1 Introduction

There is a wide array of asset pricing models that have been developed over the years to help gain insights into the pricing of assets traded in the financial market. These models identify the risk factors that are priced and thereby provide tools that guide the investor in making optimal investment decisions. However, majority of the asset pricing tests have not examined the risk factors that the investors consider before making investment decisions. The question of how investors perceive risk in financial market has received very little attention. This is mainly due to the fact that, for most of the assets, it is difficult to observe and measure investor’s reaction to various investment opportunities.

In this context, Berk and Van Binsbergen (2016) and Barber et al. (2016) propose an approach to gauge investor response. They assert that mutual fund flows can be used as a measure of investor response to identification of profitable investment opportunities. Though both the studies use different empirical technique, the basic premise is that the investors, being rational, would evaluate the performance of alternative mutual fund schemes and reward the best performing fund with additional fund flows. They argue that by assessing the sensitivity of flows to alternative risk factors or performance measures, one can conjecture the asset pricing model used by the investors. Both studies suggest that the Capital Asset Pricing Model (CAPM) performs the best in predicting the fund flows to the actively managed US equity mutual funds.

The finding that investors rely on the CAPM for evaluating mutual fund is intriguing mainly since the model is found to be inadequate in explaining the cross-sectional variation in returns. Subsequent studies that made attempts to resolve this puzzle also reached a similar conclusion. See for instance, Blocher and Molyboga (2017) and Agarwal et al. (2018) who examined the case for hedge fund investors and Dang et al. (2019) for investors in bond market. Interestingly, Ben-David et al. (2019), in contradiction to the previous studies, concludes that investors do not use any of the asset pricing models while investing in mutual funds. Rather, empirical evidence suggests that they have outsourced the risk adjustment to external entities like Morningstar (a fund rating agency).

The empirical literature that sheds light on the risk factors from the perspective of an investor is in a nascent stage. Moreover, most of the studies in the area are confined to the US market. US financial market and mutual fund industry are well developed with very little informational inefficiency. The case of underdeveloped economy may not be so. The capital market in most of the underdeveloped market are thin and underdeveloped (Rojas-Suarez, 2013). Though these markets exhibit significant growth opportunities, high political and economic risks make emerging markets more volatile than mature markets (De Santis & İmrohoroğlu, 1997). The development of mutual funds in the emerging markets has provided the masses with a means to participate in capital markets and has contributed towards the strengthening of securities and derivatives markets in these economies (Ong & Sy, 2004). Emerging markets, therefore, essentially provide alternative testing grounds for drawing inference about the asset pricing model preferred by investors.

Given this backdrop, we enquire whether Indian mutual fund investors exhibit a pattern similar to their counterparts in the developed markets when allocating funds to different mutual fund schemes. The Indian mutual fund market is quite old, history of which dates back to 1963. See Fig. 1 that traces the growth in asset under management (AUM) over the years. However, the industry is still in its growing stage, with ratio of AUM to GDP (measure of mutual fund penetration) around 13 per cent. This implies that the potential for growth in the mutual fund industry in India is very high. Moreover, the participation of individual investors in the Indian mutual fund industry has been showing an increasing trend. As of December 2019, more than half (around 54 per cent) of the industry assets are held by individual investors. It is largely the equity-oriented schemes that attract the individual investors. According to the July 2019 report on Industry Trends by Association of Mutual Funds in India (AMFI), the equity-oriented schemes generate 88 per cent of their assets from individual investors. All these factors motivate us to consider the Indian market as an ideal choice for evaluating the asset pricing model preferred by individual investors in emerging market.

Fig. 1
figure 1

Source: Association of Mutual Funds in India (AMFI)

Asset under Management (AUM).

