---
title: "Portfolio choice and financial advice"
authors:
  - name: "Alexis Direr"
    affiliation: "Laboratoire d'Économie d'Orléans (LEO), Université d'Orléans, and Paris School of Economics"
  - name: "Michael Visser"
    affiliation: "ERMES and CREST"
date: "2012-09"
doi: "10.3917/fina.342.0035"
keywords: [portfolio choice, individual investors, financial advisors, fractional response model, life insurance, unit-linked funds]
jel_codes: [C24, D12, G11]
language: en
type: research-article
---

# Portfolio choice and financial advice

**Authors**
- Alexis Direr — Laboratoire d'Économie d'Orléans (LEO), Université d'Orléans, and Paris School of Economics — *direr@ens.fr*
- Michael Visser — ERMES and CREST — *michael.visser@ensae.fr*

**DOI**: [10.3917/fina.342.0035](https://doi.org/10.3917/fina.342.0035)

**Keywords**: portfolio choice; individual investors; impact of financial advisors.
**JEL codes**: C24, D12, G11.
**Research framework**: Laboratoire d'Économie d'Orléans (LEO); Paris School of Economics; ERMES; CREST.

---

## Abstract

This paper analyzes portfolio allocation decisions of individual investors. Our dataset records how individuals allocate their money among risky funds and a money-market fund, and also the characteristics of both the investors and the financial advisors who sell the products. These data offer a unique opportunity to investigate how portfolio decisions are affected by financial advisors. Our empirical strategy consists in studying the relationship between the share of the total capital invested in risky funds and the characteristics of buyers and sellers. Since the dependent variable is bounded between zero and one, we estimate a fractional response model. We find that the share invested in risky funds is larger when the advisor is more educated. Furthermore, male advisors sell larger shares of risky funds than female advisors. We offer possible explanations for these findings.

---

## 1. Motivation and contributions

The paper documents how the supply side of retail finance — financial advisors — shapes the portfolio allocations of individual investors. As Inderst (2009) emphasizes, household finance has largely ignored advisors even though retail products are "not bought but sold." The paper contributes to a small but growing literature (Bergstresser, Chalmers & Tufano 2009; Hackethal, Haliassos & Jappelli 2009; Kramer & Lensink 2009; Gerhardt & Hackethal 2009; Bluethgen et al. 2008; Jansen, Fischer & Hackethal 2008) by exploiting an unusually rich set of advisor characteristics.

Two contributions stand out.

1. **Effect of advisor education on portfolio risk.** The paper shows that highly educated advisors sell larger shares of risky funds than lowly educated advisors. This mirrors the well-documented relationship between investor education/financial literacy and stock market participation (Rooij, Lusardi & Alessie 2011; Guiso, Haliassos & Jappelli 2003; Campbell 2006), and extends it to the supply side. The interpretation is that less-educated sellers feel less at ease with complex financial products and are therefore less inclined to recommend risky funds.

2. **Effect of advisor gender on portfolio risk.** Male advisors sell more risky portfolio allocations than female advisors. The interpretation rests on two facts: (i) women are consistently more risk-averse than men in financial decisions (Byrnes, Miller & Schafer 1999; Sunden & Surette 1998; Jianakoplos & Bernasek 1998; Barber & Odean 2001; Dwyer, Gilkenson & List 2002; Niessen & Ruenzi 2005); (ii) when advisors cannot fully apprehend a client's risk tolerance, their recommendations partly reflect their own risk preferences. The paper notes this mechanism is general (medical or legal advice may be subject to the same bias).

A subsidiary contribution is methodological: the use of the Papke & Wooldridge (1996) fractional response model on a dependent variable that is part discrete and part continuous, with mass points at 0, 0.1, 0.2, …, 0.9 and 1.

---

## 2. Data and institutional setting

**Source.** Administrative data from a large French financial company operating in insurance, personal protection, savings, retirement and financial planning. The data record contracts signed between January 2004 and December 2005 through the company's network of agencies in France.

