Biased Advise

Using data from pension funds in Italy, we document a gender gap in pension investment choices. Our analysis reveals that while women in their twenties contribute a similar proportion of their salary to pension funds as men, they tend to select less volatile investment options with lower average returns.

To explore the origins of this gap, we have obtained permission to conduct a survey experiment with pension fund advisors who provide guidance for the pension funds in our dataset. Our study aims to examine whether, and to what extent, these advisors give different advice to men and women.

To investigate this, we will conduct a name-implied vignette study in which pension fund advisors provide recommendations to hypothetical clients in their twenties. Half of the participants will advise a hypothetical female client, while the other half will advise a hypothetical male client. The clients will be identical apart from their name. First, we will ask advisors what they believe is the optimal monthly savings rate and investment allocation for their assigned client. Next, we will assess how strongly advisors perceive their clients to incorporate their recommendations. Specifically, we will ask advisors to estimate the choices they believe their clients would have made before receiving their advice and subsequently ask them about their beliefs of actual post-consultation choices of their clients. This allows us to estimate their belief impact on the choices.

Moreover, we will also collect pension advisors' beliefs about their clients’ financial literacy and risk aversion. These factors will allow us to control for and better understand potential gender differences in investment advice.

We will then introduce a second randomized treatment. Half of the advisors will be presented with results from a previous survey documenting biased investment advice—specifically, that women were, on average, advised to contribute the same percentage of their income as men but were recommended to chose less risky investment options than men. Advisors in this treatment group will also be informed about the consequences of this bias, particularly its contribution to the gender pension gap. The other half of the advisors will receive a placebo treatment, in which they are shown results from the same survey regarding clients' misconceptions about pension investments.

After receiving this information, both groups will advise an additional hypothetical client of the same gender as their initial hypothetical client. They will again provide recommendations on the monthly contribution rate and investment allocation. This design allows us to test whether a simple information intervention can mitigate potential gender biases in investment advice.

Finally, we will explore the geographic distribution of biased advice and its potential link to the observed gender gap in pension investment choices. Specifically, we will compute the extent of biased recommendations at a granular geographical level and test for spatial correlations between biased advice and gender disparities in administrative data.

The study will be conducted in the Trentino-Alto Adige/Südtirol region, a northern region of Italy. This region is particularly interesting because 33% of the population belongs to a German-speaking Austrian minority. Because of this, we will administer the survey in both Italian and German, allowing each advisor to choose their preferred language. In the analysis, we will use this choice to approximate the advisors' cultural background and examine whether gender differences in advice vary based on cultural background. In addition to cultural background, we will also explore whether advice differs by region or gender pairings.

2025-02-04

2025-03-07

The primary outcome variables will be the recommended share of income and the recommended investment line that financial advisors consider optimal for pension investment.

We have several secondary variables that we will investigate. Secondary outcome variables include advisors' beliefs about their clients' financial literacy (numeracy, diversification) and risk behavior. In addition, secondary outcomes also include advisors' beliefs about their clients' preferred share of income and the suggested investment line before their consultation, as well as advisors' beliefs about their clients' actual choices regarding the share of income and the suggested investment line after their consultation. This allows us to compute the believed effect of consultation on individuals' choices. Finally, we will consider the suggested investment lines and the suggested percentage of savings after treatment, i.e., after being confronted with information indicating the gender bias in the suggested investment lines and highlighting the possible consequences for the gender pension gap.

We carry out a vignette study involving financial advisors to gain insights into the guidance provided and investigate ways to reduce possible biases.

We will conduct a survey experiment with vignettes to investigate the origins of the gender gap in pension investment choices. Specifically, we seek to understand why women in their twenties chose to contribute a similar proportion of their salary to pension funds as men but tend to choose less volatile investment options with lower average returns. Our primary objective is to examine the role of financial advisors in shaping this gendered investment behavior.