Full size image

We examine the responsiveness of mutual fund flows to alternative performance measures to identify the risk factors that investors weigh in while making investment decision. We use six alternative performance measures. Three of these measures are based on widely used asset pricing models, i.e., Capital Asset Pricing Model (CAPM) (Lintner, 1965; Sharpe, 1964), Three-Factor Model (TFM) (Fama & French, 1993) and Four-Factor Model (FFM) (Carhart, 1997). The remaining three are based on naïve models wherein the performance of mutual fund is measured based on raw return, excess return (i.e., return in excess of the risk-free return) and market adjusted return (i.e., returns in excess of the market return). We analyze the flow–performance sensitivity following the seminal works of Berk and Van Binsbergen (2016) and Barber et al. (2016). Further, considering the argument that uncertainty in developing countries acts as a tax on investment, thereby discouraging the investors from making even profitable investment (Rodrik, 1991). We extend our analysis to take into account the impact of uncertainty on the flow–performance relation. We enquire whether uncertainty influences investors’ evaluation of alternative mutual funds. Baker et al. (2016) Economic Policy Uncertainty (EPU) Index for India is used to capture uncertainty.

Our study, based on a sample of 991 actively managed equity funds for the period that spans from the second quarter of 2006 to the last quarter of 2019, suggests that the mutual fund flows are highly sensitive to the performance measure obtained using naïve models. Both, Berk and Van Binsbergen (2016) approach and Barber et al. (2016) approach confirm that none of the factor models outperforms the naïve models in explaining flows. This implies that the mutual fund investors do not tend to adjust for any of the risk factors while evaluation alternative investment options and simply chase funds that generate high excess returns or market adjusted returns. In other words, investors do not account for the return generated by known risk factors while evaluating the performance of the funds. Our findings in this regard are contrary to that of earlier studies that suggest that investors use CAPM while evaluating alternative investment options and thus adjust for only market factor. Our enquiry into the influence of uncertainty on investors’ evaluation of alternative mutual funds suggests that uncertainty and mutual fund flows are negatively related. Surprisingly we find no significant impact of uncertainty on the flow–performance sensitivity.

Our study contributes to the existing literature in multiple ways. First, our study adds to the literature that uses flow–performance sensitivity to infer the asset pricing model or the risk factors that investors consider while making investment decisions. We extend the existing literature by examining the case of an emerging economy and present evidence contrary to those observed in developed markets. While previous studies suggest that mutual fund investors make use of CAPM for evaluating the performance of alternative funds, we find that naïve models outperform the factor models in explaining the fund flows.

Second, we examine the impact of uncertainty on the flow–performance relation and thereby take the literature forward. While there are numerous theoretical and experimental studies that suggest that investors are less willing to participate in financial market when uncertainty increases (see for instance Cao et al., 2005 and Bossaerts et al., 2010), empirical evidence for the same is rather limited. In this regard, Antoniou et al. (2015) and Li et al. (2017), using mutual fund flows, provide empirical evidence for ambiguity aversion among mutual fund investors. Adding to the existing literature, we examine whether investors use stringent measures for evaluating alternative mutual funds when faced with higher levels of uncertainty. Though, consistent with the existing literature, we document evidence for ambiguity aversion among mutual fund investors, we find that uncertainty does not induce investors to take into account risk factors while making investment decisions.

Last, we also add to the literature that seeks to identify the behavior of mutual fund investors. Empirical literature provides evidence that suggests majority of mutual fund investors are sophisticated and make informed decisions (see for instance Zheng, 1999; Gharghori et al., 2007; Chalmers et al., 2013 and Dyakov & Verbeek, 2019 among others). However, our results suggest that mutual fund investors fail to make informed investment decisions. The findings corroborate those of Capon et al. (1996), Barber et al. (2005), Cooper et al. (2005) and Hillenbrand and Schmelzer (2017) among others, who argue that mutual fund investors are naïve and subject to various behavioral biases.

The paper is organized as follows. Sect. 2 discusses the methodology and database we have used for our empirical analyses. Sect. 3 devotes on the discussion of our findings. Results of some further robustness checks are discussed in Sect. 4, and Sect. 5 concludes.

