«Oscar Stolper (University of Giessen) Andreas Walter (University of Giessen) Discussion Paper Series 1: Economic Studies No 23/2011 Discussion Papers ...»
It is conceivable that local bias is a phenomenon which is essentially driven by the area in which the investor lives. Grinblatt and Keloharju (2001), for instance, provide empirical evidence that individual investors located in rural regions exhibit a particularly strong bias towards local companies. Thus, to confirm the robustness of our results with respect to investor location, we construct subsamples according to the population structure of the household’s local area (urbanized versus rural). Panel A of Table 3 reports the corresponding numbers and reveals that the average local bias for residents of urbanized areas (9.7%) is even higher than for households living in rural areas (5.8%). Consequently, we are confident that our findings regarding the local bias extend to households living in both urbanized and rural areas.
Moreover, as discussed above in section 2.2, the geographical distribution of individual investors shows considerable variation between East and West Germany. Thus, we replicate the analysis for the subsamples of households in the Western states and the New Länder and show that households in the New Länder exhibit much less of a local equity preference than those residing in the Western states (2.9% as compared to 8.8%). Yet, even for the investors from the Eastern part of Germany, the average deviation of local stock investments from the CAPM-efficient allocation is still highly significant. One way to rationalize the higher local bias levels in urban versus rural areas and West Germany as compared to East Germany These two studies employ the same local bias measure as we do. Qualitatively, our results are corroborated by a number of other studies using alternative metrics, including Huberman (2001), Grinblatt and Keloharju (2001), and Ivkovic and Weisbenner (2005), whose results also document significant local bias among individual investors.
might involve considering the heterogeneous geographical distribution of companies throughout Germany (see section 2.2). Company clusters in agglomeration areas provide local investors with significantly better diversification opportunities in their nearby environment than, say, a single company in a rural East German area does. While the benchmark portfolio obviously requires a much smaller investment in local stocks in rural, less industrialized parts of East Germany, it might not fully adjust for differences in diversification possibilities since it accounts for company location but not for local industrial variety.
Local bias and company awareness
In addition, we would like to test whether the observed local bias is a phenomenon which is primarily attributable to stockholdings in companies whose awareness is limited to the local investment community. Due to higher wide-area media coverage or a greater exposure of individuals to company advertisements, for instance, some firms are visible to many potential investors, regardless of where they reside. For a given investor, this mitigates the asymmetry in familiarity between local and remote companies with high relative visibility. Assuming that people tend to invest in the familiar (Huberman, 2001), this effect should reduce the fraction of locally invested equity and thus the local bias. Following Ivkovic and Weisbenner (2005), we choose membership in the major national stock index DAX to distinguish companies which are nationally known from those whose awareness is likely to be regionally bounded.
DAX members are assumed to feature a relatively small potential for asymmetries in company visibility, while the opposite holds true for non-DAX companies. Panel B of Table 3 splits the sampled equity universe into stocks of the 30 companies listed in the DAX, which together account for roughly 60% of Germany’s aggregate market capitalization, and the remainder of stocks. Indeed, the rightmost column of Panel B reports average local bias levels of nearly 14% for the portion of non-DAX stocks, while for holdings of DAX-listed companies, the deviation from the benchmark comes to only slightly above one third this percentage (5.3%).
This indicates that index membership partly harmonizes the differences in the awareness of a given company between local and nonlocal investors. Alternatively, reduced local overinvestment for the subsample of DAX-listed companies might also be a result of measurement constraints. Note that we use the sampled companies’ headquarters for our distance calculations (see footnote 20) and therefore do not capture branch-related local investments but instead count them as remote stockholdings in a world where the premises of a given company are confined to its legal seat. Since DAX-members are particularly likely to have multiple premises located throughout Germany, we are in turn particularly likely to underestimate local bias levels for this subsample of firms. Again, however, the null hypothesis of no local bias is comfortably rejected for either of the two subsamples.
Local bias and employee stock ownership
Finally, we address the question of whether our implications regarding the preference for local equity might be distorted by employee stock ownership. In Germany, large publicly traded corporations offer employee share purchase plans (henceforth ESPP) where employees can buy company stock at a considerable discount, if they accept a lock-up period of several years. Assuming that most employees live in close proximity to the company they work for, household stockholdings attributable to ESPP would appear as local investments in the data.
