Sage Journals: Discover world-class research

Abstract

We study the switching of investment options by defined contribution pension fund members using a unique data set provided by a large Australian superannuation fund that spans the market volatility associated with the COVID-19 pandemic. Switching activity appears reactive to market movements, with a spike in defensive switches during the COVID-related market sell-off and other minor market downturns, and a majority of growth switches in other periods. Switching activity has a negative impact on the balances of members who switch, which appears associated with a tendency to chase returns and switch at inopportune times. Relating switches to member characteristics reveals a higher propensity for more engaged members to switch, members nearing retirement age to make defensive switches and males to make growth switches. Our analysis can assist pension funds to formulate initiatives that may help prevent members from making detrimental switches.

JEL Codes: G11, G23, G41, G51, D14

Keywords

Individual behaviour investment switching pension funds superannuation

1. Introduction

Defined contribution pension fund members in Australia and globally often do not have access to individual security selection, but instead invest through a suite of investment options with varied asset allocations. It is well known that asset allocation decisions are a significant driver of returns for mutual fund investors (Ibbotson and Kaplan, 2000). Hence, any decision to switch investment options can have a substantial impact on welfare in retirement.

We study the magnitude, direction and impact of switching of investment options using a unique data set from 1 July 2017 to 30 June 2021 provided by a large Australian superannuation (i.e. pension) fund. The analysis period captures both relatively stable and volatile market conditions, including the substantial market drop and rebound associated with the COVID-19 pandemic. Our aim is to investigate the potential predictors of investment option switching activity and to provide guidance to pension fund providers on communications that might improve member welfare. As such, we limit our analysis to data that would typically be available to the fund provider, ignoring drivers unobserved by the fund, such as member preferences, risk tolerance and external circumstances (Roszkowski and Davey, 2010; Weber et al., 2013). Of particular interest is the impact of the COVID-19-related market volatility on switching behaviour and impact.

Our study sits within a large literature on investment choice by individuals. Our specific concern is switching between investment options by members of defined contribution pension funds. This places our study within a specific strand of this literature where there has been limited research. Our study is closest to Gerrans (2012) who analyses switching data for five Australian superannuation funds over a 2.5-year period that spans the global financial crisis (GFC) of 2008–2009. Gerrans (2012) finds that less than 7% of members switched investment options during the period, and that these members were more likely to be older, female and have higher balances. Except for the result on gender, these findings are consistent with our results. Another closely related study is by Bebbington et al. (2024) who investigate switching activity of 32,000 members of a large Australian superannuation fund from 1994 to 2019. Bebbington et al. (2024) find that less than 15% of members switched investment options over the 25-year period, that switch timing is influenced by market volatility, that males are more likely to make a first switch but females are more likely to make a second switch, and that those who made defensive switches tend to reduce their subsequent return earned. We build on the analysis of Gerrans (2012) by examining switching direction and estimating the impact of switching on member financial balances, and of both Gerrans (2012) and Bebbington et al. (2024) by investigating the relation between switching and a substantially broader set of member characteristics. The latter includes a novel approach to developing proxies for engagement, as measured by non-investment and investment-related activity indices. Other studies that examine switching between pension fund investment options include Speelman et al. (2007) and Clark-Murphy et al. (2009) in Australia; and Agnew et al. (2003) in the United States. Ameriks and Zeldes (2004) analyse allocation of assets and inflows for members of a US pension fund (rather than a specific switching decision). However, switching is not the primary focus of these studies, which address switching activity as a sideline consideration.

Two other related strands of the literature include return-chasing behaviour and the potential for adverse impacts on wealth from trading activities. These two topics have been well-addressed for individuals who invest directly in markets or mutual funds; for instance, see the work by Friesen and Sapp (2007) on return-chasing and Barber and Odean (2001) for wealth impacts from trading. Friesen and Sapp (2007) describe ‘a behavioural explanation where fund investors simply chase large recent returns and flee from low recent returns’. Existing evidence in respect to pension fund members is mixed. In Australia, a relationship between the past performance and switching between superannuation funds has been found by Frino, Heaney, and Service (2005); while Peng et al. (2020) find that superannuation fund members tend to respond to poor but not good performance. Speelman et al. (2007) and Clark-Murphy et al. (2009) find evidence of return-chasing behaviour in the investment choices of the members of four superannuation funds, although do not quantify the impact. Gan et al. (2015) examine a sample of members of a large for-profit fund and find member performance is negatively related to the switching activity. Return-chasing by Swedish pension fund members with regard to mutual funds offered as investment options has been identified by Cronqvist and Thaler (2004), whereas Dahlquist and Martinez (2015) emphasize fund member inertia, arguing that it may be costly to leave members invested in poorer performing funds. Meanwhile, Sialm et al. (2015) find only a weak relationship between past performance and mutual fund allocation decisions made by US-defined contribution fund members. Unlike prior studies using pension fund member data, we calculate an indicative impact of investment switching decisions on member balances, thus quantifying evidence of the impact of return-chasing.

The overall level of switching in our member sample is low, with 9.0% of members making at least one switch over the 4-year sample period. This is broadly consistent with the related literature (e.g. Agnew et al., 2003; Ameriks and Zeldes, 2004; Bebbington et al., 2024; Byrne et al., 2009; Gerrans, 2012; Gerrans et al., 2006). We find that both the magnitude and direction of switching activity is primarily related to market conditions and moderated by member characteristics. We observe a spike in defensive switches during the COVID-related market sell-off and other minor market downturns, and a majority of growth switches during other periods. The impact of switching is estimated to be a balance reduction of 2.4% on average from the date of the first switch observed until the end of the data period. Members who switched only during the most severe COVID-19-related market volatility destroyed the most value, with those making only one switch during that period seeing an average reduction in balance of 7.6%. We observe a tendency to add more growth exposure after equity markets have risen and vice versa, consistent with the argument that members may have been engaging in return-chasing.¹

This article proceeds at follows. Section 2 describes the data used in this study. Section 3 provides summary statistics on switching activity. Section 4 sets out the empirical design, including the logit model and variable description. Sector 5 reports results for the logit model regressions of switching intensity and direction. Section 6 presents an analysis of the impact of switching on member balances. Section 7 concludes and provides implications for pension fund providers.

2. Data

2.1. Member data

De-identified monthly cross-sectional data on member characteristics spanning 1 July 2017 to 30 June 2021 was provided by Aware Super – an Australian superannuation fund with 1.2 million member accounts and $164 billion in assets under management at the end of June 2023 (Australian Prudential Regulation Authority [APRA], 2023b). The data include investment-related details and switches, demographic characteristics, activity indicators and contribution summaries. The data include switches of the current account balance; switches relating to future contributions only were not provided and are hence ignored. The use of these data is outlined in the following sections.

Members are included in the analysis sample only if their data are complete over this 4-year period. We further narrow down the sample to include only active² account members from 1 September 2019 to 30 April 2021, which is the period over which we have sufficient data to determine this. Finally, we exclude members where self-reported gender is either not specified or changes, and where age is below 18 years of age as of 1 September 2019. Our main analysis sample comprises 641,323 unique members.³

2.2. Investment options

Aware Super offered 12 investment options over the period studied, including 5 pre-mixed (i.e. diversified) options and 7 single-asset class options. These options are listed in Table 1 for Accumulation (making contributions to their account balance) and Pension (drawing down on their account balance) members, along with their target growth asset (GA) exposure based on the strategic asset allocations stated in the product disclosure statement (PDS).

Table 1.

Investment options of Aware Super and target GA exposure.

Investment option	Target GA exposure
Investment option	Accumulation (%)	Pension (%)
Cash	0	0
Australian Fixed Interest	0	0
International Fixed Interest	0	0
Conservative Growth	25	25
Balanced Growth	57	57
Diversified Socially Responsible Investment	75	57
Growth	75	75
High Growth	95	95
Australian Equities	100	100
Australian Equities Socially Responsible Investment	100	100
International Equities	100	100
Property	100	100

Members who do not make a choice on joining Aware Super are automatically enrolled in the ‘MySuper’ Lifecycle strategy, as required by the Superannuation Industry (Supervision) Act (1993). The MySuper Lifecycle strategy in place at the time of analysis commences investing 100% of the total member balance in the Growth option, then switches the entire balance to the Balanced Growth option when the member turns 60 years of age. As these automatic switches at age 60 years are recorded in the same way as member-directed switches, we create classification rules based around age and switch type to identify automatic switches and exclude them from our analysis.