2 Data and methodology

2.1 Data

We have collected the information required for the present study from the Association of Mutual Funds in India (AMFI) website and the Fama French and Momentum Factors: Data Library for Indian Market. AMFI was incorporated on August 22, 1995, and all of the 44 Asset Management Companies that are registered with SEBI, are its members. One of its objectives is to disseminate information on mutual fund industry. We have collected information on scheme-wise AUM, net asset value (NAV) and scheme details from AMFI. Data on market return, risk free return, and risk factors specified by alternative asset pricing models (market, size, value and momentum factors) are collected from the Fama French and Momentum Factors: Data Library for Indian Market. Additionally, data on EPU for India is obtained from www.PolicyUncertainty.com maintained by Scott Baker, Nicholas Bloom and Steven J. Davis.

As is already noted, we draw inference about the asset pricing model that investors use while making investment decision by examining the relation between flow of funds and alternative performance measures of mutual funds. Since our main interest is to identify the asset pricing model that is used by investors, we limit our sample to actively managed equity mutual funds. Moreover, equity-oriented schemes generate 87% of their assets from individual investors, making it perfect vehicle to gain insights on investment decisions of investors. To be specific, we include in our sample only those mutual funds that are categorized as Equity Scheme, ELSS and Growth funds by AMFI. Further, mutual funds often offer several share classes with different combinations of expense ratios, management fees and reinvestment options. These are designed to attract investors with varying levels of wealth and investment horizons and are known to influence the investment and redemption decisions of investors. Thus, we follow Jiang and Yuksel (2017) and use individual fund share classes as our unit of observation.

2.1.1 Mutual fund flow

Our main variables of interest are mutual fund flows and performance. Fund flows are calculated using scheme-wise AUM, the data for which are available only from April 2006. Hence April 2006 is chosen as the beginning of our sample period, and it spans till December 2019. However, monthly AUM data is available on the AMFI website upto September 2010. For the remaining part of the sample period, AUM data is available only on a quarterly basis. In order not to lose information, we use AUM reported for months that coincide with the end of each quarter. Consequently, our analysis is based on quarterly data, and the sample period ranges from the second quarter of 2006 to last quarter of 2019.

Following the prior literature on fund flows, flows for fund p in quarter t is defined as the ratio of net flow into the fund to lagged AUM. Formally, flow is calculated as

$${F}_{pt}=\frac{{AUM}_{pt}}{{AUM}_{p,t-1}}-\left(1+{R}_{pt}\right),$$
(1)

where \({AUM}_{pt}\) is the asset under management of fund p at the end of quarter t, \({R}_{pt}\) is the return of fund p for the quarter t.

2.1.2 Mutual fund performance

The performance of mutual funds is evaluated by considering the popular asset pricing models and also three naïve models. To be specific, we consider three of the widely discussed asset pricing models, i.e., CAPM, TFM and FFM to measure the abnormal return generated by each of the funds. Quarterly returns, calculated from the NAVs, and the return to risk factors considered in alternative asset pricing models are used to estimate the performance measures. For instance, outperformance relative to the four-factor model is measure after taking into consideration the market, size, value, and momentum factors. For each fund p in quarter t, the following time-series regression is estimated for the period t-1 to t-20:

$$\left({R}_{p\tau }-{R}_{f\tau }\right)={{\alpha }_{pt}^{FFM}}+{\beta }_{pt}{(R}_{m\tau }- {R}_{f\tau })+{s}_{pt}{SMB}_{\tau }+{v}_{pt}{HML}_{\tau }+{m}_{pt}{WML}_{\tau }+{\varepsilon }_{p\tau },$$
(2)

