NEW UPDATE: Completed trials may now upload and register supplementary documents (e.g. null results reports, populated pre-analysis plans, or post-trial results reports) in the Post Trial section under Reports, Papers, & Other Materials.
The Choice of Pension Funds
- an Information Experiment
Initial registration date
July 11, 2018
May 13, 2020 1:45 PM EDT
Other Primary Investigator(s)
Additional Trial Information
In this research project we empirically investigate why people choose and hold
strictly dominated pension funds. We discriminate between three main hypotheses:
information asymmetry, search costs, and financial illiteracy. An experiment is
conducted where letters are sent to pension savers in two strictly dominated funds
in the Swedish premium pension system. The treatment letters contain different
types of information in combination with incentivized tasks. Approximately 30 000
people save in the two funds for their premium pension, and we treat about 23 000.
Following the letter treatments we observe the real fund choices by the pension
savers, and test the relative importance of the three hypotheses. By randomizing the
different letters to individuals, we can identify potential causes why people choose to
stay with dominated fund choices. Administrative data on background characteristics
is used to study heterogeneous effects. Furthermore, we estimate distributions of
search costs and implied discount factors. By estimating discount factors across age
we test age dependent attention of the pension investment decision. Registration Citation
Intervention Start Date
Intervention End Date
Primary Outcomes (end points)
The outcome variables of interest include:
- an indicator variable for switching from the dominated to the dominating fund; - an indicator variable for switching from the dominated to another fund;
- the portfolio share invested in the dominated fund;
- an indicator variable for completing a search task (we study both those who successfully complete the search task, and those who complete the search task but provide an answer that is incorrect);
- an indicator variable for confirming reading a letter.
If possible, we will also look at the number of logins at the Pensions Agency’s website (at an individual level), and the average fund fee in the chosen portfolio. When we observe changes in the outcome variables we also analyze the
exact date at which changes were made, for the relevant individuals.
Primary Outcomes (explanation)
We consider several methods for analyzing our binary outcome variables, e.g. we may use
the cumulative normal distribution to transform expected binary outcomes into probabilities
(Probit function), and we may study shares of different groups.
Secondary Outcomes (end points)
We estimate distributions of search costs and implied discount factors.
Secondary Outcomes (explanation)
A more detailed description of how the distributions of search costs and implied discount factors are estimated is found in the Analysis Plan.
We conduct an experiment in the Swedish premium pension system. The treatments are
information letters in combination with incentivized tasks to savers in two dominated funds. We use five different treatment letters: Basic1, Basic2, A, C, and Ba. A summary of the treatment letters follows. In addition, there is a control group that receives no letter.
• Basic1: A reminder of the individual’s current choice of fund (the dominated fund)
and a statement that it is not the cheapest index fund with that investment strategy.
Information about the fee difference of the current and the cheapest index fund following
the selected investment strategy. In addition, a short guide describing how a fund
selection is implemented and how one can categorize the current fund.
• Basic2: The Basic1 letter and a statement offering one attempt where the saver
can verify that an identified fund in fact is the dominating index fund of the specific
• A: The Basic1 letter and information about the name of the cheapest index fund that
follows the same investment strategy as the currently chosen index fund.
• C: Letter A and a statement that clarifies the expected gain in the premium pension
account balance at age 65, from immediately switching to the cheapest index fund in
the chosen index fund category.
• Ba: The Basic2 letter, and a statement that clarifies the expected gain in the premium
pension account balance at age 65, from immediately switching to the cheapest index
fund in the chosen index fund category. Furthermore, we offer an immediate
search reward (four different amounts, a: 0, L, M, H) if the recipient of the letter reports the name of the
dominating fund to us. Note that we do not inform the saver of the name of the cheapest
index fund in the category.
Experimental Design Details
The randomization is done in an office by a computer, specifically, using the program Stata.
The unit of randomization is individual. The individuals in the sample (premium pension savers in the two dominated funds) are divided into strata based on: year of birth (2 groups, split by the midpoint of the age range), labor income (2 quantiles), fund choice (medium fee fund, high fee fund), and fund share (2 quantiles). Treatment groups (Control, Basic1, Basic2, A, C, B0, BL, BM, BH) are randomly assigned within strata.
Was the treatment clustered?
Sample size: planned number of clusters
We construct 16 strata, based on year of birth (2 groups, split by the midpoint of the age range), labor income (2 quantiles), fund choice (medium fee fund, high fee fund), and fund share (2 quantiles).
Sample size: planned number of observations
About 23 000 individuals are treated and roughly 7 000 individuals are left as controls.
Sample size (or number of clusters) by treatment arms
For the treatment arms we have the following number of individuals:
- Basic1: 2 500, where 250 receive reading test
- Basic2: 2 500
- A: 5 000, where 250 receive reading test
- C: 5 000
- B0: 5 000
- BL,BM,BH: 2 520 (840 for each level of reward)
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
INSTITUTIONAL REVIEW BOARDS (IRBs)
Regionala etikprövningsnämnden i Stockholm
IRB Approval Date
IRB Approval Number
Post Trial Information
Is the intervention completed?
Intervention Completion Date
August 10, 2018, 12:00 AM +00:00
Is data collection complete?
Data Collection Completion Date
February 10, 2019, 12:00 AM +00:00
Final Sample Size: Number of Clusters (Unit of Randomization)
17 960 individuals in treatment groups, 4 791 in control
Was attrition correlated with treatment status?
Final Sample Size: Total Number of Observations
17 937 individuals in treatment groups, 4 791 in the control group
Final Sample Size (or Number of Clusters) by Treatment Arms
For the treatment arms we have the following number of individuals.
- Basic1: 1 988, where 199 were sent the reading task
- Basic2: 1 992,
- A: 3 986, where 199 were sent the reading task
- C: 3 991
- B0: 3 974
- BL,BM,BH: 2 006 (669, 669, 668 for each reward level, respectively)
Reports, Papers & Other Materials
In this paper we empirically investigate potential causes of imperfect competition in the fund market, as characterized by high price dispersion among comparable funds. We discriminate between three main hypotheses on the demand side: a lack of awareness of price dispersion, search costs, and financial illiteracy. A large-scale field experiment is conducted in the Swedish Premium Pension system. Information letters are sent to pension savers in two index funds, where there exists a cheaper fund with the same index strategy. We show that an information intervention that increases the awareness of a cheaper, dominating fund, and reduces the search costs for finding such an alternative, can significantly improve households’ real investment allocations. Nonetheless, a majority of savers who are sent information about the name of the dominating fund do not switch funds. Thus, the high degree of inertia in pension investments remains even when search frictions for identifying dominating alternatives are eliminated.
REPORTS & OTHER MATERIALS