2.3. Observation periods

We analyse switching behaviour within four ‘observation periods’ of 3-month duration as set out in Table 2.⁴ The intent is to investigate how switching may differ across market conditions, while allowing some separation between the observation periods to ensure they are distinct and hence avoid any cross-over effects. The ‘pre-COVID’ period of 1 September to 30 November 2019 involved relatively stable markets and served as the baseline period used for comparison. The ‘COVID’ period from 1 February to 30 April 2020 saw equity markets suffer a sharp decline as the World Health Organization declared COVID-19 as a pandemic. The two post-COVID periods capture the recovery in equity markets to post-pandemic levels. These levels were achieved by the US S&P500 price index during the ‘post-COVID, 6 months’ period from 1 August to 31 October 2020 and the Australian S&P/ASX200 price index in the ‘post-COVID, 12 months’ period from 1 February to 30 April 2021.

Table 2.

Observation periods used to analyse switching behaviour.

Observation period (t)	Period start	Period end
Pre-COVID (1)	1 September 2019	30 November 2019
COVID (2)	1 February 2020	30 April 2020
Post-COVID, 6 months (3)	1 August 2020	31 October 2020
Post-COVID, 12 months (4)	1 February 2021	30 April 2021

Figure 1 plots the number of switches against the S&P500 and S&P/ASX200 price indices. A sharp increase in switching activity during the COVID period is clearly visible, commencing, at the same time, that global share markets started to decline, in late February 2020. Smaller spikes in switching activity also tend to be associated with other market declines (e.g. September–October 2018) and the delivery of annual statements/reports to members (e.g. August 2019).

Figure 1.

Number of switches versus equity market movements, 1 July 2017 to 30 June 2021.

2.4. Summary statistics for member sample

Table 3 presents member summary statistics for each observation period. The average balance for our member sample is consistent with industry-wide data from the Australian Prudential Regulatory Authority (APRA, 2023a). Our sample contains only 32% males reflecting the Aware Super member base, the majority of who are employees in public or healthcare sectors in New South Wales and Victoria and skewed to female-dominated occupations, such as teachers and nurses. However, the large sample size (over 600,000 members) assuages concerns over the robustness and representativeness of our findings. Most members are in the default investment option (75%–78%), with remainder either in choice options (19%–21%) or holding a pension account (3%). A small proportion of members are invested more defensively than the default option (5%–7%) or less defensively than the default option (8%–9%). See Section 4.2 for detailed descriptions of these characteristics.

Table 3.

Member characteristics at the start of each observation period.

Observation period	Pre-COVID	COVID	Post-COVID, 6 months	Post-COVID, 12 months
Period date	Sep–Nov 2019	Feb–Apr 2020	Aug–Oct 2020	Feb–Apr 2021
Members
Number of members	641,321	641,321	641,321	641,321
Average age	46.7	47.1	47.6	48.1
Males	32%	32%	32%	32%
Average account balance	$112,976	$122,344	$118,545	$129,499
Account type
Accumulation – MySuper (default)	78%	78%	76%	75%
Accumulation – choice account	19%	19%	21%	21%
Pension account	3%	3%	3%	3%
TOTAL	100%	100%	100%	100%
Risk exposure
More defensive than default	6%	5%	7%	7%
Equal to default	87%	87%	85%	84%
Less defensive than default	8%	8%	8%	9%
TOTAL	100%	100%	100%	100%

Gender and pension account status remain unchanged throughout the period.

3. Summary statistics of switches

Table 4 provides summary statistics of switch number and direction in each observation period for members who made at least one switch. Direction of switch is identified as either ‘defensive’, ‘neutral’ or ‘growth’. To calculate this, we take the difference in a member’s GA exposure prior to the first switch within an observation period and their GA exposure after the last switch in the observation period, thus capturing the net change over the period. GA exposure is calculated as the balanced-weighted average of the target GA weight of the member’s investment options (see Table 1). A switch is labelled as ‘neutral’ when the difference in GA exposure is within ± 0.5%, ‘defensive’ when the difference in GA exposure is less than −0.5%, and ‘growth’ when the difference in GA exposure is more than + 0.5%.⁵ For example, a switch by a member with a pension account from the Balanced Growth option to the Balanced Diversified Socially Responsible Investment option is classified as ‘neutral’ as both pre-mixed options have target GA exposure of 57%, thus leaving their target GA exposure unchanged.

Table 4.

Members switching in each observation period by direction and selected characteristics.

Observation period	Pre-COVID		COVID		Post-COVID, 6 months		Post-COVID, 12 months		Average of all periods
By switch number and direction	Number	%	Number	%	Number	%	Number	%
Defensive switch	1,390	26	17,768	81	1,957	33	1,462	17	54%
Growth switch	3,638	68	2,852	13	3,555	59	6,799	78	40%
Neutral switch	293	6	1,225	6	504	8	408	5	6%
Total members switching	5,321	100	21,845	100	6,016	100	8,665	100	100%
% of members	0.8		3.4		0.9		1.4		1.6
	% of members		% of members		% of members		% of members
By age (years)
<40	0.7		2.4		0.7		1.1		1.2%
40–54	0.6		2.7		0.7		1.1		1.3%
55–70	1.3		5.6		1.5		1.9		2.6%
>70	1.1		3.7		1.1		1.9		1.9%
By account balance
Low	0.4		1.5		0.5		0.6		0.8%
Medium	0.7		3.0		0.7		1.2		1.4%
High	1.4		5.5		1.5		2.0		2.6%
By gender
Female	0.6		2.8		0.7		1.0		1.3%
Male	1.3		4.6		1.4		2.0		2.3%
By account type
Accumulation – MySuper (default)	0.3		2.0		0.3		0.3		0.7%
Accumulation – choice account	2.6		8.5		3.1		4.6		4.7%
Pension account	2.3		7.6		2.7		3.4		4.0%
By relative risk exposure
More defensive than default	4.5		9.2		6.2		10.9		7.7%
Equal to default	0.5		2.4		0.4		0.5		0.9%
Less defensive than default	2.2		10.7		2.3		2.3		4.4%

Account balance is divided into three groups according to the values in each period: ‘Low’ – up to $30,000, ‘Medium’ – between $30,001 and $120,000, and ‘High’ – $120,001 and above. The groups are divided such that each group consists of approximately one-third of the observations.

There are four main takeaways from Table 4. First, switching activity is not widespread, averaging only 1.6% of members across all quarterly periods. Second, 3.4% of members switched during the COVID period, which is substantially higher than the other three periods. Third, switching activity varies noticeably with member characteristics. Switching activity is much lower for younger members, members with lower balances, females versus males, default members versus members with choice or pension accounts, and members invested less rather than more defensive than the default (except during the COVID period).⁶ Fourth, the majority of switches during the COVID period were defensive switches (81%), while the majority of switches made during the other three periods were growth switches (59%–78%).

4. Empirical design of multivariate regression analysis

We expand the univariate analysis of Section 3 to perform multivariate analysis that relates switching activity to member characteristics. Section 4.1 sets out the method. Section 4.2 describes the characteristics incorporated into the analysis. Section 4.3 outlines how we form two activity indices for non-investment and investment-related activity by combining selected member data.

4.1. Model structure

We estimate logit models with time fixed effects to relate switching activity to member characteristics and the market environment as captured by observation periods. The model allows for the relation between switching behaviour and member characteristics to vary across the four observation periods. The model is described by equation (1)

logit (p (Y_{i t})) = μ + {X^{'}}_{i t} β + I_{i 2} (μ_{2} + {X^{'}}_{i t} β_{2}) + I_{i 3} (μ_{3} + {X^{'}}_{i t} β_{3}) + I_{i 4} (μ_{4} + {X^{'}}_{i t} β_{4}) = η_{i t}

(1)

where $p (Y_{i t}) = \Pr (Y_{i t} = 1 | t, X_{i t})$ is the probability of observing the outcomes of interest for member $i$ in observation period $t$ = 1, ..., 4 (see Section 2.3); given the member explanatory characteristic vector $X_{i t}$ . $I_{i j}$ is an indicator variable equal to 1 if j = t and 0 otherwise. $μ$ is the effect for baseline time period t = 1 and $μ_{j}$ are the marginal effects for time period j. $β$ is the regression coefficient vector of the member characteristics in baseline time period t = 1 and $β_{j}$ are the regression coefficient vectors of marginal effects of the member characteristics at time period j. Explanatory characteristics are outlined in Section 4.2.

