1 Introduction

The majority of research on mutual fund performance focuses on the equity or bond fund segments, with the prevailing conclusion being that they are unable to outperform their benchmarks (e.g., Ferson et al. 2006; Heyden and Röder 2020; Mingo-López et al. 2022). In light of these findings, and given investors’ desire for higher asset class diversification, hybrid/multi-asset fundsFootnote 1 have grown in popularity among both institutional and retail investors. Indeed, total net assets of hybrid funds in the US reached $1,868.70 billion by December 2021, accounting for approximately 6% of the industry (ICI 2022). In Europe, these figures are substantially higher: by the end of December 2021, European multi-asset funds had €2,454.75 billion of total net assets, representing 17.7% of the European Undertakings for Collective Investment in Transferable Securities (UCITS) market (EFAMA 2021). However, notwithstanding the fact that they already constitute a substantial portion of the mutual fund industry, few studies can be found about the performance of these funds and most of them are focused on the US market.

According to our knowledge, Comer (2006) was the first to address hybrid fund performance. After incorporating additional equity and bond indices into the Treynor-Mazuy model, the author found that US hybrid funds had successful market timing abilities from 1992 to 2000. Some years later, using extensions of Carhart’s (1997) model that evaluated fixed-income exposures, Comer et al. (2009a) showed that hybrid funds from the US market significantly underperformed throughout the late 1990s and early 2000s by up to 1.5% per year. In contrast to Comer (2006), they reported unsuccessful timing abilities from fund managers. Additionally, based on an alternative return attribution methodology, Comer et al. (2009b) also found that US hybrid funds were unable of delivering positive returns to their investors.

For the 1998–2009 period, Herrmann and Scholz (2013) studied how 520 US funds performed by adding several bond factors to Carhart’s (1997) model. They found that hybrid funds displayed significantly negative alphas. Also, within the US market, Dass et al. (2013) showed an absence of significant performance differentials for team-managed and single-managed balanced funds, with both fund groups underperforming significantly, by approximately 1.2% per year, across the 1992–2009 period.

Khang and Miller (2022) used a weights-based measure to identify the performance components (namely, active management, forecasting success and return dispersion) of 732 US asset allocation funds from 1983 to 2013. Their results showed that, although the components were useful to explain performance differentials among funds, on average funds were unable of outperforming their benchmark. Besides, based on a dataset including four different types of US asset allocation funds, for the 2011–2021 period, Malhotra and Hadad (2024) found that fund performance was, on average, lower than that of US stocks alone. However, while cautious and moderate allocation funds exhibited neutral alphas, both aggressive and flexible funds displayed significantly negative alphas of, approximately, − 2.0% per year. Furthermore, they also reported an absence of successful timing abilities from portfolio managers.

Within the Canadian market, for an 877 hybrid fund dataset, Ayadi et al. (2016) found that the net alpha was negative by an average of 4.2% per year over the 1991–2011 time span. In addition, gross alphas were neutral but reached − 2.0% per year. They also reported that performance deteriorated after controlling for exposures to fixed-income instruments in the evaluation model, consistent with Comer et al. (2009a). Additionally, using a dataset of 1015 Canadian domestic hybrid mutual funds, Ayadi et al. (2023) found evidence of neutral (stock and bond) market timing abilities.

For a 632 fund dataset, from the US, UK and Canadian markets, Clare et al. (2016) assessed multi-asset fund managers’ abilities to time the market from 2000 to 2012. Despite employing returns-based and holdings-based methods, both demonstrated small fractions of successful timers. In addition, despite all average alphas being negative, they were approximately twice as high for UK and Canadian funds than for US funds.