where τ = t-1 to t-20,Footnote 1\({R}_{p\tau }\) is the return on mutual fund for the quarter τ, \({R}_{f\tau }\) is the risk-free return,\({R}_{m\tau }\) is the market return, \({SMB}_{\tau }\) is size factor, \({HML}_{\tau }\) is the value factor, \({WML}_{\tau }\) is the momentum factor. The estimated parameters \({\beta }_{pt}, {s}_{pt},{v}_{pt}\) and \({m}_{pt}\) represent the market, size, value and momentum tilts of fund p, respectively, and \({\alpha }_{pt}^{FFM}\) is the mean return generated by the fund that is unrelated to the risk factors considered by the FFM. The four-factor alpha for the fund in quarter t is calculated as

$$\hat{\alpha }_{pt}^{FFM} = \left( {R_{pt} - R_{ft} } \right) - \left[ { \hat{\beta }_{pt} (R_{mt} - R_{ft} } \right) + \hat{s}_{pt} SMB_{t} + \hat{v}_{pt} HML_{t} + \hat{m}_{pt} WML_{t} ].$$
(3)

This procedure is repeated so that we obtain a time-series of quarterly alpha for our sample Footnote 2 The three-factor alpha and the CAPM alpha is estimated following a similar procedure. To be specific, to obtain the three-factor alpha, we estimate the regression in Eq. (2) after dropping WML as an independent variable, and for CAPM alpha SMB, HML, and WML are dropped.

The naïve models measure the performance of fund based on the return generated by the fund not adjusted for any of the risk factors. The first naïve measure is simply the quarterly. The second measure is the excess return (ER), which is the return generated by a fund over the risk-free rate. The third naïve measure is market adjusted return (MAR), which is estimated as the difference between the fund return and market return.

Rational investors would evaluate the performance of a fund over a time horizon and hence for the purpose of our analysis we adjust our performance measure such that it reflects the performance of each mutual fund over the recent past four quarters. Thus, we compute the fund’s alphas as an average of prior four quarterly alphas:

$${ALPHA}_{pt}^{i}=\frac{1}{4}\sum_{j=t-1}^{t-4}{\widehat{\alpha }}_{pj}^{i},$$
(4)

where i stands the model based on which the performance measure is computed. A similar adjustment is made even in the case of our naïve performance measures.

The requirement of an estimation window of 20 quarters for arriving at our asset pricing models based performance measures, limits the period of our empirical analysis to begin from the second quarter of 2011. This requirement also ensures that the funds below the age of 5 are excluded from the sample, and thus incubation flows do not influence our results. Our final sample consists of an unbalanced panel of 991 mutual fund schemes.

Table 1 Panel A shows summary statistics mutual fund characteristics for the period of our empirical analysis (second quarter of 2011 to the last quarter of 2019). Average fund flow is about 2.7 per cent.Footnote 3 An average mutual funds scheme in our dataset is of age 12.49 years. The fund size (measured as the logarithm of AUM of the fund) of an average mutual fund is 8.872 and average size of an asset management company (AMC), measured as the logarithm of AUM of the AMC as a whole, is 14.863. Panel B of Table 1 reports the average factor exposure obtained from the FFM. Panel C provides the summary of the performance measures used in our study and Panel D the correlation between alternative measures. We observe that performance measures are highly correlated.

Table 1 Summary statistics
Full size table

2.2 Methodology

The primary interest of our study is to draw inference about the model that mutual fund investors use while making investment decisions. We first study the flow–performance relation for each model separately and then perform a pairwise test, which allows us to compare the relative performance of different models. The flow–performance relation for each model is examined following the methodology proposed by Berk and van Binsbergen (2016). They argue that it is possible to deduce which asset pricing model investors are using by investigating how well the signs of ALPHA match the directions of flows. For each asset pricing model i, the following regression is estimated.

