Under-Diversification

Under-Diversification

Under-diversification is the empirical pattern that households who participate in equity markets hold portfolios that are far less diversified than the rational benchmark prescribes. The benchmark — Markowitz mean-variance optimization with the global market portfolio — predicts that every investor should hold a value-weighted slice of the world equity universe. Real households deviate from this benchmark in four distinct ways, each with its own evidence base:

1. Home bias. Households allocate a large share of their equity holdings to domestic stocks. In $1989$, U.S. households held $94\%$ of their equity in U.S. stocks, U.K. households held $82\%$ in U.K. stocks, and Japanese households held $98\%$ in Japanese stocks. The rational benchmark would call for shares roughly equal to each country’s share of global market capitalization (about $30\%$ for the U.S., much less for the U.K. and Japan).
2. Local bias. Within the domestic equity allocation, households over-weight stocks of companies headquartered nearby. Ivkovich & Weisbenner (2005) use Odean brokerage data to show that the typical investor holds $31\%$ of her stocks within $250$ miles of her home address, while only $12\%$ of all U.S. stocks are within that radius. Average distance to her holdings is $917$ miles vs. $1,225$ miles for the average stock.
3. Concentrated holdings. Households often hold a small number of individual stocks. Curcuru, Heaton, Lucas & Moore (2004) define a household as “undiversified” if more than $50\%$ of its equity sits in a brokerage account holding fewer than $10$ stocks; in $1989$ about $30\%$ of U.S. households fit this definition, falling to $14\%$ by $2001$.
4. Own-company stock. Households allocate large fractions of $401$(k) contributions to their employer’s stock. Benartzi (2001) studies $154$ S&P $500$ firms and finds that across all plans, employees invest $23\%$ of discretionary contributions in company stock — $18\%$ in cash-match plans and $29\%$ in stock-match plans. The over-allocation persists even though it doubles the household’s exposure to the employer (already concentrated through wages and human capital).

The four patterns share a common thread: households over-weight what is familiar — the domestic market, nearby companies, employer stock, individually-known names — at the expense of distant, unfamiliar diversification. The leading behavioral explanations are familiarity bias (Heath & Tversky, 1991: people prefer to bet on what they feel they understand) and information advantages, which are real but quantitatively far smaller than the over-weighting they would justify.

Intuition

The mathematics of diversification is unambiguous: holding many uncorrelated assets reduces portfolio variance without sacrificing expected return. A rational investor with mean-variance preferences should therefore push toward the global market portfolio and hold no idiosyncratic risk that can be diversified away. Real households do the opposite — they hold concentrated portfolios, over-weight nearby stocks, and pile into their own employer’s equity. The rational return story cannot explain this: there is no informational reason why a Japanese household should hold $98\%$ Japanese stocks rather than the $10\%$-or-so that Japan’s share of global market cap would imply.

The cognitive shortcut at work is familiarity. Investors feel they understand companies they have heard of, employers they work for, and countries they live in. This perceived understanding gets confused with informational advantage even when no real advantage exists. Familiarity bias is closely related to ambiguity aversion: investors shrink from holding assets whose distribution they cannot picture, and prefer assets where they at least feel they could form an opinion. The result is a portfolio that is both informationally redundant (the stocks in the portfolio are correlated with each other and with the household’s labor income) and informationally narrow (the stocks the household omits cover the rest of the diversification frontier).

Properties

(Home Bias) Households over-weight their own country’s equity by a factor of $3$ to $10$.

The U.S., U.K., and Japanese figures from $1989$ ($94\%$, $82\%$, $98\%$ in domestic equity) all exceed the global-market-cap-implied weights by a factor of $3$ to $10$. The bias has narrowed somewhat as international index funds have become accessible, but it remains substantial in every developed economy. The most straightforward rational explanation — hedging the currency or political risk of consumption — predicts only modest deviation from the global benchmark and cannot generate $90\%+$ home weights for any plausible parameter values. The leading behavioral explanation is familiarity: investors feel they understand the firms whose products they consume and whose news they read.

(Local Bias) Households over-weight stocks within a few hundred miles of home.

Ivkovich & Weisbenner (2005) find that the typical Odean-data household holds $31\%$ of its equity in stocks within $250$ miles, while only $12\%$ of the U.S. equity universe sits within that radius. The over-weighting is even larger for non-S&P $500$ stocks (where information asymmetries are plausibly larger). One striking pattern: investors who hold local stocks earn modestly higher returns on their local holdings than on their distant holdings, suggesting that some informational advantage is at work — but the effect size is far too small to justify the magnitude of the local-weighting bias. Familiarity does most of the work; information does a little.