Thus, most empirical studies conducted so far, focused on the US and Canadian markets, show that multi-asset funds tend to exhibit inferior performance. However, these works are all based on mutual funds that invest in their domestic markets. The only study we are aware of about the performance of multi-asset funds investing internationally is Larrymore and Rodriguez (2007).Footnote 2 By means of a three-index style analysis, the authors examined 27 US-based asset allocation funds investing globally, across the 1999–2003 period, and found that they exhibited significantly positive average returns. In addition, the authors reported a statistically significant average alpha of approximately 3.2% per year.Footnote 3 So, the performance of international multi-asset funds is a research topic that remains broadly unexplored, but undoubtedly an interesting one. Compared to stock or bond mutual funds, multi-asset funds offer an increased diversification across asset classes, serving as a hedge against adverse events that may affect a specific class. Therefore, these funds may be a better option for investors, especially during (equity or fixed-income) market downturns. This is particularly pertinent because financial markets have experienced a number of important crises in the last twenty years, namely as a result of the early 2000s dot-com bubble, the 2007–2008 global crisis, the European debt crisis or, most recently, the Covid-19 outbreak.Footnote 4 Besides, the additional diversification benefits of an international investment universe may also reduce portfolio risk and, consequently, help fund managers achieve better risk-adjusted performance. In this way, we expand the existing literature by carrying out a comprehensive investigation into the performance of multi-asset funds investing internationally.

Though multi-asset funds are more important in Europe than in the US, the sole research that assesses the performance of multi-asset funds in a developed European market is Clare et al. (2016), whose dataset includes 80 funds domiciled in the UK.Footnote 5 Nevertheless, the weight of multi-asset funds in the net assets of the UCITS industry is considerably smaller in the UK than in most other European markets. So, we analyse a dataset of Portuguese-domiciled multi-asset funds over the 2004–2021 period. Despite being small in the European context, this market has several appealing characteristics for such a study. Firstly, it is a market where multi-asset funds represent a substantially higher proportion of total net assets than bond or equity funds. In fact, by December 2021, the Portuguese mutual fund market was worth €18,920.2 million, with multi-asset funds standing for 54.2% of the total, while bond and equity funds accounted for 15.4% and 17.9%, respectively (EFAMA, 2021). Secondly, in Portugal, multi-asset funds represent a considerably higher proportion of the fund industry than in the main European markets, like Germany (29.1%), France (17.6%) or the UK (14.4%). Thirdly, all multi-asset funds invest internationally.

Besides, we also estimate and compare fund performance throughout distinct market phases. A number of studies indicate that equity funds present a significantly higher performance during times of crisis compared to non-crisis phases (e.g., Kosowski 2011; Nofsinger and Varma 2014). Such findings might reflect an increased effort from mutual fund managers to obtain higher returns for investors when they value them most (Glode 2011) or changes in fund managers’ market timing and stock selection abilities (Kacperczyk et al. 2014; Chen et al. 2021). Consequently, it would be preferable to invest in equity funds during down-markets than during up-markets. For multi-asset funds, Comer et al. (2009b) demonstrated that US funds were able to outperform their benchmarks but only during weak stock market conditions. Yet, this issue has not been investigated beyond the US market or for funds investing internationally. This is another contribution of this paper.

Since results on fund performance can depend heavily on the underlying evaluation model, especially for multi-asset funds, where we cannot find a set of bond and equity indices that researchers widely consider as the most appropriate to explain fund returns, we begin by establishing which performance evaluation model better explains the returns of our fund dataset. Starting with a simple two-factor model, incorporating a bond and an equity index, we explore the added value of introducing more factors, which stem from the bond and the equity mutual fund performance literature. These include default and option factors (Elton et al. 1995), size, book-to-market, profitability, and investment (Fama and French 2015), momentum (Carhart 1997), global bond (Hoepner and Nilsson 2018) and global equity factors. Furthermore, we additionally integrate conditioning information to account for the time variability of alphas and betas. This stepwise process of identifying the relevant factors, as well as public information variables, that explain multi-asset fund returns is an additional contribution of this work.

Overall, our objective is to provide answers to the following research questions: Which factors better explain the returns of Portuguese-based international multi-asset funds? Are these funds able to outperform the market? Are there significant changes in fund performance and investment styles between crisis and non-crisis phases? Do multi-asset funds serve as a hedge against market downturns?

The remaining sections are divided as follows: Sect. 2 describes the research methodology. Section 3 contains a data description. Section 4 presents and discusses the results. Section 5 provides the conclusions and some future research suggestions.