We have obtained approval to conduct this study with pension fund advisors who provide guidance on retirement investments. Our study aims to assess whether, and to what extent, these advisors offer different advice to men and women. To do so, we will implement a name-implied vignette experiment, in which pension fund advisors provide recommendations to hypothetical identical clients in their twenties. Each advisor will be randomly assigned to advise either a male or a female client, with the only difference between them being their name. First, we will ask advisors to indicate the optimal monthly savings rate and investment allocation for their assigned client. At this stage, we will also collect advisors’ beliefs about their client’s financial literacy. Next, we will examine how strongly advisors believe their clients incorporate their recommendations. Specifically, we will ask advisors to estimate the investment decisions they think their clients would have made before receiving advice and then to predict their clients’ actual post-consultation choices. This will allow us to quantify advisors’ perceived impact on client decisions. At this point, we will collect advisors' beliefs about their clients’ risk aversion. These perceived factors—financial literacy and risk aversion—will help us better understand potential gender differences in investment advice.

Following this, we will introduce a second randomized treatment. Half of the advisors will be shown evidence from a previous survey documenting biased investment advice—specifically, that women were, on average, advised to contribute the same percentage of their income as men but were recommended less risky investment options. These advisors will also receive information on the potential consequences of this bias, particularly its role in exacerbating the gender pension gap. The remaining advisors will receive a placebo treatment, in which they will be shown findings from the same survey about clients' misconceptions regarding pension investments. Importantly, the placebo information is unrelated to the gender pension gap.

After receiving this information, all advisors will advise a second hypothetical client of the same gender as their initial client. They will again be asked to recommend a monthly contribution rate and an investment allocation. This design enables us to test whether a simple informational intervention can mitigate potential gender biases in investment advice.

Finally, we will examine the geographic distribution of biased advice and its relationship to observed gender disparities in pension investment choices. Specifically, we will measure the extent of biased recommendations at a granular geographic level and analyze spatial correlations between biased advice and gender gaps in administrative data.

The study will be conducted in Trentino-Alto Adige/Südtirol, a northern region of Italy. This region presents a unique research opportunity, as 33% of the population belongs to a German-speaking Austrian minority. To account for linguistic and cultural differences, we will administer the survey in both Italian and German, allowing advisors to choose their preferred language. In our analysis, we will use this choice as a proxy for cultural background and investigate whether gender differences in investment advice vary accordingly. Additionally, we will explore whether advice patterns differ by region or by gender pairings between advisors and clients.

Randomization done directly by the survey system.

Participants will be randomly assigned to different branches with equal probability. In this case, the participants are financial advisors, and there are four distinct branches in total. In the first stage, 50% of participants will be assigned a female client, while the other 50% will be assigned a male client. In the second stage, 50% of participants will receive a treatment, in which they will be provided with information about biases in investment advice—specifically, that female clients are typically advised to choose less risky and less profitable investment options, along with the potential consequences of such advice. The remaining 50% will receive information about common client misconceptions.

No

We contact the population of financial advisors working for Pensplan (383). We expect a minimum of 140 participants.

We contact the population of financial advisors working for Pensplan (383). We expect a minimum of 140 participants.

We expect at least 140 complete observations. 70 meeting a female client. 70 meeting a male client. In addition, 70 being treated, 70 being not treated. This gives us at least 35 female client treated, 35 female client non-treated, 35 male client treated, 35 male client non-treated.

For the recommended share of income in the first randomization, we compute our minimum detectable effect size (MDE) using a two-sided t-test with N=70, a power of 0.80, and a significance level of 0.05. We find that we can detect an effect size of at least 0.48, which in the conventional Cohen's classification falls in the range of medium effects. For the recommended investment line in the first randomization, we compute our minimum detectable effect size (MDE) using a chi-squared test with N=70, a power of 0.80, and a significance level of 0.05. We find that we can detect an effect size of at least 0.39, which in the conventional Cohen's classification falls in the range of medium effects. For the recommended share of income in the second randomization, we compute our minimum detectable effect size (MDE) using a two-sided t-test with N=35, a power of 0.80, and a significance level of 0.05. We find that we can detect an effect size of at least 0.68, which in the conventional Cohen's classification falls in the range of medium-large effects. For the recommended investment line in the second randomization, we compute our minimum detectable effect size (MDE) using a chi-squared test with N=35, a power of 0.80, and a significance level of 0.05. We find that we can detect an effect size of at least 0.56, which in the conventional Cohen's classification falls in the range of large effects.