$Y_{i t}$ is one of the two outcomes of interest:

Switch – Indicator variable equal to 1 for a member who has made at least one switch during the observation period, and 0 otherwise. This analysis uses the full member sample.

Direction – Indicator variable equal to 1 for a member who has made an overall growth switch during the observation period, and 0 otherwise. This analysis uses the sample of members who made at least one switch altering their GA exposure.⁷

4.2. Explanatory variables – member characteristics

We investigate characteristics that were available to Aware Super – and hence many pension fund providers – that may explain switching activity and direction. The eight explanatory variables used in the modelling are listed and discussed below. See Table 4 for the potential influence of some of these variables from a univariate perspective. Unless stated otherwise, variables are observed or estimated at the start of the observation period.

Age – An indicator variable for a member who is aged 55–70 years.⁸

Account balance – Account balance is continuous, log-transformed and normalized with mean of zero and standard deviation of one.

Gender – Gender is captured by the male indicator variable.

Use of choice options – The choice account indicator variable identifies accumulation members who are not entirely invested in the MySuper default, capturing accumulation members who have previously made a non-default investment decision.

Pension account – The pension account indicator variable identifies members who are in the decumulation phase. Pension members must choose to open their account⁹ and hence the default investment strategy is not applicable for this group.

Relative risk exposure – The variable difference in GA is a continuous variable of the difference between the GA exposure across the member’s investments and the GA exposure of the MySuper default (see Section 2.2). For pension account members, we reference the MySuper default GA exposure.

Non-investment activity – A non-investment activity index is constructed to measure member activity by combining data on beneficiary nominations, communication activity and various cash flow activities. Section 4.3.1 describes the estimation of the index, which is continuous and normalized to a mean of zero and standard deviation of one.

Investment activity – An investment activity index is constructed to measure willingness to partake in investment activities by combining data on prior switching activity and the number of investment options used. Section 4.3.2 describes the estimation of the index, which is continuous and normalized to a mean of zero and standard deviation of one.

Table 5 reports correlations between the explanatory variables in the baseline pre-COVID period, noting there is no material difference in the correlations across the four observation periods. While all correlations are statistically significant, this is due to the large sample size. The most material collinearity observed is the 0.494 correlation between the choice account and investment activity index. This mostly reflects that investment activity index incorporates the number of investment options held, and that those with more than one investment option must also be choice account members. Choice account is a member characteristic that a fund would likely use to segment members for communication purposes and so is maintained separately in the model. Taking the large sample size into account, we are not concerned with the multicollinearity seen in Table 5.

Table 5.

Correlation between explanatory variables in the pre-COVID observation period.

	Age 55–70 years	Account balance	Male	Choice account	Pension account	Difference in GA	Non-investment activity index	Investment activity index
Age 55–70 years	1.000
Account balance	0.099***	1.000
Male	0.006***	0.036***	1.000
Choice account	0.220***	0.170***	0.080***	1.000
Pension account	0.154***	0.157***	0.015***	−0.085***	1.000
Difference in GA	−0.053***	−0.003*	0.020***	−0.053***	−0.085***	1.000
Non-investment activity index	0.138***	0.312***	0.024***	0.262***	0.117***	−0.007***	1.000
Investment activity index	0.096***	0.129***	0.079***	0.494***	0.119***	−0.081***	0.200***	1.000

Significance (two-tailed) is indicated by *p < 0.1, **p < 0.05, ***p < 0.01.

4.3. Construction of activity indices

Deetlefs et al. (2019) find that more engaged members are more likely to make non-default investment choices, although Bateman et al. (2014) find non-investment activities to be an unreliable proxy for interest in superannuation. The data supplied by Aware Super provide a range of member activity indicators that may indicate a level of engagement or possible interest in switching. However, the influence of these particular indicators is difficult to interpret when included individually as explanatory variables, in part due to significant collinearity. We hence combine the indicators to generate two overall measures of activity – a non-investment activity index and an investment activity index – that might be respectively interpreted as proxies for the overall level of engagement with superannuation in general and investment decisions more specifically.

To do this, we describe in this section the use of models from item response theory (IRT), also known as latent response theory. This method refers to a family of mathematical models that seek to explain the relationship between latent traits (e.g. the unobservable engagement level of the members) and their manifestations (e.g. observed actions of the members).¹⁰ Xue et al. (2019) apply IRT in examining financial literacy among elderly Australians.

4.3.1 Non-investment activity index

The non-investment activity index is generated using the two-parameter logistic model (2-PL) model. The general form of the 2-PL model is described by equation (2)

P_{i k} (r_{i k} = 1 | θ_{i}, α_{k}, β_{k}) = \frac{e^{α_{k} (θ_{i} - β_{k})}}{1 + e^{α_{k} (θ_{i} - β_{k})}}

(2)

where $r_{i k}$ is an indicator variable that the $i t h$ member has undertaken item (activity) and $θ_{i}$ is the non-investment activity index of the $i t h$ member. Activity indices have an average value of zero, with more ‘active’ members having an index value greater than zero and vice versa. $β_{k}$ is the ‘difficulty’ parameter for the $k t h$ item, reflecting the frequency with which the activity is undertaken in the data set. From the right-hand side of equation (2) it can be seen that $β_{k}$ represents a non-investment activity index $θ$ where there is a 50% likelihood of undertaking the $k t h$ activity (i.e. $β_{k} = θ$ ). $α_{k}$ is the ‘discrimination’ parameter for the $k t h$ item, which represents the extent to which the $k t h$ item determines the non-investment activity index. The IRT model is used to estimate $β_{k}$ and $α_{k}$ using maximum likelihood. These in turn support identification of the latent variable $θ_{i}$ that coincides with the observed member actions, which is expressed as an index value for use in the logit model described in Section 4.1.

The non-investment activity index combines the five activity indicators listed below. Dates for the time period that the indicator is observed are also shown,¹¹ noting that the data provided for these activities commence on 1 July 2019 and are limited to the time periods described below:

Nominated beneficiaries (observed by 28 February 2021)

Registered on the online portal (observed by 28 February 2021)

Chose to receive email instead of paper-based communication (observed by 28 February 2021)

Made at least one phone call (observed between 1 July 2019 and 30 April 2021)

Undertaken at least one discretionary cash flow activity, including a salary sacrifice contribution, a voluntary contribution or a pension drawdown (observed between 1 July 2019 and 30 April 2021).

To extract values of the non-investment activity index for members using the fitted 2-PL model, we first obtain 2⁵ = 32 unique activity index values. Based on their observed set of actions, each member is assigned 1 of the 32 unique values, which are normalized and range from −1.01 to 1.63. A member with value −1.01 is observed to undertake none of the above activities, while a member with value 1.63 is observed to undertake all of the above activities.

Figure 2 depicts the relationship graphically across the range of possible index values. The activities that are ‘difficult’ (i.e. those with lower rates of occurrence and hence higher $β_{k}$ ) have lower probabilities. For instance, a member with average activity levels (i.e. non-investment activity index of zero) has a probability of about 88% of having registered on the portal (blue line) and 16% of some discretionary cash flow activity (black line). A highly active member with a latent activity score of 1.63 would have about a 100% chance of having registered on the portal and about 45% chance of some discretionary cash flow activity. Actions with steeper curves have higher ‘discriminating’ power (hence higher $α_{k}$ ), meaning that they exert higher influence on the activity index. For instance, being registered on the portal has higher discriminating power than making a phone call, as engaged members are more likely to register for the portal, while making a phone call is undertaken by both engaged and disengaged members.

Figure 2.

Estimated probability of member engaging in non-investment activities by estimated activity index.

4.3.2 Investment activity index

The investment activity index is generated by combining past investment activities into a single indicator using the graded response model (GRM) as developed in the work by Samejima (1969). GRM is an extension of the two-parameter logistic model (2-PL) that facilitates converting continuous variables into ordered categories (also known as discretizing, or binning). The activities listed below represent investment activities over and above a member who made a single investment option choice on entry to the fund several years previously. As discussed in Section 4.2, we keep ‘choice account’ as a separate explanatory variable:

Number of switches made in the past 12 months,¹² which is converted into five ordered categories of 0, 1,2, 3–5 and 6 or more switches.

