«Oscar Stolper (University of Giessen) Andreas Walter (University of Giessen) Discussion Paper Series 1: Economic Studies No 23/2011 Discussion Papers ...»
Ultimately, we compare against an additional benchmark which is specific to the location of a given investor. This way, we address heterogeneity in the geographic distribution of industries and households, and avoid that some local investors earn superior (inferior) returns on their local stockholdings simply because they reside in an area, where certain industries experience higher (lower) relative returns during the period under review. Consider, for instance, the period from mid-2007 to end-2008. Those stocks in our sample, which are related to the financial sector likely exhibit significant underperformance during that time. Recalling the strong geographic concentration of finance-related companies in the city of Frankfurt, chances are that we would erroneously document a significant underperformance of households in the area surrounding Frankfurt when applying the standard benchmark. To prevent The calendar-time portfolio approach dates back to the work of Jaffe (1974) and Mandelker (1974) and has proved suitable for the analysis of risk-adjusted performance of investors. See Hoechle et al. (2009) for a review of empirical finance studies applying this methodology. Our regression model largely follows Seasholes and Zhu (2010).
flawed results, we therefore regress the returns of local holdings not only on a broad market return, but also on a local benchmark. This investor-specific reference portfolio is composed of the value-weighted market capitalization local to a given investor, i.e. stemming from companies domiciled within 100 kilometers of her place of residence.
We form calendar-time portfolios based on both the holdings and the transactions of the individual investors under review, and present our results in the following section.
4.3 Results Holdings-based calendar-time portfolios Table 4 reports the results of the holdings-based regression analysis. Regression 1 documents the average raw excess return ( Rlocal,i − Rf ) which households earn on their local stockhold
both benchmarks as value-weighted indices including all stocks in our sample. Note that the market benchmark is the same for all investors, whereas the local benchmark is specific to the geographic location of a given investor. Regression 4 represents the full specification including both the nationwide and the local benchmark. Panel A shows the results for the full stock universe; Panel B reports the corresponding numbers for the subsample of non-DAX stocks.
We find a number of interesting results. The average quarterly excess return amounts to a negative 1.3 basis points (bp) per quarter, which can be ascribed to the down market in the second half of our sample period. Regressions 2 to 4 show that abnormal returns further converge to zero after adjusting for the different market betas; however, they are negative irrespective of the reference portfolio. Note that neither regression model produces significant alphas―be it economically or statistically―, regardless of the specification we estimate. In fact, even on an annualized basis, the gross loss before benchmark adjustment amounts to only roughly five basis points. Returns decrease when we use the investor-specific local benchmark as a reference portfolio (regression 3), but in terms of economic significance, results do not materially differ from those for the overall market index (regression 2).
The findings of our holdings-based analysis for the full stock universe qualitatively support the evidence provided by Seasholes and Zhu (2010), who also document economically and statistically insignificant alphas for U.S. individual investors (albeit positive ones). Unlike Seasholes and Zhu (2010), however, we also investigate holdings in the subsample of nonDAX firms (see Panel B of Table 4) and find that investors’ portfolio share of those stocks (with presumably lower visibility for the investment community) does not generate significant positive alpha, either. Quite on the contrary, excess returns even decrease across the board when we replicate the performance analysis for the sample of those companies for which we hypothesize that information asymmetries―if present―are highest. This is contrary to what one would expect to see in case of an information-based preference for nearby equity.
Next, we test the robustness of our results by dissecting households according to how strongly their stockholdings are tilted towards nearby companies. Assuming that investors with superior ability to pick local stocks concentrate their investments locally, whereas investors with no such abilities hold a better diversified portfolio, it could be the case that we find abnormal returns from nearby investments only for those households, which exhibit a high relative local bias. To investigate this issue, we rank all households according to their investment locality and assign them to local bias quartiles. Table 5 summarizes the results for the four resulting portfolios. Interestingly, average local bias levels differ sharply across the quartiles. At -0.9% on average, the 25% least locally invested households in the sample effectively show a slight remote bias, while mean local bias levels exceed 20% for households in the top quartile. Yet, differences across the four portfolios virtually vanish when focusing on investment performance: we re-estimate the full regression model (as described above) for each of the quartiles and find that alphas are all indistinguishable from zero and do not differ significantly, as can be seen in the rightmost column of Table 5. This implies that our main finding, i.e. private households do not significantly outperform the market with their local holdings, applies to all households in the sample, regardless of how strongly locally biased they are.