2 Methodology

2.1 Fund performance evaluation model

Our base model for assessing multi-asset fund performance is the following 8-factor regression:

$${r}_{p,t}={\alpha }_{p}+{\beta }_{1} {Bond}_{t}+{\beta }_{2} {Equity}_{t}+{\beta }_{3}{ Default}_{t}+{\beta }_{4} SM{B}_{t}+{\beta }_{5} {HML}_{t}+{\beta }_{6} {WML}_{t}+{\beta }_{7} {RMW}_{t}+{\beta }_{8} {WEquity}_{t}+{\varepsilon }_{p,\text{t}}$$
(1)

where \({r}_{p,t}\) is the portfolio’s excess return during month \(t\), \({Bond}_{t}\) is a high-quality bond index excess return, \({Equity}_{t}\) represents an equity index excess return, \({Default}_{t}\) is a default spread, \(SM{B}_{t}\) represents the return difference among small and big stocks, \({HML}_{t}\) corresponds to the difference in returns between stocks with high and low book-to-market ratios, \({WML}_{t}\) is a return differential between past winner and past loser stocks, \({RMW}_{t}\) is the return difference among stocks with robust and weak profitability, \({WEquity}_{t}\) represents a global equity index excess return, \({\varepsilon }_{p,\text{t}}\) is a residual term.Footnote 6 Portfolio’s performance is given by the model’s intercept.

However, unconditional approaches, which assume constant risk exposures, may produce biased performance estimates. Therefore, we extend the model presented in (1) to a conditional framework by incorporating public information variables. Following Christopherson et al. (1998) and Ayadi et al. (2016), among others, risk factor loadings and performance are both time-varying in response to a vector of lagged, demeaned, information variables (\({z}_{t-1}\)). This model writes as:

$${r}_{p,t}={\alpha }_{0p}+{A}_{p}{\prime} {z}_{t-1}+{\beta }_{01} {Bond}_{t}+{\beta }_{1p}{\prime}\left({z}_{t-1} {Bond}_{t}\right)+{\beta }_{02}{Equity}_{t}+{\beta }_{2p}{\prime}\left({z}_{t-1} {Equity}_{t}\right)+{\beta }_{03}{Default}_{t}+{\beta }_{3p}{\prime}\left({z}_{t-1} {Default}_{t}\right)+{\beta }_{04}SM{B}_{t}+{\beta }_{4p}{\prime}\left({z}_{t-1} SM{B}_{t}\right)+{\beta }_{05}{HML}_{t}+{\beta }_{5p}{\prime}\left({z}_{t-1} {HML}_{t}\right)+{\beta }_{06}{WML}_{t}+{\beta }_{6p}{\prime}\left({z}_{t-1} {WML}_{t}\right)+{\beta }_{07}{RMW}_{t}+{\beta }_{7p}{\prime}\left({z}_{t-1} {RMW}_{t}\right)+{\beta }_{08}{WEquity}_{t}+{\beta }_{8p}{\prime}\left({z}_{t-1}{WEquity}_{t}\right)+{\varepsilon }_{p,t}$$
(2)

where the average of the conditional alphas is given by \({\alpha }_{0p}\) and the averages of the conditional betas are given by \({\beta }_{01}, \dots , {\beta }_{08}\). In the above regression, vectors \({A}_{p}{\prime}\) and \({\beta }_{1p}{\prime}, \dots , {\beta }_{8p}{\prime}\) indicate how the alphas and the betas, respectively, change with regard to each information variable.

2.2 Performance in crisis and non-crisis phases

To investigate if multi-asset fund performance changes over different market phases, we must identify the crises that occurred during the 2004–2021 time horizon. Since market crises are usually associated to bear stock market periods, we follow a number of preceding studies (e.g., Wang et al. 2022; Xu et al. 2023) and employ the Pagan and Sossounov (2003) algorithm for this task. As the authors recommend using eight-month windows for financial asset prices, a stock index peaks at time t if \(\mathit{ln}\left({P}_{t-8},...,{P}_{t-1}\right)<\mathit{ln}\left({P}_{t}\right)>\mathit{ln}\left({P}_{t+1},...,{P}_{t+8}\right)\), where Pt stands for the index value. If \(\mathit{ln}\left({P}_{t-8},...,{P}_{t-1}\right)>\mathit{ln}\left({P}_{t}\right)<\mathit{ln}\left({P}_{t+1},...,{P}_{t+8}\right)\) we have a trough. Subsequently, every decrease of at least 20% from peak to trough is classified as a bear market phase.