**Volumes.**
- 24,375 life-insurance contracts after exclusion of agencies directed by more than one manager and contracts with missing covariates.
- 1,215 distinct agencies, each directed by one manager (so 1,215 distinct sellers).
- 98 French administrative districts; 95 with active agencies in the data.
- Average contracts per manager: 20.06 (range 1 to 175).
- Clients: average age 49.0 years (SD 15.0), 52% male.
- Managers: average age 44.7 years (SD 8.3), 93% male.

**Institutional context — French life insurance ("assurance-vie").** Despite the name, the product is a tax-advantaged investment vehicle. Subscribers split their capital between a near-riskless euro-based fund (*fonds euro*) — short-term debt with a minimum guaranteed return — and eight preselected unit-linked funds (*fonds en unités de compte*) mostly invested in equities. After eight years of holding, taxes on accrued gains are nearly zero; capital can be bequeathed with full inheritance-tax exemption up to a ceiling. About 13 million people held such products at the time of writing, accounting for more than one third of aggregate French financial wealth (Couleaud & Delamarre 2009).

**Distribution channel.** Agencies are run by *agents généraux* — independent managers mandated by the company, paid via commissions that scale with the type of product and premium size (front-loads on payments and level loads on outstanding capital). Managers are the ones who counsel and negotiate with clients.

**Unit of observation.** A contract signed at the enrollment date. The data record the total capital and the split between the money-market fund and the aggregate of the eight unit-linked funds (the eight individual unit-linked allocations are not separately observed). Subsequent investments are not observed; the paper focuses on the initial allocation, when client and manager are known to have met.

### 2.1 Returns of the underlying funds (2003–2009)

| Fund | Mean (%) | SD (%) | Skewness |
|:---|:---:|:---:|:---:|
| Euro-based fund | 4.33 | 0.47 | 0.58 |
| European equity fund (1) | 16.00 | 34.81 | -1.08 |
| European bond fund | 3.45 | 4.81 | 0.42 |
| European equity fund (2) | 8.09 | 23.04 | -1.79 |
| French equity fund | 8.83 | 22.87 | -1.89 |
| International equity fund (1) | 3.79 | 21.29 | -1.50 |
| International bond fund | 0.33 | 6.09 | 0.53 |
| International equity fund (2) | 4.21 | 13.59 | -1.79 |
| International equity fund (3) | 5.02 | 17.88 | -1.88 |

The euro-based fund's standard deviation is at least an order of magnitude lower than that of any unit-linked fund.

### 2.2 Capital invested per contract

| Variable | Mean (€) | SD (€) | Min (€) | Max (€) |
|:---|:---:|:---:|:---:|:---:|
| Total capital | 9,430 | 30,282 | 120 | 900,000 |
| Money-market fund | 7,842 | 27,321 | 0 | 900,000 |
| Unit-linked funds | 1,588 | 9,335 | 0 | 612,245 |

### 2.3 Distribution of the risky share

The share of total capital invested in unit-linked funds has mass points at 0, 1, and at the deciles 0.1, 0.2, …, 0.9. About 95% of observations take these eleven discrete values; the remaining 5% are continuously distributed. About 58% of clients invested zero in unit-linked funds; more than 5% invested 100%.

| Subsample | Mean | p50 | p75 | p90 | p95 |
|:---|:---:|:---:|:---:|:---:|:---:|
| Full sample | 0.194 | 0 | 0.30 | 0.60 | 1.00 |
| Lowly educated manager | 0.184 | 0 | 0.30 | 0.50 | 0.80 |
| Highly educated manager | 0.200 | 0 | 0.30 | 0.60 | 1.00 |
| Female client × Female manager | 0.151 | 0 | 0.21 | 0.50 | 0.60 |
| Female client × Male manager | 0.174 | 0 | 0.30 | 0.50 | 0.80 |
| Male client × Female manager | 0.174 | 0 | 0.30 | 0.50 | 0.70 |
| Male client × Male manager | 0.218 | 0 | 0.40 | 0.70 | 1.00 |

Equality of means across subgroups is rejected at the 1% level for all pairs except the "client female × manager male" vs. "client male × manager female" pair. The lowest risky share is observed when both client and manager are female; the highest when both are male.