Number of investment options held,¹³ which is converted into five ordered categories of 1, 2, 3, 4 and 5 or more options held.

Discretizing each of the variables into five categories creates 25 groups (i.e. 5²) and separates out the few members with a very high number of past switches or high number of options into distinct groups. Unlike the non-investment activity index, the investment activity index is updated at the start of each observation period.

Table 6 reports the percentage of total observations (i.e. members by observation periods) in Panel A and the associated investment activity index values for each of the 25 groups. It is worth noting that the largest group (91.52% of observations) has the lowest activity index (−0.13), and by design includes MySuper ‘default’ accumulation members who hold one investment option and have made zero switches and members who made a switch to a single investment option more than 12 months prior. Investment activity index values sit at 1.12 or above for all other groups, with a maximum of 3.17 for members that are in the most active category on both indicators.

Table 6.

Percentage of members and investment activity index values by number of switches and number of investment options.

Panel A: Percentage of members
Number of switches made in the past 12 months	Number of investment options					TOTAL
Number of switches made in the past 12 months	1	2	3	4	5 or more
0	91.49%	2.88%	0.85%	0.37%	0.38%	95.97%
1	1.88%	0.61%	0.26%	0.11%	0.11%	2.96%
2	0.37%	0.15%	0.07%	0.03%	0.04%	0.66%
3–4	0.14%	0.07%	0.03%	0.02%	0.02%	0.28%
5 or more	0.07%	0.03%	0.02%	0.01%	0.01%	0.14%
TOTAL	93.95%	3.75%	1.22%	0.53%	0.55%	100.00%
Panel B: Investment activity index
Number of switches made in the past 12 months	Number of investment options held
Number of switches made in the past 12 months	1	2	3	4	5 or more
0	−0.13	1.12	1.25	1.32	1.37
1	1.38	1.92	2.09	2.18	2.27
2	1.58	2.16	2.36	2.48	2.62
3–4	1.67	2.29	2.52	2.67	2.86
5 or more	1.74	2.40	2.67	2.87	3.17

5. Regression results

This section presents results from the regression analysis described in Section 4, with switching activity in Section 5.1 and switching direction in Section 5.2.

5.1. Switching activity

Table 7 presents regression outputs with switch as the dependent variable. The baseline member has the following characteristics: not aged 55–70 years, an average (log) account balance, female, invested in the MySuper accumulation default (i.e. not a choice account or pension account member), exposure to GA in line with the default, and an average non-investment activity and investment activity index value. Coefficient estimates for observation periods 2 (COVID), 3 (Post-COVID 6 months) and 4 (Post-COVID 12 months) are marginal effects relative to the Pre-COVID baseline across periods and member characteristics. The grey text below the coefficient estimates shows the total effect expressed as a probability of switching in a given quarter, estimated by combining the baseline and period effects and member characteristic effects for the baseline and that period. For example, equation (1) estimate for a male member with other characteristics in line with the baseline member during COVID is: period effect + member characteristic effect = (−5.822 + 1.675) + (0.435 − 0.104) = −4.147 + 0.332 = −3.815. Conversion into the probability of a switch is calculated as exp[−3.815] / (1 + exp[−3.815]) = 2.2%.

Table 7.

Regression results – switch as dependent variable.

Observation period	Pre-COVID	COVID	Post-COVID,6 months	Post-COVID,12 months
	Baseline	Marginal effects versus the baseline
Period effects	−5.822***	1.675***	−0.104	0.147**
Estimated probability	0.3%	1.6%	0.3%	0.3%
Member characteristic effects
Aged 55–70 years	0.022	0.406***	0.223***	0.067
Estimated probability	0.3%	2.4%	0.3%	0.4%
Account balance	0.141***	0.165***	−0.060	0.073
Estimated probability	0.3%	2.1%	0.3%	0.4%
Male	0.435***	−0.104*	−0.025	−0.054
Estimated probability	0.5%	2.2%	0.4%	0.5%
Choice account	0.463***	−0.146*	−0.276***	0.020
Estimated probability	0.5%	2.1%	0.3%	0.6%
Pension account	0.665***	−0.348***	−0.097	0.004
Estimated probability	0.6%	2.1%	0.5%	0.7%
Difference in GA	−1.742***	1.854***	−0.059	−1.386***
Estimated probability	0.2%	1.6%	0.2%	0.2%
Non-investment activity index	0.645***	−0.090***	−0.164***	−0.250***
Estimated probability	0.6%	2.7%	0.4%	0.5%
Investment activity index	0.419***	−0.082***	0.199***	0.139***
Estimated probability	0.4%	2.2%	0.5%	0.6%
Number of observations	641,321	641,321	641,321	641,321
(Members / period)

Significance (two-tailed) is indicated by *p < 0.05, **p < 0.01, ***p < 0.001. See Section 4 for model description. The grey text below the coefficient estimates shows the estimated probability of switching for a baseline member with that characteristic in that period; see the discussion prior to Table 7 for an example calculation. The effect of account balance, and the two activity indices are expressed as one standard deviation above the baseline. The effect of difference in GA is expressed as 20% above the default level, similar to the differences between the Balanced Growth, Growth and High Growth options in Table 1.

Looking first at period effects, the pre-COVID effect of −5.822 gives an estimated probability of switching of 0.3% in the pre-COVID period for a member with baseline characteristics. The COVID effect of 1.675 shows, for the baseline member, a significant increase in the estimated probability of switching during the COVID period to 1.6%. Relatively minor impacts are observed in the two post-COVID periods. Thus, the baseline propensity to switch, prior to taking into account member characteristics, spikes during COVID but is relatively consistent across the other three periods.

We discuss the member characteristic effects below, noting that we are referring to members who hold the characteristic in question but are in line with the baseline on all other characteristics:

Age 55–70 years – This characteristic is unrelated to the propensity to switch pre-COVID. However, the 0.406 estimate of the marginal effect during COVID suggests that age 55–70 years is associated with a higher propensity to switch over and above the period impact during COVID. The estimated probability of switching for a member who is age 55–70 years is 2.4% during COVID, compared with 1.6% for baseline members during COVID. Prior survey evidence suggests that members who are nearing or at retirement may be more prone to switch because they are more engaged and perhaps more risk averse (Bateman et al., 2011). Our results are consistent with this propensity manifesting specifically during the market uncertainty associated with the COVID pandemic.

Account balance – Members with a higher than average account balance had a higher propensity to switch than the baseline member during all observation periods, consistent with the findings of Agnew et al. (2003) and Gerrans (2012). A potential reason is that members with larger balances may be more engaged due to their superannuation balance forming a more meaningful portion of their wealth. The most material impact is observed in the COVID period, with a marginal effect estimate of 0.165 and an estimated probability of switching of 2.1% for a member with a (log) account balance that is one standard deviation above average, as compared to 1.6% for baseline members during COVID.

Male – Males have a higher propensity to switch than females across all periods, with a coefficient estimate of 0.435 pre-COVID and immaterial marginal effect estimates in other periods. This is consistent with the results of Agnew et al. (2003) but opposite to that of Gerrans (2012). This corresponds with an estimated probability pre-COVID of switching of 0.5% for male members compared to 0.3% for baseline (female) members.

Choice account and pension account – Both choice account and pension account members have a higher propensity to switch than baseline members across all periods, with pre-COVID coefficient estimates of 0.463 and 0.665, respectively. However, the respective baseline pre-COVID effects are dampened during COVID (marginal effect estimates of −0.146 and −0.348, respectively), and in post-COVID 6 months for choice account (marginal effect estimate of −0.276). As an example of this dampening impact, for the baseline and pension account members respectively, the estimated probability of switching doubles from 0.3% to 0.6% in pre-COVID and has a relatively smaller increase from 1.6% to 2.1% in COVID. This indicates that the market volatility associated with the COVID period, and additional time available due to the national lockdown commencing on 22 March 2024,¹⁴ may have incited some previously disengaged members to become more engaged with their superannuation.