Transactions-based calendar-time portfolios
In a second step, we aim to explore if buys and sells of local stocks predict positive and negative future returns, respectively. Overall, purchases of individuals have been found to underperform their sales.30 Hence, it might be interesting to examine whether this still holds when focusing on the portfolio fraction of geographically close stocks. To this end, we now focus on changes in stock positions compared to the previous quarter. Note that these changes reflect net quarter-to-quarter turnover of aggregated portfolios which combine the trading decisions of many individual investors at the bank level. Certainly, this entails that opposite trades See, for instance, Odean (1999), who infers this finding from analyzing the accounts of discount brokerage customers.
,and―analogously―(ii) the difference in returns of the buy- and sell-positions for the portion buy sell of nonlocal stocks she holds ( Rremote,i - Rremote,i ). Returns are computed before transaction costs.
We assume that each stock is held for 12 months which has been reported as the average holding period in related research.31 Table 6 reports regression results regarding the performance of households’ purchases and sales of nearby stocks, at a 12-month horizon. Similar to what we see for the holdings-based analysis, we find negative alphas: local buys underperform local sells. This time, however, the return differential turns out to be statistically significant for both the full sample and the subsample of non-DAX companies. Interestingly, this finding again corroborates the empirical evidence of Seasholes and Zhu (2010) for U.S. individual investors. However, they report economically significant losses, whereas our analysis yields excess losses adding up to no more than 5.3 bp p.a. for the full sample and 4.8 bp p.a. for the subsample of non-DAX companies. With respect to the performance of remote stocks bought minus sold, we also document a marginally negative return for the entire stock universe (-0.70 bp p.a.) as well as for the subsample of non-DAX companies (-7.85 bp p.a.), which turns out statistically significant for the latter group. Compared to the returns from the local portfolio fraction, this points to a slightly better (worse) performance of the remote portfolio for the full sample (the subgroup of non-DAX companies). However, at roughly ±3 bp p.a., this effect is marginal in magnitude and thus economically negligible.
In sum, the results of the holdings-based as well as the transactions-based performance analysis conclusively reject the proposition that private households possess a ‘home-field advantage’ which manifests itself in value-relevant information about local companies. This has a number of implications. First, the findings document that it is by no means rational for private households to actively pick local stocks. In fact, returns do not compensate investors for the concentration of diversifiable risk they hold in their portfolios when tilting them towards local equity. Second, judging from the consistently negative return differential between local buys Seasholes and Zhu (2010) document an average holding period of one year; Doskeland and Hvide (2011) state an average holding period of 300 days.
and local sells it appears that, if anything, translating her local information into an investment decision turns out to be detrimental to an individual investor’s assets.
5 Testing the familiarity hypothesis: Investor ambiguity aversion and local bias
5.1 General intuition So far, we have not explicitly addressed changes in local bias levels over time. In order to examine the purely information-driven behavior by means of a performance analysis, it suffices to mitigate a potential time series selection bias, which we have been careful to do by considering all holdings and transactions over the entire 2005 to 2009 sample period (see section 4.2). Also, this rational behavior does not provide for changes in local bias over time, since it is unrealistic to assume that, on aggregate, investors systematically possess more information advantages at a certain point in time than before or after this date. In this section, we investigate whether, empirically, we observe changes in individual investors’ local bias over time and seek to rationalize them.
Our period under review is substantially different from others in that it includes extreme market cycles. Continued GDP growth in Germany between the last quarter of 2005 and the first quarter of 2008 is followed by four consecutive quarters of severe economic decline, with annualized GDP plummeting by 8 percent in the last quarter of 2008 and again 14 percent in the subsequent three months. Finally, this crisis period is replaced by moderate GDP growth from mid-2009 onwards. These heavy fluctuations are accompanied by unprecedented stock market volatility: The broad German stock index CDAX crashes by 43% in 2008 while in the other three years, it rises by more than 20% per annum.
We expect households to take measures in response to this strong economic downturn, i.e.
to rebalance their equity portfolios, where otherwise inertia would have prevailed.32 We are interested in whether different business cycles have an impact on individuals’ propensity to overinvest in nearby companies and how this teaches us new insights regarding the root cause of local bias. Specifically, our data set allows us to test a key implication of a portfolio selection model developed by Boyle et al. (2011). The authors extend the classic Markowitz model In a recent contribution, Cao et al. (2011) show analytically that investors are reluctant to trade away from investments that they currently hold (‘status-quo bias’). In their model, a threshold has to be exceeded for a given investor to be willing to leave this status quo.