$$sign\left({F}_{pt}\right)={\gamma }_{0}^{i}+{\gamma }_{1}^{i}sign\left({\alpha }_{pt}^{i}\right)+{\epsilon }_{pt},$$
(5)

where \({F}_{pt}\) is the fund flow to the mutual fund p in the quarter t, \({\alpha }_{pt}^{i}\) denotes the alpha computed from Eq. (4), sign is a function that returns the sign of a real number, say x, and takes the value 1 if x > 0, – 1 if x < 0 and 0 if x = 0. Coefficient \({\gamma }_{1}^{i}\) measures the sensitivity of flows to the performance measure. Intuitively, if investors find that a fund performs well (poorly) then they will invest in (withdraw investment from) that particular fund. Lemma (2) of Berk and van Binsbergen (2016) shows that a linear transformation of the regression slope \({(\gamma }_{1}^{i})\) helps in recovering the average probability that conditional on outperformance being positive (negative), the sign of fund flow will also be positive (negative). This allows us to ascertain the percentage of fund flows explained by the performance measure from each of the models. The linear transformation of \({\gamma }_{1}^{i}\) takes the form:

$$\frac{{\gamma }_{1}^{i}+1}{2}=\frac{Pr\left[sign\left({F}_{pt}\right)=1|sign\left({\alpha }_{pt}^{i}\right)=1\right]+Pr\left[sign\left({F}_{pt}\right)=-1|sign\left({\alpha }_{pt}^{i}\right)=-1\right]}{2}.$$
(6)

The pairwise test that allows us to compare two alternative measures of performance is also conducted following the method proposed by Berk and van Binsbergen (2016). To empirically distinguish between two contesting models, say i and j, we estimate the following regression.

$$sign\left({F}_{pt}\right)={\delta }_{0}+{\delta }_{1}\left(\frac{sign\left({\alpha }_{pt}^{i}\right)}{var\left({sign(\alpha }_{pt}^{i})\right)}-\frac{sign\left({\alpha }_{pt}^{j}\right)}{var\left({sign(\alpha }_{pt}^{j})\right)}\right)+{\omega }_{pt}.$$
(7)

The sign of the slope coefficient indicates the model that better explains investor behavior. If \({\delta }_{1}\) is positive and significant, then we can conclude that model i is better that model j in explaining mutual fund flows.

The method proposed by Berk and van Binsbergen (2016) focuses only on the sign of the flow and performance and ignores their magnitude. To overcome this limitation, we follow the method proposed by Barber et al. (2016) and estimate the following regression:

$${F}_{pt}= {\varphi }_{p}+ {\varphi }_{1}{\alpha }_{pt}^{i}+{\varphi }_{2}^{\intercal }{X}_{i,t-1}+ {\mu }_{t}+{\vartheta }_{pt},$$
(8)

where, \({X}_{i,t-1}\) is a vector of control variables that include fund specific controls— logarithm of age of the fund, fund size; AMC size to control for the AMC networks in attracting flows; and type flow (total flows to all funds that belong to same type)Footnote 4 to control for flows driven by popularity of the type of the fund. \({\varphi }_{p}\) and \({\mu }_{t}\) are fund and time fixed effects, respectively. We standardize all the independent variables (except age) to facilitate direct comparison of alternative performance measures in explaining the fund flows. Further, following Barber et al. (2016), all our inferences are based on double-clustered standard errors by fund and quarter. This allows us to mitigate the issue of cross-sectional dependence.

The final set of empirical analysis in our study aims at examining the flow–performance sensitivity in connection with uncertainty. We examine whether uncertainty alters the explanatory power of competing performance measures by estimating the below regression.

$${F}_{pt}= {\theta }_{p}+ {\theta }_{1}{\alpha }_{pt}^{i}+{\theta }_{2}{EPU}_{t-1}+{\theta }_{3} {{(\alpha }_{pt}^{i}\times EPU}_{t-1})+{\theta }_{4}^{\intercal }{X}_{i,t-1} +{\eta }_{pt}.$$
(9)

Our variable of interest here is the interaction term of measure of performance and uncertainty. A statistically significant coefficient for our variable of interest would imply that uncertainty influences the relation between flows and performance measures.