### 2.4 Manager education and other characteristics

Education is observed in seven categories. In the regressions it is collapsed into a binary "highly educated" indicator (= 1 for the three highest categories, covering 67% of managers).

| Education category | Share |
|:---|:---:|
| Autodidact | 0.93% |
| Low-level vocational training | 5% |
| General certificate of secondary education | 5% |
| High-school diploma | 22% |
| Two years of higher education | 33% |
| Two to four years of higher education | 14% |
| Four years of higher education and more | 20% |

Other manager characteristics observed: job tenure (mean 8.48 years), prior experience in insurance (56%), sales (79%), administration (76%), family situation, and whether the manager has financially dependent children (75%).

---

## 3. Econometric strategy

The dependent variable $s_{ijt}$ is the share invested by client $i$ in unit-linked funds at the date of contract $t$ with manager $j$. Bounded in $[0,1]$ and discrete-continuous mixed, it cannot be modelled by OLS (predicted values may fall outside the unit interval) nor by a doubly-censored Tobit (which assumes a continuous interior).

The paper adopts the **fractional response model** of Papke & Wooldridge (1996):

$$E(s_{ijt} \mid x_{ijt}) = G(x_{ijt}\beta)$$

with $G(\cdot)$ the logistic CDF. The parameter $\beta$ is estimated by **quasi-maximum likelihood** maximizing a Bernoulli log-likelihood:

$$l_{ijt}(b) = s_{ijt} \log G(x_{ijt}b) + (1-s_{ijt}) \log[1 - G(x_{ijt}b)].$$

Because the Bernoulli log-likelihood belongs to the linear exponential family, consistency follows from Gouriéroux, Monfort & Trognon (1984) even when the true distribution of $s_{ijt}$ is not Bernoulli. Asymptotic standard errors are computed using a sandwich formula $\hat{A}^{-1}\hat{B}\hat{A}^{-1}$.

Goodness of fit is reported as $R^2 = 1 - \sum \hat{u}_{ijt}^2 / \sum s_{ijt}^2$.

| Specification | Controls added | $R^2$ |
|:---:|:---|:---:|
| (1) | Client age, gender interactions, manager education | 0.331 |
| (2) | (1) + other manager characteristics + proxy for client wealth | 0.335 |
| (3) | (2) + 23 month dummies | 0.358 |
| (4) | (3) + 95 district dummies | 0.393 |

The proxy for client wealth is the average total capital invested by **other** clients of the same manager (the *i*-th observation is dropped from the average to avoid mechanical correlation with the dependent variable).

**Identification logic.** Under a first-best benchmark with informed, optimizing investors, only client characteristics (age, wealth, risk tolerance) should matter. A statistically significant impact of seller characteristics is therefore evidence of frictions in the decision process.

---

## 4. Results

### 4.1 Baseline fractional logit (Table 6)

| Variable | (1) | (2) | (3) | (4) |
|:---|:---:|:---:|:---:|:---:|
| Constant | -1.076*** | -1.281*** | -1.837*** | -2.217*** |
| Age client (×100) | -1.465*** | -1.506*** | -1.467*** | -1.494*** |
| Client female × manager male | 0.171** | 0.180** | 0.152** | 0.146** |
| Client male × manager female | 0.152 | 0.145 | 0.141 | 0.136 |
| Client male × manager male | 0.424*** | 0.435*** | 0.416*** | 0.406*** |
| Highly educated manager | 0.096*** | 0.072*** | 0.067** | 0.056* |
| Other controls | No | Yes | Yes | Yes |
| Month dummies | No | No | Yes | Yes |
| District dummies | No | No | No | Yes |

*Reference category: client female × manager female. ***/**/* denote 1%/5%/10% significance.*

**Headline findings.**

- **Client age** has a strong negative effect: older clients hold less risky portfolios, in line with Papke (1998) and Agnew, Balduzzi & Sunden (2003).
- **Manager gender — female clients.** Female clients invest more in risky funds when contracting with a male advisor (coefficient 0.171) than with a female advisor.
- **Manager gender — male clients.** Male clients invest substantially more with male advisors (coefficient 0.424) than with female advisors (0.152, not significant). The Wald test rejects equality of these two coefficients at the 1% level.
- **Asymmetric gender effect.** The gender-of-manager effect is larger for male clients (0.424 - 0.152 = 0.272) than for female clients (0.171 - 0 = 0.171), but this difference is not statistically significant in the simplest specification (Wald statistic 1.119; 5% critical value 3.84).
- **Manager education.** Highly educated managers sell more risky funds. The coefficient remains positive across all specifications but its significance level weakens as fixed effects are added (1% in (1)–(2), 5% in (3), 10% in (4)).