For the MSCI AC Europe, which is the stock index used in our analysis, this process leads to three crises/bear market periods: May 2007 to February 2009 (− 56.67%), February 2011 to September 2011 (− 21.36%) and December 2019 to March 2020 (− 23.54%). These periods are related to the global financial crisis, the European debt crisis, which itself affected equity markets, and the Covid outbreak, respectively.

In the same way as Leite and Cortez (2015), to split alpha and beta estimates for crisis and non-crisis phases, we included two dummy variables in regression (1) and obtained the equation below:

$${r}_{p,t}={\alpha }_{NC}{D}_{NC,t}+{\alpha }_{C}{D}_{C,t}+{\beta }_{1NC }Bon{d}_{t}{D}_{NC,t}+{\beta }_{1C} Bon{d}_{t}{D}_{C,t}+{\beta }_{2NC} Equit{y}_{t}{D}_{NC,t}+{\beta }_{2C} Equit{y}_{t}{D}_{C,t}+{\beta }_{3NC} Defaul{t}_{t}{D}_{NC,t}+{\beta }_{3C} Defaul{t}_{t}{D}_{C,t}+{\beta }_{4NC} {SMB}_{t}{D}_{NC,t}+{\beta }_{4C} {SMB}_{t}{D}_{C,t}+{\beta }_{5NC} {HML}_{t}{D}_{NC,t}+{\beta }_{5C} {HML}_{t}{D}_{C,t}+{\beta }_{6NC} {WML}_{t}{D}_{NC,t}+{\beta }_{6C} {WML}_{t}{D}_{C,t}+{\beta }_{7NC} {RMW}_{t}{D}_{NC,t}+{\beta }_{7C} {RMW}_{t}{D}_{C,t}+{\beta }_{8NC} {WEquity}_{t}{D}_{NC,t}+{\beta }_{8C}{ WEquity}_{t}{D}_{C,t}+{\varepsilon }_{p,t}$$
(3)

where \({D}_{C,t}\) equals 1 for crisis periods and 0 otherwise, and \({D}_{NC,t}\) equals 1 for non-crisis periods and 0 otherwise. As for regression (3), the alpha for non-crisis phases is given by \({\alpha }_{NC}\) and the alpha for market crises is\({\alpha }_{C}\). \({\beta }_{1NC }, \dots , {\beta }_{8NC}\)(\({\beta }_{1C }, \dots , {\beta }_{8C}\)) stand for the factor exposures across non-crisis (crisis) times.Footnote 7 Employing dummy variables to separate the coefficients for both market phases is a different form of allowing for the variation of fund performance estimates, as well as risk exposures, over time, which can be considered as an alternative to conditional performance evaluation models (Leite and Cortez 2017).Footnote 8

3 Data

3.1 Fund dataset

Our survivorship bias-free dataset includes all (40) funds classified by APFIPP—the Portuguese Investment Fund, Pension Fund and Asset Management Association—as multi-asset funds, with a minimum of 24 monthly observations between January 2004 and December 2021. All funds invest internationally in more than an asset class, namely in equities and bonds.Footnote 9

Considering funds’ investment styles, APFIPP further divides funds into four sub-categories: defensive funds, where the equity exposure reaches a maximum of 15% of Total Net Assets (TNA); moderate funds, which have an equity exposure between 16 and 35% of TNA; balanced funds, where the equity exposure lies between 36 and 65% of TNA; and aggressive funds, where the equity exposure is at least 66% of TNA. However, since most funds belong to the two intermediate categories, we opted to merge the two more defensive (23 funds) and the two more aggressive (17 funds) sub-categories in our empirical analysis. Therefore, funds investing predominantly in bonds will be denominated multi-asset bond funds, while funds investing primarily in equities will be designated multi-asset equity funds.

Mutual fund net asset values were obtained from CMVM—the Portuguese Securities Market Commission. Monthly fund returns, denoted in Euros, are computed including operating expenses but excluding sales charges. The 1-month Euribor represents the risk-free rate.