3 Results and discussion

3.1 Flow performance relation: Berk and van Binsbergen (BvB) approach

In our attempt to gain an insight into the risk factors that the investors consider while making investment decisions, we implement the signed flow–performance analysis explained in Sect. 2.2. First, we estimate the performance of each fund using a rolling regression considering each of the six alternative models. The performance measures based on each of these models are then used to estimate Eq. (5), using panel regression. We control for both fund and time specific fixed effects. The coefficient estimates obtained from this estimation are reported in Table 2. Each row corresponds to one of the alternative models considered for evaluation of funds, with the best performing model being reported first. The first column reports the coefficient estimates for \({\gamma }_{1}\). Double-clustered standard error by fund and month are reported in the parenthesis. The second column presents \(\frac{{\gamma }_{1}+1}{2}\), which shows the percentage of fund flows explained by the performance measure from each of the models. The results confirm that there exists a statistically significant relation between the performance of a fund and flows. We also observe that none of the models explains more than 60 per cent of fund flows, i.e., a portion of the flows remain unexplained. The results indicate that a naïve model, wherein the performance of a fund is evaluated solely based on the excess returns generated over the risk free rate, is used by investors in India. It explains around 60 per cent of the mutual fund flows, which is considerably higher than the next best model, MAR that explains 58 per cent of the flows. Two naïve measures of performance outperform CAPM. This is contrary to the findings of Berk and van Binsbergen (2016), Barber et al. (2016) and Blocher and Molyboga (2017) among others, where CAPM is identified as the model that is closest to the one that is used by investors while making investment decisions. The results suggest that investors fail to adjust for exposure to the risk factors when choosing between funds.

Table 2 Flow–performance relation: BvB approach
Full size table

3.2 Pairwise comparison of performance measures

The results of the previous analysis suggest that investors compare the average excess return generated by funds over the preceding 4 quarters while making investment decisions. In other words, Indian investors do not adjust for the known risk factors while evaluating alternative investment options. However, we conduct a pairwise test to assess the relative performance of the alternative models considered in our study. This allows us to deduce whether the difference in the performance of alternative models is statistically significant. The fixed effects estimates of Eq. (7) are presented in Table 3. As is already noted, it is the sign of the coefficient that indicates the superior of the two models that are compared. In terms of Eq. (7), a positive and significant coefficient implies that model i outperforms model j in explaining the sensitivity of flows. In the table below, models in the row (represent model i in Eq. (7)) are compared with those in the columns (represents model j in Eq. (7)). Examining the results, we observe that naïve measure based on market adjusted return outperforms all the alternative models considered in the study. The results of the pairwise comparison confirm the findings of our previous analysis, wherein the flow–performance sensitivity was estimated for each model separately. Overall, the results suggest that investors behave naively while evaluating competing investment options.

Table 3 Pairwise comparison of performance measures
Full size table

The finding that the CAPM significantly outperforms the FFM suggests that the investors discount only the returns associated with the market risk while evaluating alternative investment opportunities. In other words, on an average, investors are more likely to adjust for market risk while assessing the performance of a fund. An average investor in our sample overlooks the risk associated with size, value and momentum.

3.3 Flow performance relation: Barber, Huang and Odean (BHO) approach

The analysis in the previous section draws conclusions regarding flow–performance relation based on the sign of flows and performance measure. As against this, the methodology proposed by Barber et al. (2016) takes into consideration the magnitude of flow and performance measures. The results of our analysis based on this approach are presented in Table 4. Each column corresponds to the estimation result for Eq. (8), taking into consideration each of the performance measures separately. Consistent with the central results of the previous section, we find that the naïve measures outperform the measures based on factor models in explaining fund flows. To specify, a one standard deviation increase in the naïve performance measures leads to more than 5 per cent increase in fund flows. Of the naïve measures, a one standard deviation change in Return and ER based measures are associated with close to 10 percent change in flows.Footnote 5 In contrast to this, one standard deviation increase in performance measures based on CAPM, TFM and FFM increases fund flow by 4.3, 3.3 and 3.1 per cent, respectively.