Difference in GA – The coefficient estimate of −1.724 for the pre-COVID period reflects the lower propensity to switch during that period for members with a GA bias, with an estimated probability of 0.3% and 0.2% for a baseline member and one with a difference in GA weight that is 20% above the default, respectively. The COVID period marginal effect estimate of 1.854 directly offsets the pre-COVID estimate, meaning that current asset allocation has little effect on switching activity (1.6% estimated probability for both groups) during COVID. The significant negative marginal effect estimate for the post-COVID 12 months period reflects an even greater impact of asset allocation during this period compared to pre-COVID.

Non-investment and investment activity indices – The respective pre-COVID coefficient estimates of 0.645 and 0.419 for these two variables indicate that more active members have a higher propensity to switch, consistent with the results of Deetlefs et al. (2019). For example, pre-COVID there is an estimated probability of switching of 0.6% for members with a non-investment activity index that is one standard deviation above average compared to 0.3% for baseline members. The marginal effect estimates relative to the pre-COVID values indicate that the impact of these two indices differs in intensity across periods, but not in a material way. Noting that the non-investment activity index, investment activity index and account balance are all normalized variables, the coefficient estimates suggest that the activity indices are more significant predictors of switching activity than account balance, especially when considered together as two related indicators of engagement.

The main finding from Table 7 is that the most influential factor for switching activity over our analysis period is the period itself, with the period impact being particularly significant during COVID. Member characteristics also have an additional impact over and above any period effects, with higher switching activity associated with members with age 55–70 years (i.e. around retirement, for COVID and post-COVID 6 months only), higher account balance (all periods), male (all periods), invested in non-default (i.e. choice account or pension account, with magnitude of impact varying across periods) and have tended to be more active in terms of both the investment activity index and non-investment activity index (all periods).

5.2. Switching direction

Table 8 presents regression outputs with direction as the dependent variable. Interpretations of the model structure are comparable to Section 5.1. We follow a similar structure in discussing the parameter estimates.

Table 8.

Regression results – direction as dependent variable.

Observation period	Pre-COVID	COVID	Post-COVID,6 months	Post-COVID,12 months
	Baseline	Marginal effects versus the baseline
Period effects	1.219***	−3.293***	−1.052***	−0.053
Estimated probability	77.2%	11.2%	54.2%	76.2%
Member characteristic effects
Aged 55–70 years	−0.909***	−0.023	−0.236	−0.073
Estimated probability	57.7%	4.7%	27.4%	54.6%
Account balance	−0.106	−0.235***	−0.010	0.187*
Estimated probability	75.3%	8.2%	51.3%	77.7%
Male	0.251**	0.289**	0.144	−0.004
Estimated probability	81.3%	17.8%	63.7%	80.4%
Choice account	0.073	−0.151	−0.298	−0.227
Estimated probability	78.5%	10.4%	48.6%	73.3%
Pension account	−1.505***	1.484***	1.303***	0.674**
Estimated probability	42.9%	11.0%	49.1%	58.3%
Difference in GA	−6.857***	1.756***	−0.657	−2.045***
Estimated probability	46.2%	4.3%	20.8%	35.1%
Non-investment activity index	−0.085	0.200**	0.502***	0.399***
Estimated probability	75.7%	12.4%	64.2%	81.5%
Investment activity index	0.076**	0.063*	−0.132***	−0.033
Estimated probability	78.5%	12.6%	52.8%	77.0%
Number of observations	5,028	20,620	5,512	8,261
(Members / period)

Looking first at period effects, the pre-COVID effect of 1.219 implies an estimated probability of a growth switch of 77.2% in the baseline period for a member with baseline characteristics. The COVID effect of −3.293 reveals a large reduction in this probability during COVID, implying an estimated probability of a growth switch of 11.2% for a member with baseline characteristics. A potential reason for this outcome could be a reduction in risk tolerance during the COVID market volatility, consistent with prior studies that found risk tolerance reduced in response to the GFC (Bateman et al., 2011; Gerrans et al., 2015). However, there is some debate over whether the primary influence is related to changes in return expectations or risk perceptions with the market environment, with risk tolerance remaining a relatively stable attribute (Hoffmann et al., 2013; Hoffman & Post, 2015; Roszkowski and Davey, 2010; Weber et al., 2013). Compared to pre-COVID, there is a reduction in the estimated probability of a growth switch to 54.2% for the baseline member associated with the post-COVID, 6 months effect of −1.052, but an insignificant impact in the post-COVID, 12 months period. We discuss the impact of these trends in switching direction in Section 6.

We discuss the member characteristic effects below, noting again that we are referring to members who hold the characteristic in question but are in line with the baseline on all other characteristics:

Age 55–70 years – Members in this age group have a lower propensity for growth switches across all periods. This is consistent with prior literature indicating that GA exposure in the context of pension funds is related to age, although there is some debate over whether this reflects participant choice versus other influences (see Agnew et al., 2003; Ameriks and Zeldes, 2004; Bateman et al., 2011; Gerrans et al., 2010).¹⁵ The pre-COVID coefficient estimate of −0.909 is equivalent to an estimated probability of a growth switch of 57.7%, compared with 77.2% for baseline members. No significant marginal effects are observed in other periods. The lack of significant marginal effects is consistent with market conditions and age having no meaningful interactive impact on likely switch direction.

Account balance – A significant marginal effects estimate of −0.235 during COVID indicates that members with a higher account balance had a lower propensity to make a growth switch during the COVID period, and thus were even more likely to make a defensive switch than members with lower balances. The estimated probability of a growth switch during COVID for someone with a (log) account balance one standard deviation higher than average is 8.2%, compared to 11.2% for baseline members. No material impact is observed for account balance in other periods.

Male – Males have a higher propensity to make a growth switch relative to females across all periods. The coefficient estimate of 0.251 in pre-COVID implies that males have an 81.3% probability of making growth switches in that period, versus 77.2% for female baseline members. This propensity was exacerbated during COVID where a marginal effect estimate of 0.289 leads to an estimated probability of a growth switch for males of 17.8%, compared to 11.2% for female baseline members, indicating that males may be less likely than females to be influenced by market conditions in terms of switch direction. Prior studies have shown that gender is correlated with willingness to accept financial risk (see Gerrans et al., 2006, in the context of Australian superannuation fund members), and a range of behavioural attributes that might influence switching behaviours, such as overconfidence (e.g. Barber and Odean, 2001). Our results are consistent with these findings.

Choice account – This characteristic has no material impact on switch direction.

Pension account – The baseline pre-COVID coefficient estimate of −1.505 is substantial and indicates that members with a pension account were much less likely to make a growth switch in pre-COVID, with an estimated probability of a growth switch of 42.9% compared to 77.2% for baseline members. However, this estimate is almost fully offset by marginal effect estimates during COVID and post-COVID 6 months of 1.484 and 1.303, respectively, and partly offset by a marginal estimate during post-COVID 12 months of 0.674. For example, the estimated probability of a growth switch in post-COVID 6 months is 49.1% for members with a pension account, compared to 54.2% for baseline members. Thus, while pension account members reveal a tendency to de-risk relative to other members, their behaviour was similar to other switchers in COVID and post-COVID 6 months.

Difference in GA – The coefficients on this variable need to be interpreted carefully due to a natural boundary condition. Specifically, members invested 100% in GAs (and hence having the largest possible positive difference in GA can only make a defensive switch, and vice versa. This helps explain the baseline pre-COVID estimate of −6.857, which implies that members with a high weighting to defensive assets were more likely to make growth switches, and vice versa. For a baseline member with GA allocation 20% higher than the default, this leads to an estimated probability of a growth switch in pre-COVID of 46.2% compared to 77.2% for baseline members. After considering period effects, the impact of difference in GA is strongest in post-COVID, 12 months and weakest in COVID.¹⁶

Non-investment and investment activity indices – The estimates point towards some propensity for growth switches by members with a higher non-investment activity index in the two post-COVID periods, with marginal effect estimates of 0.502 and 0.399 for post-COVID, 6 months and post-COVID, 12 months, respectively. For example, in post-COVID, 6 months, the estimated probability of a growth switch is 64.2% for a baseline member with non-investment activity index one standard deviation higher than average, compared to 54.2% for baseline members. The impacts for the investment activity index are not material and inconsistent over periods and between indices. While Table 7 reveals a clear tendency for more active members to be more likely to engage in switching, the results of Table 8 suggest that the activity indices are less helpful for understanding the likely switch direction.

