, as well as coherent and spectral risk measures for a portfolio optimization (e.g. Adam, Houkari, Laurent 2001; Alexander, Baptista, 2002). The relationship between the inflow of funds for the performance and profitability and risk fund was shown by many researchers. One of the first and most comprehensive studies in this area is the paper of Chevalier and Ellison (1997). Authors confirm the non-linear nature of the flow-performance relation. This nonlinearity creates risk incentives, which dispose fund manager to actively operate risk. In the paper of Sirri and Tufano ( 1998) asymmetric disposition of individual investors to invest in funds, that very well performed previously explained by the existent of individual investor’s "search costs", that are compensated by the successful fund manager in exchange for the investments through the additional advertising. In this paper, the character of the flow-performance relation for Russian open-end mutual funds is studied. A model analogous to (Chevalier, Ellison, 1997) is used. We also consider the flow-risk performance for different risk measures. 2) Flow-performance relation a) Model To analyze the flow-performance relationship for Russian open-end mutual funds two different models for return and risk performance were considered. The first model (flowreturn model) is close to the semiparametric model, considered in (Chevalier, Ellison, 1997): πΉπππ€π,π‘+1 = ∑ πΎπ π΄πππππ‘ π(πππ‘ − πππ‘ ) + ∑ πΏπ π΄πππππ‘ π π +πΌ1 (ππ,π‘−1 − ππ,π‘−1 ) + πΌ2 (ππ,π‘+1 − ππ,π‘+1 ) + πΌ3 πΌπΊπ‘+1 + πΌ4 πππ,π‘+1 , where πΉπππ€π,π‘+1 – net fund π flow in period π‘ + 1, (1) ππ,π‘ – fund π return in period π‘, ππ,π‘ – broad-base index (micex) return in period π‘ , πΌπΊπ‘+1 – open-end mutual funds industry grow in period π‘ + 1, πππ,π‘+1 – fund i market share in the beginning of period π‘ + 1, π΄ππππ,π‘ - dummy variable, characterize age category of the fund. π΄ππππ,π‘ 1, 2, ={ 3, 4, ππ ππ’ππ πππ ≤ 2 π¦ππππ ππ 4 ≥ ππ’ππ πππ > 2 π¦ππππ ππ 6 ≥ ππ’ππ πππ > 4 π¦ππππ ππ ππ’πππππ > 6 π¦ππππ Under the net fund flow we assume relative fund growth ratio, that is πΉπππ€π,π‘+1 = ππ΄ππ,π‘+1 −ππ΄ππ,π‘ ππ΄ππ,π‘ − ππ,π‘+1 , here ππ΄π is net assets value. The attractiveness of this model lies in the fact that the usage of the coefficients πΎπ and πΏπ allows us to compare the degree of the elasticity of the flow-performance relationship for different fund age categories. Estimation of the model (1) parameters is carried out in three stages. On the first stage consistent estimates of πΌ’s were obtained by nonparametric1 partialing-out procedure, described in (Robinson, 1988): obtaining residuals from nonparametric regression of πΉπππ€π,π‘+1 and ππ,π‘ on ππ,π‘ − πππ‘ , then regress the residuals on the residuals. On the next stage πΎ and πΏ estimates were derived as follows. First of all nonparametric estimation of the function πΜπ = πΎπ π + πΏπ were obtained by kernel regression of πΉπππ€π,π‘+1 − πΌΜππ,π‘ on ππ,π‘ − ππ,π‘ (πΜ0 obtained similar to πΜπ when dummy variable is omitted). Then πΎ and πΏ obtained by simple regression of πΜπ on πΜ0 . On the last stage we obtain an estimate of π by kernel2 regression of πΉπππ€π,π‘+1 − ∑π πΏΜπ π΄ππππ,π‘ − πΌΜπ π¦Μπ‘ ≡ 1 + ∑π πΎΜπ π΄ππππ,π‘ on (ππ,π‘ − ππ,π‘ ). The second model used in this section simulates the relationship between risk and fund flows in the future. 1 All nonparametric regressions computed with optimal bandwidth selection, described in Li and Racine (2013) 2 Gaussian kernel was used in all nonparametric regressions. πΉπππ€π,π‘+1 = πΎβπ ππ,π‘ + πΌ1 ππΉπ + πΌ2 πππ,π‘+1 + πΏ , where βπ ππ,π‘ – fund π relative risk measure change in period π‘, ππΉπ – fund π management fee. βπ ππ,π‘ = π ππ,π‘ −π ππ,π‘−1 π ππ,π‘−1 (2) , here As risk measures (π ππ,π‘ ) we consider empirical variance, semi-variance, value at risk and expected shortfall. As the period for the assessment of risk, we chose a two-year interval in order to obtain more robust estimates. b) Data The flow-return model is estimated on the set of 383 open-end mutual funds in the period between 1999 and 2015. We use quarterly flow-return data. There are several “pro-s” to use shorter time interval, than in (Chevalier, Ellison, 1997) 3. First, in Russia individual investors, for the most part, oriented on the shorter investment horizon. Second, managers respond to risk incentives more often than once a