Table 1 provides summary statistics for three equally-weighted fund portfolios (one including all funds, one for multi-asset bond funds and another for multi-asset equity funds). During our evaluation period, all portfolios exhibit positive mean excess returns. As expected, funds that invest predominantly in bonds display considerably lower excess returns, as well as lower standard deviations, than funds investing mostly in equities.Footnote 10

Table 1 Summary Statistics for Multi-Asset Fund Returns and Risk Factors
Full size table

3.2 Benchmarks and factors

The bond indices we use are from Intercontinental Exchange (ICE) and correspond to the former Bank of America Merrill Lynch (BofAML) Total Return (TR) indices, while TR equity indices are from Morgan Stanley Capital International (MSCI). They were all obtained in Datastream.

The bond factor, which measures exposures to investment grade bonds, is the excess return of the ICE BofAML Euro Broad index, while the equity factor corresponds to the MSCI AC Europe index excess return. The default factor, which captures default risk compensation, is created by subtracting the returns of the ICE BofAML Euro Government index from the returns of the ICE BofAML Euro High-Yield index.Footnote 11 European size, book-to-market, investment, profitability, and momentum factors were all collected in Kenneth French’s website and, subsequently, converted into Euros.Footnote 12 The MSCI AC World index excess return represents the global equity factor. Table 1 shows that practically all risk factors exhibit positive mean excess returns for the evaluation period, with the only exception being the book-to-market factor.

Since some of the factors were highly correlated,Footnote 13 we orthogonalized the regressors, in order to prevent multicollinearity. Initially, following Hoepner et al. (2011), we ran an auxiliary regression between the impacted factors. Then, new factors were constructed by adding the intercept from those regressions with the respective residuals. Following this procedure, the pair-wise correlations between the eight relevant factors are all reasonably low, ranging from − 0.57 to 0.46.

3.3 Conditional variables

We consider three public information variables that many prior studies (e.g., Ayadi and Kryzanowski 2011; Banegas et al. 2013) have proven suitable to forecast stock and/or bond returns: (1) dividend yields; (2) short-term rates; (3) term structure slope. The first variable is related to the MSCI AC Europe index; the second corresponds to the annualized 3-month Euribor; the third variable is obtained by subtracting the 3-month Euribor from the annualized 10-year Eurozone government bond yields. All variables are lagged 1-month. To prevent spurious regressions, we perform a stochastic detrending on the three instruments, as recommended by Ferson et al. (2003), using their previous 12-month moving averages.

However, unlike most studies, we previously analyse the predicted power of these variables using simple and multiple regressions. In this way, redundant variables can be excluded, so that they will not influence our findings. Our results are shown in Table 2, where we can confirm that only dividend yields and short-term rates have some predictive power of fund returns. Therefore, only these information variables, which exhibit a very low correlation (0.10),Footnote 14 will be included in the conditional model.

Table 2 Multi-Asset Fund Return Predictability
Full size table

4 Results

4.1 Factor model selection

To find which specification better describes the returns of multi-asset funds in our dataset, we investigate the goodness of fit of several multi-factor models. This analysis is performed at an aggregate level, based on an equally-weighted portfolio that includes all funds.

Our first model is an unconditional two-factor model incorporating a European high-quality bond index and a European equity index (model M1).Footnote 15 Then, we explore the additional value of incorporating more factors, one at a time. In line with Elton et al. (1995) and Hoepner and Nilsson (2018), we start by incorporating three bond-related factors: a default factor, to evaluate funds’ exposures to high-yield instruments (model M2); an option factor, to control for exposures to mortgage-backed securities (model M3); a global bond factor, to measure funds’ exposures to global bonds (model M4). Subsequently, we incorporate the six equity-related factors proposed by Carhart (1997) and Fama and French (2015): size (model M5), book-to-market (model M6), momentum (model M7), profitability (model M8) and investment (model M9). We also include a global equity factor, to account for funds’ exposures to global equities (model M10). Finally, we incorporate conditioning information and allow for the time variation of alphas and betas (model M11). In this stepwise process, every factor that does not improve the model’s explanatory power significantly is dropped. Following Otten and Bams (2004), a significant improvement in the adjusted R2 between two models occurs when two times the difference in their log likelihoods exceeds the corresponding critical value of a χ25% (d.f.) test statistic. Table 3 reports the factor exposures for each of the 11 models considered.