Similar to the propensity to switch results in Section 5.1, the results for switch direction suggest that period tends to dominate member characteristics in determining the likelihood of members making switches in either a growth or defensive direction. A number of the characteristics were found to have no clear relation with switch direction including account balance (outside of COVID), choice account and the investment activity index. Overall, only certain member characteristics appear to have a readily interpretable impact on switch direction. Members nearing or in retirement tend to make defensive switches relative to other members, with age 55–70 years having an impact in all periods and pension account in the pre-COVID and post-COVID, 12 months period. Gender is also influential, with female members tending to make defensive switches relative to other members. These results suggest that member characteristics may be of limited assistance in helping to predict with confidence whether a particular member is much more likely to make, for example, a defensive switch in response to market volatility than other members.

6. Impact of switching on member balances

We now examine the impact of switching activity through estimating a proxy for the percentage change in balance that arises from switching activity. We also undertake regression analysis to relate the impact of switching to the broader set of member characteristics and switching periods. The analysis should be viewed as indicative only. We do not account for differences in risk or the impact of contributions or withdrawals. Furthermore, we are unable to gauge if members may be making rational switching decisions in response to changes in circumstances, objectives or preferences. Our aim is to capture whether switching has impacted on balance over a relatively short and volatile investment period.

6.1. Sample

For this analysis, we make use of data for the full 4-year data period from 1 July 2017 to 30 June 2021. We examine the 57,744 members (9.0% of the member sample) who made at least one switch. Calculating the impact on member welfare of switching activity is complicated by the fact that our data allow us to observe only the dollar value of switches, but not the portion of the total portfolio that is switched at a particular point of time. We also do not observe all cash flows, ruling out a portfolio tracking approach. Given these data hurdles, we calculate the dollar impact of each switch from the time of switching¹⁷ to the end of the sample period (30 June 2021). The impact of an observed switch is estimated for each member as the difference in accumulated dollar-value that arises from investing in the option(s) chosen less the dollar-value that would have arisen if the original option(s) had been retained. We sum these dollar-value impacts across all observed switches made by a member to calculate the total dollar impact of all switches for that member. To scale this total dollar impact, we divide by the maximum observed account balance over the analysis period adjusted for the total dollar impact, to give an impact of switches expressed as percentage. Using the maximum account balance for scaling generates a conservative estimate of the percentage impact of switching activity.

To evaluate the impact of switches made during different market phases, we partition the entire data into time of switch based on three windows: prior COVID from 1 July 2017 to 31 January 2020, COVID from 1 February to 30 April 2020, and after COVID from 1 May 2020 to 30 June 2021.¹⁸ Table 9 describes the sample breakdown based on member account type and the windows during which the member switched. The member account type is measured at the time of the first switch; hence, default members were in the MySuper investment option before this switch. The default group is singled out for closer attention on the basis that default members are likely to be less engaged or less informed than non-default members (see the work by Butt et al., 2018) to gauge if this group is associated with poorer switching decisions. In terms of the total sample, the largest group is the 37% of members who switched only during prior COVID, which reflects the longest time period covered. However, 22% of members switched only during after COVID, while 13% switched only during COVID. The remaining 28% of members switched across multiple windows. Default members who switched were more likely to only switch during the COVID or after COVID periods than choice account and pension account members.

Table 9.

Member groups based on time of switching.

Group	When switched	Members (Total)		Default		Choice Account		Pension Account
Group	When switched	Number	%	Number	%	Number	%	Number	%
1	Prior COVID only	21,623	37	11,966	37	8,675	38	982	37
2	COVID only	7,353	13	5,249	16	1,961	9	143	5
3	After COVID only	12,798	22	8,393	26	4,064	18	341	13
4	Prior COVID and COVID	1,674	3	585	2	979	4	110	4
5	Prior COVID and after COVID	3,514	6	1,146	4	2,013	9	355	13
6	COVID and after COVID	6,806	12	3,870	12	2,609	11	327	12
7	Switched during all periods	3,976	7	1,005	3	2,587	11	384	15
	Total	57,744	100	32,214	100	22,888	100	2,642	100

6.2. Summary statistics of impact of switches

Table 10 reports the average impact of switches for the member groups listed in Table 9, further categorized into members who made single versus multiple switches in Panel B and whether their activity comprised a switch during COVID or during non-COVID periods only in Panel C. In general, switching activity leads to a loss of value for members. The average impact of switches across all members who switched is −2.4%, which represents a 2.4% loss of balance due to switching activity. The only members to increase their balance through switching activity were choice account and pension account members who switched during non-COVID periods only (see Panel C); in particular, those who switched only once during the prior COVID period (see Panel B). Members who switched during COVID generally fared poorly, with the impact of switches averaging −5.6% (see Panel C). This is unsurprising given the large majority of defensive switches during COVID (see Table 4) and the subsequent market rebound. The worst performing group were members who switched once only during COVID and never switched again, for whom the impact of switches averages −7.6% (see Panel B).

Table 10.

Impact of switches by member group and time of switching.

Green shading indicates a positive impact. Yellow shading indicates a reduction smaller than −5%, and orange a reduction of more than −5%.

Default members fared worse than choice account and pension account members within every sub-group analysed. This aligns with the idea that default members may be less engaged and informed, and hence may make poorer financial decisions (Lusardi, 2012). Pension account members fared slightly better than choice account members on average.

Care needs to be taken not to over-emphasize the value lost with respect to switches made during the COVID window, given that it was a single market event of a somewhat unusual nature given the relatively abrupt market decline and recovery. If the market had continued to fall further before recovering, perhaps because the policy reaction was less extreme or the pandemic turned out to be even more severe, then results may have been different. More compelling is the finding that switching activity outside of the COVID window also led to reduced balances. Table 4 indicates that most switches made during the observation periods outside of the COVID period were growth switches. Nevertheless, with notable exception of choice account and pension account members that switched in the prior COVID window, switches made during prior COVID and after COVID period acted to reduce member balances on average. This is consistent with switching activity generally being value-destroying for the typical member, indicating that our findings are not solely a function of the large transitory market decline that occurred during this particular sample period.¹⁹

Figure 3 provides an indication of the underlying source of consistent value loss by overlaying net switching behaviour on the S&P/ASX200 (Australian share) price index. This chart plots net direction switch numbers during weekly periods, with net defensive switches in red and net growth switches in blue. Darker shades indicate a greater magnitude of net switching, with ‘high’, ‘medium’ and ‘low’ each representing one-third of the total observations of each colour. It illustrates how the direction of switching activity not only moves with the market direction but also increases in intensity as the market approaches a near-term turning point. Defensive switches peak during and at the bottom of market downturns, while growth switches are more likely during a rising market. This suggests a tendency for switches to often occur at inopportune times within each sub-period and is consistent with the return-chasing behaviour described in the work by Friesen and Sapp (2007).²⁰

Figure 3.

Net switching direction and Australian equity market.

6.3. Regression analysis of impact of switches

We undertake regression analysis to investigate the relation between impact of switches and member characteristics, and the windows during which members switched. Member characteristics are the same as applied in Section 5, except that the investment activity index is removed since it cannot be calculated for switches occurring in the first 12 months of the data.²¹ The window during which members switched, as defined in Table 9, are included as fixed effects. Equation (3) describes the regression

I o S_{i} = {I^{'}}_{i} δ + {X^{'}}_{i} β + ϵ_{i}

(3)

where $I o S_{i}$ is the impact of switches for member $i$ who switched at least once in the period of 1 July 2017 to 30 June 2021, multiplied by 100 to scale the coefficient estimates to a percentage magnitude; $I_{i}$ is a vector of indicator variables indicating the period in which switches were made; $δ$ is the regression coefficient vector for period in which switches were made; $X_{i}$ is the member characteristic vector, calculated at the time of the first switch; $β$ is the regression coefficient vector for member characteristics and $ϵ_{i}$ is the error term. The regression coefficient vector $δ$ does not assume a baseline, but rather by the period effects being mutually exclusive and exhaustive, the coefficient estimates are uniquely identified as the impact from that period.