year. Funds, oriented on institutional investor (fund with minimum initial purchase of 200 000 rubles), were excluded from the analytical set. Funds that merged with other funds also were eliminated. Finally we obtain 4800 data points to estimate model (1). The flow-risk model (2) was estimated on the same set of 383 open-end mutual funds. As we need to estimate empirical risk measures (on 2 years intervals) in model 2, we consider only funds of age 3 and greater. Therefore the amount of applicable data points is substantially lower than for the first model (only 1 102 fund-years). c) Results Flow – return model is estimated separately for «young» (of age less 4 years) and «old» (of age greater than 4 years) funds. Figure 1 presents non-parametric regression results of function π obtained from the young funds and it’s 90% confidence bands. It should be noted that flow-return relation on the 1st graph is normalized (it is cleared from the effect of mutual fund industry growth, and other parameters under πΌ coefficients). 3 They use annual data. Year t+1 net flow 1 0.8 0.6 0.4 0.2 0 -0.4 -0.3 -0.2 -0.1 -0.2 0 -0.4 0.1 0.2 0.3 0.4 Year t excess return -0.6 -0.8 -1 Figure 1 – Flow-performance relationship for funds of age<=2 πΜ with 90% confidence bands. Graph on the figure 1 may be interpreted as presented the expected growth rate in year π‘ + 1 for young funds as a function of their excess return in year π‘. For example, such a fund would be expected to grow by approximately 10 percent per quarter if it demonstrate no exceed return. If it’s return exceed market by 10% expected growth rate is near 20%. If we take a look on the left side of the 1st graph we will see that at extremely negative returns fund contraction accelerated inversely large excess return lead to the acceleration of fund inflows. If in the last quarter excess return of the 2-year old fund was near 25% than manager of the fund should slightly increase risk in the following quarter. In a good position fund will receive extra inflow, in the bad situation the inflow rate will not change substantially. Fund manager should reduce the risk of the fund if fund flows excessively sensitive to small negative changes in excess return. If we consider funds of age 6 and greater years (Figure 2), it is not If we take a look on the flow-return relation of «old» funds (Figure 2) it is become obvious that the «π» - curve is much flatter for them than for young funds, especially in the part of negative returns. It goes to show that individual investors have more trust in older funds. Their clients believe that in spite of bad quarter the fund will stand and next quarter will be batter. Year t+1 net flow 1 0.8 0.6 0.4 0.2 0 -0.4 -0.3 -0.2 -0.1 -0.2 0 0.1 -0.4 0.2 0.3 0.4 Year t excess return -0.6 -0.8 Figure 2 - Flow-performance relationship for old funds (age>6) πΜ with 90% confidence bands. Model (1) estimated parameters presented in the table 1. To interpret table 1 correctly we should remember that πΎ and πΏ coefficients had relative nature. They characterize the relation between nonparametric parts of the model (1) for different age categories. For the first and fourth age category πΎ = 1 and πΏ = 0 because we choose first age category as a benchmark for the second and 4th as a benchmark for the 3rd. To obtain πΜ for the second age category we should multiply πΜ for the first age category (figure 1) by (1-0.29) and then subtract 0.04 from that product. Negative πΎ2−4 in couple with positive πΎ4−6 indicate that the older fund flows are less sensitive to the past return performance. Table1 – Coefficients from flow-return model Independent variables πΎ2−4 πΏ2−4 Young funds (age <=4) N = 1 343 -0.29 (0.33)* -0.04 (0.15) πΎ4−6 πΏ4−6 πΌ1 πΌ2 πΌ3 0.82 (0.29) 1.53 (0.20) 1.21 (0.10) Old funds (age > 4) N = 3 459 0.33 (0.59) 0.02 (0.35) 0.44 (0.11) 1.01 (0.35) 0.83 (0.06) πΌ4 -0.18 (0.07) -0.08 (0.04) Note – estimated standard errors are in parentheses If we take a look on the πΌ1 and πΌ2 coefficients we could see that excess return in π‘ + 1 and π‘ − 1 periods also have influence on fund flows in π‘ + 1 period. In table 2 we had flow-risk regression results. As it seen from the table the most strong Table 2 – Coefficients from flow-risk model. Risk measure Variance (empirical) Semi-variance (empirical) Value at Risk (empirical, πΌ = 0,25) Expected shortfall (empirical, πΌ = 0,25) πΎ -0,9* πΏ 0.23* πΌ1 1.24* πΌ2 0.21** -1.44** 0.05* 1.15 0.34 -0.81 0.14 0.95 0.18 -1.19* -0.02** 0.77 0.17 Note: * significant at the 0.1 level ** significant at the 0.15 level It is seen from the table that all coefficients 3) Dose Russian fund managers actually manipulate risk in response to incentives? In the previous section it was shown that the relationship between the yield of an openend mutual fund and fund flows is non-linear. This nonlinearity causes incentives to manipulate risk for fund manager. a) Risk incentives estimation To estimate the risk incentives for fund managers we use the same approach as in (Chevallier, Ellison, 1997). The only difference – is the shorter time interval (quarter instead of the year). According to (1): πΈ[πΉπππ€π,π‘ ] = πΈ[(1 + πΎπ )π (π2π + π’) + πΏπ + πΌππ,π‘ ], where 1+ π2π - fund’s excess return in the first two months of the quarter; π’ – random variable representing fund’s monthly return. Assuming that π’ distributed with mean zero and standard deviation π, and another variable π£, has mean zero and st. dev. π + βπ we can write risk incentives as follow: π πΌπ (ππ(π‘) ; π, βπ) = πΈ[(1 + πΎπ )[π(ππ‘ + π£ ) − π(ππ‘ + π’)]] (3) Risk incentive graph for funds of th 1st age category is presented in the figure 3. The function is interpreted as the expected increse in the 3rd month growth rate of a 1st age category fund that results from increasing its risk in the 3rd month of the quarter from the sample average to 50% above this average. Risk Incentive 0.025 0.02 0.015 0.01 0.005 -0.4 -0.3 -0.2 0 -0.1 0 -0.005 0.1 0.2 0.3 0.4 2 month return -0.01 -0.015 Figure 3 – Risk incentives for «young» mutual funds. b) Model and data To examine fund managers reaction to risk incentives the following model is used βπΏπ,π‘+1 = πΌπ πΌπ,π‘ + π½πΉπππ€π,π‘ + πΎπππ , where (4) π πΌπ,π‘ - risk incentive at the end of the period π‘, measured from (3), βπΏπ,π‘+1 = πΏπ,π‘+1 −πΏπ,π‘ πΏπ,π‘ – change in std. dev. in the period π‘ + 1. To estimate this «risk-flow» model we use quarter data of fund flows and risk parameters (4800 observations). Monthly risk indicators (funds π½ and volatility) obtained from InvestFunds (http://pif.investfunds.ru/analitics/coefficients/). Fund flows quarter data obtained from nlu.ru (http://www.nlu.ru/export-excel.htm). The results of the regression (4) presented in the table 4. πΌ fund age <=2 4>=fund age >2 1.21* (0.64) 0.92** π½ 0.05 (0.10) 0.60 πΎ 0.12 (0.08) 0.02 6>=Fund age>4 Fund age>6 (0.53) 0.88** (0.21) 0.21** (0.19) (0.48) 0.74 (0.32) 0.51 (0.47) (0.04) 0.03 (0.00) 0.40 (0.34) Note: * significant at the 0.1 level ** significant at the 0.15 level Model (4) parameters estimates suggest that young funds managers pay more attention to risk incentives than the older ones. These results are not surprising, young funds managers often 4) Conclusion From this study we could derivate several basic results. First of all, the dependence between past performance of the fund and its net flow in the next period obviously exists. This relation has non-linear character, as a result risk incentives for fund managers appears. Flow-performance relation as well as risk incentives are stronger for the young funds. Young funds reaction to the risk incentives is much stronger than the reaction of the older funds. The flow-risk performance is most stronger for expected shortfall and semi variance risk measures. References G. Alexander, A. Baptista “Economic implications of using mean-VaR model for portfolio selection: A comparison with mean-variance analysis”, Journal of Economic Dynamic & Control, 26, 1159-1193. A. Adam, M. Houkari, J-P. Laurent “Spectral risk measures and portfolio selection”, Journal of Banking & Finance, 32, 1870-1882. E. Sirri, P. Tufano “Costly Search and Mutual Fund Flows”, The Journal of Finance, 5, 1589-1622. J. Chevalier, G. Ellison “Risk Taking by Mutual Funds as a Response to Incentives”. R. Campbell, R. Huisman, K. Koedijk, “Optimal portfolio selection in a Value-at-Risk framework”, Journal of Banking & Finance, 25 (2001), 1789-1804. Y. Lan, N. Wang, J. Yang (2013), “The economics of hedge funds”, Journal of Financial Economics, 110 (2013), 300-323. Li, Q. and J. Lin and J.S. Racine (2013), “Optimal Bandwidth Selection for Nonparametric Conditional Distribution and Quantile Functions”, Journal of Business and Economic Statistics, 31, 57-65.