Table 3 Factor Models for Multi-Asset Fund Returns
Full size table

Based on model M1, which presents a reasonably high adjusted R2 (87.29%), we can confirm that funds have significant exposures to both European (investment grade) bonds and equities. Model M2 has a significantly higher explanatory power than model M1, with the adjusted R2 increasing substantially (90.34%). The coefficient obtained for the default factor shows that funds are not only significantly invested in investment grade bonds, but also in high-yield debt. In the next two models (M3 and M4), the additional factors display insignificant coefficients and, in comparison to model M2, neither leads to a significantly higher explanatory power. This means that multi-asset funds in our dataset do not have significant exposures to mortgage-backed securities or global bonds, the reason for which both factors will not be considered in the subsequent specifications.

Models M5, M6 and M7 confirm the relevance of incorporating the SMB, HML and WML factors since each of them leads to significant increases in the adjusted R2s, which reaches 91.04% in the last specification. The coefficients obtained for these factors reveal significant exposures to small and growth stocks, together with momentum strategies. In Model M8 we can see that the profitability factor further increases the model’s explanatory power significantly, with the negative coefficient representing a significant exposure to weak profitability stocks. On the contrary, model M9 shows that the investment factor is not statistically significant and, therefore, will be dropped. The last unconditional model (M10) also leads to a significant increase in the adjusted R2, which now reaches 92.76%. The positive coefficient of the additional factor denotes that funds are significantly invested not only in European but also in global equities.

Finally, we incorporate conditioning information in model M10, resulting in our last specification (model M11), which also exhibits a significantly higher explanatory power (93.42%). Average conditional betas confirm that most previous relevant factor exposures remain statistically significant. Though average conditional betas of the SMB and HML factors lose their statistical significance, these factors along with their cross-products with the lagged instruments are not redundant for the model, as confirmed by a Wald test,Footnote 16 and therefore will be kept.

4.2 Multi-asset fund performance

Our performance evaluation results, based on the conditional model (M11) presented in Eq. (2), are shown in Table 4.

Table 4 Performance and Risk Exposures of International Multi-Asset Funds
Full size table

For the “all funds” portfolio, we can see that performance estimates are negative and statistically significant, reaching − 0.2185% per month. Thus, international multi-asset funds in our dataset underperform the market by 2.62% per year, in clear contrast to the results of Larrymore and Rodriguez (2007). Besides, compared to studies on domestic funds, our result is considerably worse than the ones reported for US funds by Comer et al. (2009a), Dass et al. (2013) and Malhotra and Hadad (2024), but substantially better than the one obtained for Canadian funds by Ayadi et al. (2016) using a similar (conditional multi-factor) model.Footnote 17 Factor loadings confirm significant exposures to European and global equities, investment grade and high-yield bonds, and low profitability stocks.

Since our dataset includes funds with diverse investment styles, we also evaluate performance for each style, i.e., multi-asset bond funds and multi-asset equity funds. Additionally, to improve comparability, we create a “difference” portfolio using the returns of both fund groups. Consistent with the results obtained for the “all funds” portfolio, both multi-asset fund sub-categories underperform significantly. Funds investing predominantly in bonds underperform by − 0.1354% per month (− 1.62% per year), while funds investing mainly in equities underperform by − 0.2765% per month (− 3.32% per year). This result aligns with the conclusions drawn by Malhotra and Hadad (2024), which suggest that funds with smaller proportions of equities achieve better performance than funds with higher proportions of equities. Nevertheless, for the overall period under analysis, neither fund group seem to be an appealing investment. Yet, multi-asset bond funds perform significantly better than multi-asset equity funds by 0.1411% per month (1.69% per year).Footnote 18

As to risk exposures, both fund groups show significant exposures to investment grade bonds, high-yield bonds and both European and global equities. Besides, they also exhibit significant negative loadings on the profitability factor. However, only funds that invest predominantly in equities show significant exposures to small-caps. The more aggressive funds have significantly higher exposures to (European and global) equities, high-yield bonds, and small-cap stocks than their more defensive counterparts,Footnote 19 as expected.Footnote 20

4.3 The impact of fees on performance

Until now, we have used net fund returns, already corrected for operating expenses. However, some authors (e.g., Elton et al. 1995) show that mutual funds have a propensity to underperform the market by the fees they charge. So, to assess how fund fees influence the performance of multi-asset funds, we added them back to each portfolio return. We started by gathering information about the annual total expense ratio of each individual fund from APFIPP. Afterwards, we divided it by twelve and added the value to each fund’s monthly return. Management fees reach an average of 1.71% per year and, as expected, are higher for multi-asset equity funds (1.97% per year) than for multi-asset bond funds (1.52% per year). In Table 5 we present the performance estimates after and before management fees.

Table 5 Fund fees and performance
Full size table

For the “all funds” portfolio, alphas before fees remain negative, reaching − 0.0755% per month (− 0.91% per year), but are only significant at the 10% level. Similar evidence is found for multi-asset equity funds. However, when management fees are added back, multi-asset bond funds’ performance shifts from significantly negative to neutral, with a value of only − 0.0084% per month (− 0.10% per year), meaning that these funds perform well enough to compensate their expenses. So, the underperformance of funds in our dataset is, at least partially, due to the expenses they charge. Furthermore, fund fees alone cannot justify the performance differential between the two investment style portfolios since these remain significant both on a before-fee and an after-fee basis.

4.4 Performance in crisis and non-crisis phases

In Table 6 we report multi-asset funds’ performance, as well as their factor loadings, for crisis and non-crisis phases separately.Footnote 21

Table 6 Performance in Crisis and Non-Crisis Phases
Full size table

The “all funds” portfolio presents a significantly negative alpha for non-crisis times and a neutral alpha during crises, as seen in Panel A of Table 6. The performance differential between crisis and non-crisis phases reaches 0.4307% per month (5.17% per year) and is statistically significant. Therefore, international multi-asset funds display significantly higher alphas during down-markets than during up-markets, a result that is somewhat consistent with the findings of Comer et al. (2009b) for domestic US funds. Besides, throughout market crises, funds perform well enough (1.34% per year) to cover most of their fees, although they are still unable to outperform significantly.

With regard to factor loadings, funds display significantly lower exposures to global equities, as well as a significantly higher HML factor loading, during crises than during less troubled times. These results mean that, during market crises, funds become more focused on the European region and prefer investing in value stocks, which may hold up better during downturns because they are usually from larger well-established companies. Besides, during periods of increased volatility, we can also see considerably higher exposures to investment grade bonds and lower exposures to high-yield debt, which are slightly compatible with a flight-to-quality phenomenon.

Table 6 also reports our findings for both investment style portfolios (Panels B and C). While multi-asset funds that invest predominantly in bonds perform similarly across distinct market phases, funds that invest predominantly in equities present significantly higher alphas during crisis than in non-crisis phases. For multi-asset equity funds, differences in performance between market phases are significant at the 5% level, reaching 0.6676% per month (8.01% per year). Besides, these funds exhibit an alpha of 0.2824% per month (3.39% per year) during crises, a figure that, although not statistically significant, is economically relevant. Furthermore, this improved performance during crises is more than enough to cover their fees.

Since the performance improvements of multi-asset funds during market crises are driven by the more aggressive funds, we now focus on investigating potential justifications for what can be considered a puzzling finding. In fact, more defensive funds, with higher bond holdings, are expected to hold up better during market downturns than funds with higher equity holdings. Shifts in factor exposures between market phases could help to justify this result, though we only find a couple of statistically significant differences concerning this issue. On the one hand, Panels B and C of Table 6 show that both fund groups exhibit significantly higher exposures to value stocks in crisis than in non-crisis phases. Besides, their HML factor coefficient during crises is very similar in terms of both magnitude and statistical significance. On the other hand, while multi-asset bond funds show a similar exposure to global equities during both market phases, multi-asset equity funds not only significantly reduce this factor exposure during market downturns but, perhaps most importantly, seem to drop global equities from their portfolios during periods of higher instability. Though the more aggressive funds also display higher exposures to investment grade bonds, small caps, and stocks of firms with lower profitability in crisis than in non-crisis phases, none of these differences is statistically significant.