The results are presented in Table 11. The negative coefficient estimates for the period effects reflect value-destroying switches for all periods, while their relative values confirm that members who switched during COVID tended to destroy more value. Meanwhile, the estimate for members who switched in prior COVID only is relatively small and is consistent with the small loss in prior COVID only in Table 10. Member characteristic effects show a higher propensity for value-destroying switches by members who were aged 55–70 years, possessed a higher account balance, and had a positive difference in GA (i.e. GA exposure greater than the MySuper default). These characteristics are associated with members that were more likely to make defensive switches as indicated by the results in Table 8. There is a higher propensity for value-creating switches by members who were male, held a choice account or pension account and had a higher non-investment activity index. The latter two results suggest that members who are likely to be more engaged with their superannuation were better at switching. The male results reflect a greater tendency for males to make growth switches relative to females as indicated in Table 8. The relative values of the period effects compared to the member characteristic effects show that the period has a much larger influence on impact of switches than the member characteristic effects.²² This is consistent with the results from Section 5.

Table 11.

Regression results – impact of switches as dependent variable.

Variable	Coefficient estimate
Period effects
Prior COVID only	−1.658***
COVID only	−7.267***
After COVID only	−2.221***
Prior COVID and COVID	−5.601***
Prior COVID and after COVID	−3.260***
COVID and after COVID	−6.525***
Switched during all periods	−5.360***
Member characteristic effects
Aged 55–70 years	−1.604***
Account balance	−0.523***
Male	1.005***
Choice account	0.897***
Pension account	1.063***
Difference in GA	−13.478***
Non-investment activity index	1.051***
Number of observations	57,744

Significance (two-tailed) is indicated by *p < 0.05, **p < 0.01, ***p < 0.001.

6.4. Further analysis of multiple switchers

As mentioned at the start of this section, one limitation of the impact of switches measure is that we are unable to gauge if members may be making rational switching decisions in response to changes in circumstances, objectives or preferences, or instead engaging in return-chasing behaviour. In this section, we make some adjustments to the sample to address this issue. First, we consider multiple switchers only. Second, we amend the impact of switches variable to calculate the dollar impact of switching up until the date of the final switch rather than 30 June 2021, hence removing the impact of the final switch made by the member. We are effectively treating the initial and final investment options in the data as representing the true preference of the member, and assuming that all switches between these two observed allocations represent return-chasing. We call these intervening switches ‘intermediate switching’. While this is a broad assumption, one factor strengthening the assumption is the observation that 84% of multiple switchers made both growth and defensive direction switches. This seems less likely to occur purely because of changing circumstances, objectives or preferences; and more likely to be due to trading activity due to return-chasing. A further advantage of this approach is that the impact of switches is no longer dependent on a fixed end period of 30 June 2021, but rather is measured for each member until the date of their final switch. Results are presented in Tables 12 –14, which are the equivalent of Tables 9 –11.

Table 12.

Member groups based on the time of switching – multiple switchers only.

Group	When switched	Members (Total)		Default		Choice Account		Pension Account
Group	When switched	Number	%	Number	%	Number	%	Number	%
1	Prior COVID only	2,323	4	735	2	1,344	6	244	9
2	COVID only	807	1	461	1	310	1	36	1
3	After COVID only	1,426	2	806	3	573	3	47	2
4	Prior COVID and COVID	1,674	3	585	2	979	4	110	4
5	Prior COVID and after COVID	3,514	6	1,146	4	2,013	9	355	13
6	COVID and after COVID	6,806	12	3,870	12	2,609	11	327	12
7	Switched during all periods	3,976	7	1,005	3	2,587	11	384	15
	Total	20,526	36	8,608	27	10,415	46	1,503	57

To aid comparisons, percentages are expressed as a percentage of all switchers from Table 9.

Table 13.

Impact of switches – Intermediate switching for multiple switchers.

Green shading indicates a positive impact. Yellow shading indicates a reduction smaller than −5%, and orange a reduction of more than −5%.

Table 14.

Regression results – impact of switches with as dependent variable, Intermediate switching for multiple switchers.

Variable	Coefficient estimate
Period effects
Prior COVID only	−2.882***
COVID only	−1.749***
After COVID only	−3.839***
Prior COVID and COVID	−2.821***
Prior COVID and after COVID	−4.131***
COVID and after COVID	−7.406***
Switched during all periods	−6.722***
Member characteristic effects
Aged 55–70 years	0.567***
Account balance	−0.619***
Male	0.855***
Choice account	0.750***
Pension account	0.928***
Difference in GA	−6.791***
Non-investment activity index	0.822***
Number of observations	20,526

Significance (two-tailed) is indicated by *p < 0.05, **p < 0.01, ***p < 0.001.

Table 12 shows that 36% of switchers made multiple switches over the sample period. Much higher proportions of choice account and pension account members were multiple switchers than default members.

Table 13 reveals that the average impact of switches was negative for intermediate switching across all groups, with MySuper default members again suffering the largest average reduction in balance. The overall numbers are similar to the equivalent Total multiple switchers row in Table 10, indicating that balance reductions for multiple switchers are caused mostly by intermediate switches and are hence independent of the period of analysis and its end timing. Exceptions can be found in Panel B for COVID only and prior COVID and COVID where the reductions in Table 10 were caused mostly by a final defensive switch made during COVID after the intermediate prior switch(es). However, this occurs within a relatively small proportion of the multiple switcher sample at 1% and 3%, respectively.

Table 14 as expected shows similar coefficient estimates to Table 11. Consistent with the discussion above, the effects of COVID only and prior COVID and COVID are substantially smaller than seen in Table 11 where the coefficient estimates are caused mostly by a final defensive switch during COVID. The other significant change is the aged 55–70 years coefficient estimate is now positive rather than negative, which is reflective of the most substantial balance reduction for this age group being caused by a final defensive switch after intermediate prior switch(es).

In summary, the results for multiple switchers reported in Tables 12 –14 are broadly consistent with those in Tables 9 –11. They provide further evidence that the average balance reductions seen in the full sample do not depend on the end timing of the sample period and that the results accord with return-chasing behaviour.

7. Conclusion and implications

Our analysis of switching by Aware Super members offers several insights. To start, they confirm some widely held perceptions. We find that switching activity is generally low, that is, most members are inactive; members nearing or at retirement have a propensity to make defensive switches; males are more likely to make growth switches and there is more switching by members who are more engaged. However, we also find that switching activity and switching direction are by far most strongly influenced by market conditions as captured by our observation periods, with member characteristics having only a moderating impact.

We provide evidence on the impact of switches on member welfare. We find that switches reduce the account balances of the average member, particularly those who switched during the COVID-related market downturn, and that the losses can be substantial. Furthermore, the loss of value for the average member occurs not just during this period but is also associated with switches made during other periods. This suggests that the average member makes consistently bad switching decisions, with evidence that many members who switch appear to be reacting to market movements (return-chasing) and switching at inopportune times. This manifests as a strong skew towards defensive switches during the COVID-related market downturn and a tendency towards growth switches during other periods when equity markets were rising. While the negative impact of the defensive switching behaviour during the COVID period is unsurprising, the finding that the largely growth switches made during generally rising markets also destroy value is particularly notable as evidence that the average member makes switches that destroy value.

Our results have important practical implications for members, trustees and regulators of defined contribution pension funds, highlighting the need for, and the potential value of, member communications and possibly interventions to encourage members to identify and retain the asset mix that is most appropriate for their needs and not to react to market movement. We recommend that this communication not be left only to periods of market volatility, where emotions around potential financial losses may be high, but instead occur throughout all market conditions. Given that member characteristics have a secondary effect on switching activity and impacts compared to period, we recommend that communications be directed towards all members rather than targeted at specific member types.²³

Finally, we acknowledge some limitations of our study. We only analyse members of one pension fund, which although of large size, has some differences to the general population. While the large size and broad member base of the fund analysed give us confidence that the results are likely to be generalizable, and where comparable our results are broadly consistent with the literature, some caution still needs to be exercised. It would be particularly interesting in future research to see if the results can be replicated for other funds with different member profiles. Furthermore, our analysis spans a limited and relatively unique time period, with the COVID-related sell-off being somewhat atypical in the terms of the speed of the market decline and subsequent recovery fuelled by an aggressive policy response. This acted to amplify the losses from switching in a manner that may not translate to other periods of market volatility. Future research could consider the results for other time periods. We also have incomplete information on members, which limits our analysis of the member characteristics that may be associated with switching activity. In particular, we do not observe member preferences and goals, risk tolerance, assets outside of superannuation, family situation and more. As discussed in the literature, these factors can influence switching activity. Relatedly, we cannot identify the underlying motivations for switching, which could stem from changed personal circumstances or justifiable adjustments to expectations or risk perceptions. While our analysis of members who switched multiple times, the majority of whom switched in both growth and defensive directions provide supporting evidence for return-chasing, other explanations remain possible. Despite these shortcomings, our study nevertheless contributes new insights to the existing body of knowledge on the behaviour of the members of pension funds. We encourage funds to engage further with membership to collect additional data to better understand influences on switching and other activity.