Our data set does not allow for a distinction of shareholders according to their affiliation to the company they are invested in. However, this does not pose a problem, since ESPP-related stockholdings are typically aggregated in a collective deposit held by the company for account of their employees.24 In other words, they do not appear in the SecuStat filings we examine in this study.
4 Testing the information hypothesis: Do German individual investors yield excess returns on their local stock investments?
4.1 General intuition In this section, we investigate whether the information hypothesis is able to explain local bias among individual investors, i.e. whether they possess value-relevant information about local stocks and earn abnormal returns from stock-picking. To answer this question, we analyze the long-run performance of investors’ local stockholdings and compare it to different benchmarks. Although prior literature generally documents a poor performance of individual investors as stock market participants,25 there are several reasons why those investors might possess advantageous information for local stocks. First, since households are more exposed to regional shocks if they tilt their portfolios towards local equity, uninformed investors should eschew such a local overweight. Hence, the holdings in geographically close equity should See Dorn and Huberman (2005, p. 469).
Barber (1999) and Barber and Odean (2000), for instance, document that the stockholdings of the average broker client in their sample yield negative abnormal returns. Dorn et al. (2008) report related evidence for German private investors.
reflect the investments of informed households, which is why we can expect that, on average, local stock investments should be accompanied by excess returns if information advantages apply. Second, by investigating nearby holdings of individual investors, we focus on the portfolio segment in which value-relevant informational asymmetries―if present―should be most pronounced. Third and finally, we replicate the performance analysis for the subsample of non-DAX-listed stocks in order to guarantee that our results are not weakened by the more nationally known companies with less potential for information asymmetries between local and nonlocal stockholders.
We divide each investor’s portfolio into a local and nonlocal portion, using the 100 kilometer threshold. Next, we calculate quarterly returns of the local and remote portion of her portfolio.26 These returns are then regressed on the performance of two reference portfolios.
Measuring long-run abnormal performance of individual investors
Note that several methodological issues should be considered when studying the performance of individual investors.27 One set of pitfalls concerns the calculation of a valid test statistic.
First, we have to account for cross-sectional dependence in portfolio returns across individuals. This is necessary because our data comprises 27,819 investor-quarter observations, while our equity universe consists of 1,317 different stocks. Thus, we have cross-correlation in returns whenever two investors hold the same stock over the same quarter. Hoechle et al. (2009) find that test statistics which ignore cross-sectional dependence in the sample of investors’ returns can produce t-values which are three and more times higher than their correctly specified counterparts and thus are unusable. Second, we require an appropriate benchmark against which to compare households’ returns on their local equity investments.
Moreover, empirical evidence suggests that individuals hold rather poorly diversified portfolios, typically composed of only a handful of different stocks.28 Put differently, chances are that the local fraction of a given investor’s holdings consists of a single stock only. Thus, the For details on how the returns are computed, the reader is referred to the appendix.
See, for instance, Lyon et al. (1999) and Hoechle et al. (2009) for problems with measuring long-run abnormal returns of individual investor’s stockholdings as well as methodological approaches to resolve them.
See, for instance, Dorn and Huberman (2005) for empirical evidence of under-diversification among German private investors.
monthly return of a sole stock may be counted as an observation in a standard regression analysis, which would mean that small, volatile stocks can overly influence results.
Finally, recent empirical evidence has revealed a potential time-series selection bias when investigating individual investors’ preference for nearby companies. As mentioned in section 1, Seasholes and Zhu (2010) re-estimate the results of Ivkovic and Weisbenner (2005) and reach directly contradicting conclusions. They partly ascribe this to the fact that Ivkovic and Weisbenner confine their analysis to a (arbitrarily chosen) cross-section of holdings data.
We make use of calendar-time portfolios in order to circumvent the problems discussed above.29 Owing to the structure of our dataset, we build 1,715 bank-level portfolios, each of which aggregates the stockholdings of all private households affiliated with the respective bank, and thus ensure that the impact of small numbers of stocks is not unduly high in the performance analysis. For each of the 1,715 portfolios, we then calculate the value-weighted return of its local holdings. We estimate pooled ordinary least squares regressions and compute Rogers (1993) standard errors that are robust to heteroscedasticity and contemporaneous correlation (clustered by quarter).
Also, by analyzing a time span of more than four years―with utterly different stock market periods of boom and bust, including an unprecedented financial crisis―we implicitly avoid arbitrary ‘snapshot’ results.