> **Authors' critical reading.** The well-documented gender effect among investors (men hold more equities than women) only shows up here when the seller is male: male and female clients buy similar shares of risky funds when dealing with a female seller. This suggests the seller's gender mediates how much of the standard "male investors take more risk" pattern is actually expressed in observed portfolios.

### 4.2 Robustness — dropping small contracts

Re-estimation on subsamples that progressively drop contracts smaller than 1,000 €, 2,000 €, …, 30,000 € (the latter eliminating 91% of the sample). The interaction-variable coefficients are stable or slightly increasing; the first and third interactions remain significant for nearly all subsamples; the second becomes significant once contracts below 7,000 € are dropped. The education coefficient stays roughly constant in magnitude but loses significance once contracts below about 5,000 € are dropped. The main results survive sample-restriction concerns about managers possibly delegating small clients to assistants.

### 4.3 Robustness — endogenous matching

Two diagnostic exercises.

**(i) Auxiliary regressions of manager characteristics on client characteristics (Table 7).** Of the regressors tested:
- *Manager gender* — no client variable is significant.
- *Manager age* — only client age has a significant coefficient (older clients match with older managers).
- *Manager education* — only total capital invested has a significant positive coefficient (wealthier clients match with more educated managers).

Overall, observable cross-side correlations are weak.

**(ii) Manager fixed-effects estimation.** The conditional expectation is rewritten $E(s_{ijt}|x_{ijt}) = G(x_{ijt}\beta)\delta_j = G(x_{ijt}\beta + \mu_j)$ and approximately one thousand seller fixed effects are jointly estimated with $\beta$. Time-invariant manager variables (education, gender of manager interacted with client female) are absorbed and no longer identified. Observations from managers with a single contract or with $s_{ijt} \in \{0\}$ (resp. $\{1\}$) for all $i,t$ are dropped.

| Coefficient | Without fixed effects (same reduced sample) | With manager fixed effects |
|:---|:---:|:---:|
| Age client (×100) | -1.490 (0.079) | -1.664 (0.085) |
| Client male × manager female | 0.135 (0.092) | 0.136 (0.094) |
| Client male × manager male | 0.390 (0.074) | 0.266 (0.024) |

The two interaction coefficients remain significantly different from each other at the 1% level. Magnitudes are comparable to the pooled estimates. Endogenous matching does not appear to drive the gender result.

> **Authors' critical reading.** The fixed-effects design controls for manager-level unobservables but cannot identify client-specific or client/manager-specific effects (each client appears once in the data). Unobserved client variables (education, risk profile, wealth beyond the proxy) remain a possible source of bias.

---

## 5. Conclusion

Three main takeaways.

1. **Advisor characteristics matter for portfolio choice.** In a first-best benchmark, only investor characteristics should drive the equity share. The data reject this benchmark.

2. **Education channel.** More-educated advisors sell riskier portfolios. A plausible mechanism is that less-educated advisors are imperfectly informed about complex financial products (equities, fixed-income securities) and steer clients toward the riskless euro-based fund.

3. **Gender / risk-aversion channel.** Female advisors sell less risky portfolios than male advisors. Combined with the literature documenting greater female risk aversion (Byrnes, Miller & Schafer 1999; Barber & Odean 2001) and lower risk-taking by female fund managers (Dwyer, Gilkenson & List 2002; Niessen & Ruenzi 2005), the natural reading is that advisors' own risk preferences leak into the recommendations they give to clients whose risk tolerance they cannot perfectly observe.