Key practical and research implications

Investment switching activity of members of Aware Super is generally low, that is, most members are inactive. For those who do switch, this reduces the account balances of the average member, particularly for those who switched during the COVID-related market downturn, but also at other times. This period impact is much larger than the impact of member characteristics on switching outcomes. While the negative impact of the defensive switching behaviour during the COVID period is unsurprising, the finding that the largely growth switches made during generally rising markets also destroy value is particularly notable.

Our results highlight the need for, and the potential value of, member communications and possibly interventions to encourage members to identify and retain the asset mix that is most appropriate for their needs and not to react to market movement.

We do not observe factors typically unavailable to funds, such as member preferences and goals, risk tolerance, assets outside of superannuation, family situation, and more. We encourage funds to engage further with membership to collect additional data to better understand influences on switching and other activity.

Footnotes

Acknowledgements

The authors thank Aware Super for supplying their member data in support of this research. The research and findings are solely attributable to the authors and should not be taken as representing the viewpoints of Aware Super.

Final transcript accepted on 1 January 2025 by Tom Smith (AE Finance).

Ethical considerations

No ethical approval is required.

Declaration of conflicting interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: Shang Wu is employed by Aware Super, who provided the data used in this study. All other authors declare that they have no conflict of interest.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Adam Butt

Geoffrey J. Warren

Notes

References

Agnew

Balduzzi

Sunden

(2003) Portfolio choice and trading in a large 401 (k) plan. American Economic Review 93: 193–215.

Ameriks

Zeldes

(2004) How Do Household Portfolio Shares Vary with Age. Working Paper, Columbia University, New York. Available at: https://www0.gsb.columbia.edu/mygsb/faculty/research/pubfiles/16/Ameriks_Zeldes_age_Sept_2004d.pdf

Australian Prudential Regulation Authority (2023a) Annual Superannuation Bulletin – Superannuation Entities. Available at: https://www.apra.gov.au/annual-superannuation-bulletin (accessed form 13 June 2023).

Australian Prudential Regulation Authority (2023b) Annual Fund-Level Superannuation Statistics. Available at: https://www.apra.gov.au/annual-fund-level-superannuation-statistics (accessed from 21 February 2022).

Barber

Odean

(2001) Boys will be boys: Gender, overconfidence, and common stock investment. Quarterly Journal of Economic 116: 261–292.

Bateman

Deetlefs

Dobrescu

, et al. (2014) Just interested or getting involved? An analysis of superannuation attitudes and actions. Economic Record 90: 160–178.

Bateman

Islam

Louviere

, et al. (2011) Retirement investor risk tolerance in tranquil and crisis periods: Experimental survey evidence. Journal of Behavioural Finance 12: 1–218.

Bebbington

Durand

Khuu

(2024) Investor decision making within retirement savings schemes. Pacific-Basin Finance Journal 83: 102203.

Butt

Donald

Foster

, et al. (2018) One size fits all? Tailoring retirement plan defaults. Journal of Economic Behaviour and Organization 145: 546–566.

10.

Byrne

Blake

Mannion

(2009) Financial Sophistication and Pension Plan Decisions. Discussion Paper PI-0903, Pensions Institute.

11.

Clark-Murphy

Gerrans

Speelman

(2009) Return chasing as a driver in individual retirement savings investment choices: Evidence from Australia. Journal of Family and Economic Issues 30: 4–19.

12.

Cronqvist

Thaler

(2004) Design choices in privatized social-security systems: Learning from the Swedish experience. American Economic Review 94: 424–428.

13.

Dahlquist

Martinez

(2015) Investor inattention: A hidden cost of choice in pension plans? European Financial Management 21: 1–19.

14.

Deetlefs

AMJ

Bateman

Dobrescu

, et al. (2019) Engagement with retirement savings: It is a matter of trust. Journal of Consumer Affairs 53: 917–945.

15.

Friesen

Sapp

(2007) Mutual fund flows and investor returns: An empirical examination of fund investor timing ability. Journal of Banking and Finance 31: 2796–2816.

16.

Frino

Heaney

Service

(2005) Do past performance and past cash flows explain current cash flows into retail superannuation funds in Australia? Australian Journal of Management 30: 229–244.

17.

Gan

Heaney

Gerrans

(2015) Individual investor portfolio performance in retirement savings accounts. Australian Journal of Management 40: 652–671.

18.

Gerrans

(2012) Retirement savings investment choices in response to the global financial crisis: Australian evidence. Australian Journal of Management 37: 415–439.

19.

Gerrans

Clark-Murphy

Speelman

(2010) Asset allocation and age effects in retirement savings choices. Accounting and Finance 50: 301–319.

20.

Gerrans

Faff

Hartnett

(2015) Individual financial risk tolerance and the global financial crisis. Accounting and Finance 55: 165–185.

21.

Gerrans

Gardner

Clark-Murphy

, et al. (2006) An investigation of home bias in superannuation investment choices. Economic Papers 25: 17–31.

22.

Himelfarb

(2019) A primer on standardized testing: History, measurement, classical test theory, item response theory, and equating. The Journal of Chiropractic Education 33: 151–163.

23.

Hoffmann

Post

(2015) How return and risk experiences shape investor beliefs and preferences. Accounting and Finance 57: 759–788.

24.

Hoffmann

Post

Pennings

(2013) Individual investor perceptions and behaviour during the financial crisis. Journal of Banking and Finance 37: 60–74.

25.

Ibbotson

Kaplan

(2000) Does asset allocation policy explain 40, 90, or 100 percent of performance? Financial Analysts Journal 56: 26–33.

26.

Lusardi

(2012) Numeracy, financial literacy, and financial decision-making (No. w17821). National Bureau of Economic Research 5:2.

27.

Peng

Alpert

Hsu

GCM

(2020) Switching between superannuation funds: Does performance and marketing matter? Pacific-Basin Finance Journal 63: 101431.

28.

Roszkowski

Davey

(2010) Risk perception and risk tolerance changes attributable to the 2008 economic crisis: A subtle but critical difference. Journal of Financial Service Professionals 64: 42–53.

29.

Samejima

(1969) Estimation of latent ability using a response pattern of graded scores. Psychometric Society 34: 100.

30.

Sialm

Starks

Zhang

(2015) Defined contribution pension plans: Sticky or discerning money? Journal of Finance 70: 805–838.

31.

Spearman

(1904) General intelligence objectively determined and measured. American Journal of Psychology 15: 201–293.

32.

Speelman

Clark-Murphy

Gerrans

(2007) Decision making clusters in retirement savings: Preliminary findings. Australian Journal of Labour Economics 10: 115–127.

33.

Thurstone

(1925) A method of scaling psychological and educational tests. Journal of Educational Psychology 16: 433.

34.

Thurstone

(1928) Attitudes can be measured. American Journal of Sociology 33: 529–554.

35.

Weber

Nosić

(2013) Who takes risks when and why: Determinants of changes in investor risk taking. Review of Finance 17: 847–883.

36.

Xue

Gepp

O’Neill

, et al. (2019) Financial literacy amongst elderly Australians. Accounting and Finance 59: 887–918.

Investment option switching behaviour and impact for pension fund members around the COVID pandemic

Abstract

Keywords

1. Introduction

2. Data

2.1. Member data

2.2. Investment options

2.3. Observation periods

2.4. Summary statistics for member sample

3. Summary statistics of switches

4. Empirical design of multivariate regression analysis

4.1. Model structure

4.2. Explanatory variables – member characteristics

4.3. Construction of activity indices

4.3.1 Non-investment activity index

4.3.2 Investment activity index

5. Regression results

5.1. Switching activity

5.2. Switching direction

6. Impact of switching on member balances

6.1. Sample

6.2. Summary statistics of impact of switches

6.3. Regression analysis of impact of switches

6.4. Further analysis of multiple switchers

7. Conclusion and implications

Key practical and research implications

Footnotes

Acknowledgements

Ethical considerations

Declaration of conflicting interests

Funding

ORCID iDs

Notes

References