Another possible reason for the improved performance of multi-asset equity funds during crisis periods, which our multi-factor model may not detect, are funds’ cash holdings. In fact, more aggressive funds may hold higher proportions of cash, especially during market downturns, to be able to hedge against higher outflows.Footnote 22 To find if this is the case, we hand-collected information on the monthly cash holdings of each fund from the CMVM website and, subsequently, divided it by the fund’s TNA. Our results, presented in Table 7, show that, on average, multi-asset equity funds hold significantly higher proportions of cash than multi-asset bond funds during both crisis and non-crisis phases. However, the magnitude of these differences substantially differs across different market conditions and is more than two times higher during crisis (1.67% per month) than during non-crisis phases (0.74% per month).Footnote 23 Therefore, the more aggressive multi-asset funds in our dataset seem to reduce their exposure to global equities and increase their cash holdings during market crises, and this may help to explain their superior performance across these market phases.

Table 7 Funds’ cash holdings in crisis and non-crisis phases
Full size table

5 Conclusions

This work assessed the performance of multi-asset funds that invest internationally, during the 2004–2021 period. Since there is no consensus on the combination of bond and equity indices that are most appropriate to explain multi-asset fund returns, we followed a stepwise process to identify the relevant factors and conditional variables, to determine the most suitable model for assessing performance. We found that a conditional eight-factor model, incorporating bond, equity, default, size, book-to-market, profitability, momentum, and global equity factors, best describes the return dynamics of funds in our dataset.

For the overall sample period, our results showed that Portuguese-based international multi-asset funds underperformed significantly by − 2.62% per year. This significant underperformance holds even on a before-fee basis, reflecting fund managers’ inability to provide enough gross alphas to cover their management expenses. Funds exhibited significant exposures to investment grade and high-yield bonds, European and global equities, weak profitability stocks, and momentum strategies. In terms of investment styles, both multi-asset funds investing predominantly in bonds and multi-asset funds with a preference for equities underperformed significantly during the full sample period. However, the more defensive funds performed significantly better than their more aggressive counterparts and the difference is not justified by the lower expenses they charge.

Then, we investigated performance for crisis and non-crisis phases separately and uncovered some interesting results. Based on the “all funds” portfolio, international multi-asset funds performed significantly better during down-markets than during up-markets, by a sizeable 5.17% per year. Indeed, alphas were significantly negative during non-crisis phases (-3.82% per year), but insignificantly different from zero during crises (1.34% per year). Shifts in risk exposures can, at least to some extent, explain these results. In fact, during times of increased volatility, funds displayed a higher preference for value stocks and reduced their investments in global equities. Besides, they were also considerably more exposed to investment grade bonds and less exposed to high-yield debt. Taken together, these results indicate that, during market crises, funds tended to focus on a more restricted investment universe (Europe), preferred value investing strategies and, consistent with a flight-to-quality behaviour, reduced their exposure to default risk.

At the investment style level, we observed that multi-asset funds with higher equity holdings exhibited significantly higher alphas in crises than in non-crisis phases and by a noticeable 8.01% per year. In contrast, funds investing mostly in bonds displayed similar performance across different market phases. Therefore, our findings suggest that multi-asset funds that favour bond holdings can be a better option during non-crisis periods, while funds that invest predominantly in equities should be preferred during market downturns. A potential reason for this somewhat puzzling finding is that, on average, and especially during market crises, funds with higher proportions of equities also exhibit significantly higher cash holdings, which may help to explain their superior performance. So, our evidence shows that, in a multi-asset fund context, more attractive investment options depend not only on how fund managers weight the two main asset classes, but also on how they combine risky investments with their cash holdings.

This research is important not only within the academic community but also for financial institutions and investors interested in the factors that influence the performance of international multi-asset funds in different market conditions. Extending this work to other international multi-asset fund datasets, in order to corroborate or refute our findings, would be an interesting path for further research. Besides, another relevant topic would be to consider the market timing component of performance, using returns-based tests or approaches that employ portfolio holdings information.