A general implication: investors do not have firm preferences over the equity share, and let outside advice substitute for those missing preferences. This echoes peer-effect evidence (Duflo & Saez 2002; Hong, Kubik & Stein 2004; Lyons, Neelakantan & Scherpf 2008).

The content is positive, not normative: the paper does not assess whether advisors improve or degrade client welfare, returns, or risk-profile suitability.

### Limitations and research extensions

- **Limited client information.** Education, risk profile and wealth of clients are not observed; client wealth is proxied by the manager-level average of other clients' total invested capital. Unobserved client heterogeneity can bias the estimated advisor effects even after manager fixed effects.
- **No subsequent returns.** The data record only the opening allocation; portfolio reshuffling and realized returns are not observed.
- **No outside accounts.** Other investment or savings vehicles held by the same client are not in the data.
- **No advisor portfolio.** The advisor's own personal allocation, which would directly test the "advisors project their own risk preferences" mechanism, is not observed.
- **Suggested extensions.** Tie advisor characteristics to subsequent portfolio performance and to suitability with clients' risk profiles; investigate whether advice attenuates investor behavioral biases (overtrading, underinvestment).

---

## Acknowledgments

The authors thank an anonymous referee, Mohamed Baccouche, Georges Bresson, Xavier D'Haultfoeuille, Philippe Février, Donatien Hainaut, Edwin Leuven, and Laurent Linnemer for comments and suggestions, and Karim Arslan and Rim Ennajar-Sayadi for help in constructing the dataset.

---

## Main references

Agnew, J., Balduzzi, P., & Sunden, A. (2003). Portfolio choice and trading in a large 401(k) plan. *American Economic Review* 93(1), 193–215.

Barber, B. M., & Odean, T. (2001). Boys will be boys: Gender, overconfidence, and common stock investment. *Quarterly Journal of Economics* 116(1), 261–292.

Bergstresser, D., Chalmers, J., & Tufano, P. (2009). Assessing the costs and benefits of brokers in the mutual fund industry. *Review of Financial Studies* 22(10), 4129–4156.

Byrnes, J. P., Miller, D. C., & Schafer, W. D. (1999). Gender differences in risk taking: A meta-analysis. *Psychological Bulletin* 125(3), 367–383.

Campbell, J. Y. (2006). Household finance. *Journal of Finance* 61(4), 1553–1604.

Dwyer, P. D., Gilkenson, J. H., & List, J. A. (2002). Gender differences in revealed risk taking: Evidence from mutual fund investors. *Economics Letters* 76(2), 151–158.

Gerhardt, R., & Hackethal, A. (2009). The influence of financial advisors on household portfolios: A study on private investors switching to financial advice. Working Paper, Goethe University Frankfurt.

Gouriéroux, C., Monfort, A., & Trognon, A. (1984). Pseudo maximum likelihood methods: Theory. *Econometrica* 52(3), 681–700.

Guiso, L., Haliassos, M., & Jappelli, T. (2003). Stockholding in Europe: Where do we stand and where do we go? *Economic Policy* 18(36), 123–170.

Hackethal, A., Haliassos, M., & Jappelli, T. (2009). Financial advisors: A case of babysitters? Working Paper, Goethe University Frankfurt.

Inderst, R. (2009). Retail finance: Thoughts on reshaping regulation and consumer protection after the financial crisis. *European Business Organization Law Review* 10(3), 455–464.

Jansen, C., Fischer, R., & Hackethal, A. (2008). The influence of financial advice on the asset allocation of individual investors. Working Paper, Goethe University Frankfurt.

Kramer, M., & Lensink, R. (2009). The impact of financial advisors on individual investor portfolio performance. Working Paper, University of Groningen.

Niessen, A., & Ruenzi, S. (2005). Sex matters: Gender and mutual funds. Working Paper, University of Cologne.

Papke, L. E., & Wooldridge, J. M. (1996). Econometric methods for fractional response variables with an application to 401(k) plan participation rates. *Journal of Applied Econometrics* 11(6), 619–632.

Rooij, M. van, Lusardi, A., & Alessie, R. (2011). Financial literacy and stock market participation. *Journal of Financial Economics* (forthcoming at time of writing).

*The full reference list appears in the PDF.*