(Concentrated Holdings) Many households hold fewer than $10$ stocks.

Curcuru, Heaton, Lucas & Moore (2004) find that in $1989$, about $30\%$ of U.S. households with brokerage accounts held more than half their equity in fewer than $10$ stocks. By $2001$, the share had fallen to $14\%$, mostly because mutual fund investing had become widespread. The remaining $14\%$ of “undiversified” households are not all behavioral mistakes — some are inheritances, founder shares, or stock-option vestings — but a large portion reflect deliberate concentration that costs the household substantial Sharpe-ratio efficiency.

(Own-Company Stock) $401$(k) participants hold $23\%$ of contributions in employer stock.

Benartzi (2001) studies the $154$ S&P $500$ firms with sufficient $401$(k) data and finds that employees invest $23\%$ of discretionary (non-matched) contributions in their employer’s stock. The figure is $18\%$ at firms whose match is paid in cash and $29\%$ at firms whose match is paid in employer stock — meaning employees not only fail to offset the imposed concentration but actively amplify it. The behavior is doubly mistaken: the household is already heavily exposed to the employer through wages, benefits, and (for many workers) defined-benefit pensions, so adding equity exposure pushes the household far from the diversification frontier. The Enron collapse in $2001$ produced one of the most dramatic case studies of this risk: thousands of employees lost both their jobs and the bulk of their retirement savings simultaneously.

(Familiarity vs. Information) The bias is too large to be informational.

A possible defense of under-diversification is that households have private information about the assets they over-weight. Empirically, this defense is weak. The return differential on locally over-weighted stocks is positive but small (around $2\%$ per year on the over-weighted slice), while the diversification cost of concentrating in those stocks is much larger (the Sharpe-ratio loss from holding $10$ correlated local stocks instead of a market index is substantial). For home bias and own-company stock, even this small informational advantage is hard to find. The conclusion is that the bias is mostly familiarity rather than information — investors prefer what they know, even when knowing produces no measurable return advantage.

(Trend Toward Diversification) Index funds and target-date defaults have softened the bias.

The Curcuru et al. drop from $30\%$ undiversified in $1989$ to $14\%$ in $2001$ reflects the rise of mutual-fund investing, which mechanically diversifies even when the underlying investor would not. The post-$2006$ Pension Protection Act made target-date funds the default option in $401$(k) plans, further mechanizing diversification. The remaining bias is concentrated in directly-held brokerage accounts and in employer-stock holdings, both of which remain stubbornly home-biased and concentrated. Institutional design has done what investor education could not.

Examples

An Enron-era employee

A long-tenured Enron engineer in $1999$ has $$800,000$ in $401$(k) wealth, $80\%$ of which is in Enron stock — partly because Enron matches in stock, and partly because the engineer voluntarily directed his discretionary contributions into Enron as well. His wages also depend on Enron continuing to operate. His diversification picture is therefore: human capital exposed to Enron, $401$(k) exposed to Enron, and a smaller amount of Enron stock in a personal brokerage account. By the end of $2001$, Enron has filed for bankruptcy, the engineer’s $401$(k) is worth essentially nothing, and his job is gone. The rational benchmark would have called for him to hold less Enron exposure than the average investor (because his human capital was already highly correlated with Enron’s fate), not more. The familiarity bias and own-company-stock pattern produced exactly the opposite outcome.

Computing the Sharpe-ratio cost of home bias

Consider a U.S. investor with all of her equity in the U.S. market, vs. a globally diversified benchmark. Assume the U.S. and rest-of-world equity markets each have expected return $7\%$, volatility $20\%$, and correlation $0.6$. The U.S.-only portfolio has Sharpe ratio $\frac{0.07-0.02}{0.20}=0.25$. A $50$-$50$ U.S./international portfolio has variance $0.5^2\cdot 0.04+0.5^2\cdot 0.04+2\cdot 0.5\cdot 0.5\cdot 0.6\cdot 0.04=0.032$, so volatility $0.179$, and Sharpe ratio $\frac{0.05}{0.179}=0.279$. The diversified portfolio dominates the U.S.-only portfolio in Sharpe-ratio terms by about $11\%$. Over a $30$-year working life, this translates into roughly a $40\%$ difference in terminal wealth — substantial, and entirely available to the household at zero cost via an international index fund.

See Also

• 8.10. Behavioral Household Finance — parent hub.
• 8.10.1. Stock Market Non-Participation — sibling: the prior-step bias of staying out of equity entirely.
• 8.10.3. Disposition Effect — sibling: bias in selling decisions, complementing the bias in portfolio composition.