Table 4 Flow performance relation: BHO approach
Full size table

Our findings suggest that the CAPM outperforms both the TFM and the FFM in explaining the mutual fund flows. While our conclusion in this regard corroborates that of the existing studies, our main result contradicts these studies that conclude that CAPM based measure best explain fund flows (Berk & Van Binsbergen, 2016; Barber et al., 2016; Blocher & Molyboga, 2017; and Agarwal et al., 2018). We believe that the results differ from the existing studies as we explore the case of a developing market. We conjecture uncertainty prevalent in such markets causes the behaviour exhibited by mutual fund investors in developing market to differ substantially from that of their counterparts in developed markets. To confirm, we control for uncertainty and revisit the flow–performance relation.

3.4 Uncertainty and flow–performance sensitivity

Given that theoretical and experimental studies confirm uncertainty is a significant determinant of investment decisions, we examine the impact of uncertainty on mutual fund flows. Analyzing the behaviour of mutual fund investors under uncertainty Ben-Rephael (2017) documents evidence for flight-to-liquidity. Based on this evidence, we hypothesize that higher uncertainty would encourage investors to make non-naïve investment decisions. An informed investor would account for all the risk factors while evaluating alternative investment options. For the purpose of our analysis, we use EPU Index proposed and constructed by Baker et al. (2016). The index is constructed based on newspaper articles regarding policy uncertainty and is widely used in empirical literature as a measure of uncertainty (see for instance Wang et al., 2014; Li, 2017; Ftiti & Hadhri, 2019 and Zhang et al., 2019 among others).

For the purpose of analysis, we compute quarterly EPU as average of monthly index values that constitute the quarter. Further, as is the case with other explanatory variables, we use standardized EPU and estimate Eq. (9). The results presented in Table 5 confirm uncertainty discourages investment in financial markets. We find that one standard deviation increase in uncertainty leads to fall of around 3 per cent in mutual fund flows. However, the coefficient of interaction term is insignificant in all the cases. Nonetheless, our central result of investors using naïve measure to evaluate alternative funds remains unaltered even after controlling for uncertainty.

Table 5 Uncertainty and flow–performance sensitivity
Full size table

To summarize, we find that the mutual fund investors in India places highest weight on returns, adjusted for market return or risk-free return, while evaluating alternative funds. This implies that investors fail to discount (or place little weight on) returns associated with market, size, value, and momentum factors; and reward the fund managers with additional funds.

4 Robustness check

We have performed four alternative robustness checks for our results. First, we consider alternative time horizons for performance evaluation. There is no consensus in literature regarding the horizon over which investors evaluate performance. While Berk and van Binsbergen (2016) consider different time horizons to establish the robustness of their results, Barber et al. (2016) uses a weighted average of performance over preceding 18 months. We calculate alpha over 2, 4, 6, and 8 quarters; and examine if the choice of horizon changes our result by estimating Eq. (5) using these alphas. Table 6 ranks each of the performance measures over different time horizons considered. We find that our central conclusion that investors use naïve model to evaluate performance remains unaltered.

Table 6 Model ranking over different horizons
Full size table

Second, we examine the relation between flow and performance measures by estimating regression of fund flow on all the performance measures.Footnote 6 In the preceding sections, we use each of the performance measures separately due to the high correlation among alternative measures. Including all the performance measures allows us to evaluate the partial coefficients and draw robust results. The results presented in Table 7 shows that only ER and CAPM alpha are significant impact on mutual fund flows. The partial coefficient of ER is considerable greater than that of CAPM alpha. This further strengthens our conclusion that mutual fund investors largely rely on naïve measurers for evaluating alternative funds.

Table 7 Model horse race
Full size table

Third, we address the issue of asymmetric impact of uncertainty by estimating a quantile regression. It is widely accepted in the literature that the effect of uncertainty is asymmetric and highly related to the market conditions (see for instance You et al., 2017). Table 8 reports the coefficient for the alternative performance measures considered in our study. Each column corresponds to a different quantile. The results confirm that the impact of uncertainty varies depending on the market conditions, and that in general uncertainty has a negative impact. However, it does not negate our conclusion regarding the naïvety of mutual fund investors. In all the quantile considered we find the coefficient of the naïve measures is greater than that of the asset pricing based measure.Footnote 7 This further strengthens our conclusion regarding the naïve behaviour of mutual fund investors in India.

Table 8 Asymmetric impact of uncertainty of flow–performance sensitivity
Full size table

Lastly, we re-examine the flow–performance relation using VIX as a measure of uncertainty. While EPU measures uncertainty related to economic policy, VIX measures market uncertainty (Chung & Chuwonganant, 2014). The result reported in Table 9 confirms that flow–performance sensitivity is not influenced by the level of uncertainty and investors continue to rely on naïve performance measures.

Table 9 Uncertainty and flow–performance sensitivity: VIX
Full size table

5 Conclusion

The asset pricing models aids an investor in making investment decisions by identifying the risk factors that determine the price of an asset. Studies over the years have empirically tested alternative asset pricing models. While some studies confirm that the models explain the cross section of returns adequately some find them to be inadequate. Given such equivocal evidences regarding the validity of the asset pricing models, it is intriguing to identify the risk factors that individual investors while making capital allocation. However, it is rather difficult to observe how investors identify profitable investment opportunities and how they respond to such opportunities. In this regard, Berk and van Binsbergen (2016) and Barber et al. (2016) suggest that the mutual fund flows can be used for measuring investor response to opportunities identified by them. Following different methodologies both the studies conclude that the Capital Asset Pricing Model (CAPM) is closest to the true model used by investors in the US to evaluate alternative investment options.

We have investigated into the risk factors that Indian investors consider while making investment decisions. Majority of the previous studies are confined to US markets and our study on an emerging market provides a fresh outlook into the investor behavior since Indian market is very distinct for that of US. We have used fund flows as a measure of investor reaction and have taken into consideration six alternative measure of performance. Three of the performance measures are based on the factor-based asset pricing models—CAPM (Lintner, 1965; Sharpe, 1964); Fama French (1993) Three-Factor Model and Carhart (1997) Four-Factor Model. The remaining three measure are based on naïve models, wherein alternative funds are evaluated based on returns generated, excess return over risk free rate and market adjusted return, respectively. Our sample consists of actively managed equity schemes for the period from April 2006 to December 2019.

Following the methodology proposed by Berk and van Binsbergen (2016), we analyze the relation between quarterly performance measures and fund flows. Our results considering each of the performance measure separately suggests that the investors use naïve measures to compare alternative investment opportunities. Even after pairwise comparison of the performance measures, we find that naïve model outperforms the factor models in explaining fund flows. Analysis using methodology proposed by Barber et al. (2016) also confirms that investors overlook the risk factors while making investment decisions. Thus, the study suggests that majority of the mutual fund investors in India are naïve and make investment decisions based on superficial information.

Our result are contradictory to that of previous studies that examine the case of mutual fund investors (Barber et al., 2016; Berk & van Binsbergen, 2016) and hedge fund investors (Agarwal et al., 2018; Blocher & Molyboga, 2017). We hypothesize that the contradictory results are driven by the level of uncertainty in the market and proceed to examine whether the flow–performance relation is sensitive to the level of uncertainty. Using Economic Policy Uncertainty (EPU) Index of Baker et al. (2016) we confirm that uncertainty has a negative impact on mutual fund flows. However, we do not find evidence that suggests flow–performance relation are sensitive to uncertainty.