¹ Credit Suisse First Boston Technology Group (Burnham and Earle (1999)) reports a 70% drop in the average commission charged by the top ten online trading firms from first quarter 1996 to fourth quarter 1997, though commissions have remained largely unchanged from fourth quarter 1997 through first quarter 1999.

² In each of these surveys, investors were asked two questions: “What overall rate of return do you expect to get on your portfolio in the NEXT twelve months,” and “Thinking about the stock market more generally, what overall rate of return do you think the stock market will provide investors during the coming twelve months?” Across the 13 surveys, the average investor expects a return of 15 percent on their own portfolio, while they expect the market to return 13 percent.

³ Odean (1998a) points out that overconfidence may result from investors overestimating the precision of their private signals or, alternatively, overestimating their abilities to correctly interpret public signals.

⁴ The other six reasons are: cost, speed and availability, convenience, easy access to reliable information, lack of trust in and unsatisfactory experiences with traditional brokers, and investor discomfort when communicating directly with traditional brokers.

⁵ Previous studies of the behavior of individual investors include Lewellen, Lease, and Schlarbaum's (1977); Schlarbaum, Lewellen, and Lease (1978a, 1978b); Odean (1998b, 1999); Barber and Odean (2000a, 2000b); Grinblatt and Keloharju (1999); and Shapira and Venezia (1998). These studies do not analyze online trading.

⁶ The month-end position statements for this period allow us to calculate returns for February 1991 through January 1997. Data on trades are from January 1991 through November 1996. See Barber and Odean (2000a) for a detailed description of these data.

⁷ In auxiliary analyses, we matched the online sample to households with the most similar gross return in the 12 months prior to the switch to online traded. Our main results are very similar to those reported later in the paper. Based on this analysis, the self-attribution bias alone cannot explain the increased trading and resulting poor performance of online investors. It may be that the successful investors most prone to self-attribution bias are the ones who, in anticipation of increased trading, go online. But we believe that other factors specific to the online environment, such as information-based overconfidence and the illusion of control, also contribute to trading increases and poor performance.

⁸ If trades valued at less than $1,000 are included in these calculations, average round-trip commissions drop from 5.1 percent to 3.4 percent when investors go online. Average round-trip spreads drop from 1.2 percent to 0.9 percent.

⁹ Since 1996, online commissions have continued to drop while investor trading has increased. To determine whether investors have benefited from these offsetting trends will require additional research.

¹⁰ The general tenor of our results are the same if we weight the returns of the online and size-matched samples by account size rather than equally.

¹¹ When calculating this benchmark, we begin the year on February 1. We do so because our first monthly position statements are from the month end of January 1991. If the stocks held by a household at the beginning of the year are missing CRSP returns data during the year, we assume that stock is invested in the remainder of the household's portfolio.

¹² We estimate the logistic regression by pooling over time so that we can more precisely measure turnover and return performance in the period preceding a switch to online trading. When we estimate the regression using households, rather than household months, as the unit of observation, the results for the remaining independent variables are qualitatively similar.

¹³ Using the full sample six-year sample period to estimate household investment preferences assumes that preferences of online households do not change significantly after going online. In section 4.6, we document that there is not a significant change in investment style once households begin trading online.

¹⁴ Results for which p < 0.10 are generally considered moderately statistically significant, p < 0.05 is generally considered statistically significant and p < 0.01 highly significant.

¹⁵ Before going online, the gross performance of online traders is positive for all of our return measures except the own-benchmark measure which is essentially zero (-0.014, t = -0.55). Thus the superior returns earned before going online were due primarily to the portfolios these investors held at the beginning of our evaluation period, not to the trades they made during this period.

¹⁶ There is also an increase in the turnover of the median online household after the switch to online trading, though it is much smaller than the increase in average turnover (1.4 percent, p < 0.10). Twenty-five percent of the online households increase their turnover by 35 percent or more; Ten percent of the online households increase their turnover by 109 percent or more. Thus, most households increase their trading activity, but some increase their trading dramatically.

¹⁷ The daily returns on this portfolio are calculated as:

where

is the aggregate value of all speculative purchases in security i from day τ-63 through τ-1 and

is the gross daily return of stock i on day τ. We compound the daily returns within a month, which yields a time-series of monthly returns for four portfolios: one for speculative purchases before going online

, one for speculative purchases after going online

one for speculative sales before going online

, and one for speculative sales after going online

. The t-statistic to test whether the stocks purchased speculatively outperform those sold is calculated as:

.

¹⁸ If we look at all trades, not just speculative trades, the stocks online investors buy before going online outperform those they sell by an insignificant 29 basis points a month (p = 0.15). After going online their buys underperform their sells by a significant 31 basis points a month (p=0.05).

¹⁹ An analogous analysis for the size-matched households yields differences in net performance (after online less before online) ranging from 8 basis points (own-benchmark abnormal return) to -14 basis points (market-adjusted returns). The change in net performance of the online sample (after online less before online) less the change in performance for the size-matched households range from -21 basis points (Fama-French alpha) to -26 basis points (own-benchmark abnormal return); all of the differences between the online and size-matched samples are statistically significant at less than the 5 percent level (two-tailed).

²⁰ These conclusions are based on the coefficient loadings and associated test statistics from the time-series regressions that employ the Fama-French three-factor model. The online investors also have a preference for stocks with poor recent return performance relative to their size-matched counterparts. This inference is drawn by adding a zero-investment portfolio that is long stocks that have performed well recently and short stocks that have performed poorly (see, Carhart (1997)). None of our conclusions regarding performance are altered by the inclusion of this price momentum variable. We thank Mark Carhart for providing us with these return data.

²¹ An additional bias that may encourage some investors to trade online is loss aversion. Kahneman and Tversky (1979) argue that, when faced with losses, people will accept additional risks in hopes of recovering to the former status quo. While we do not believe that most investors are motivated to go online by loss aversion, some may be. An E*TRADE advertisement reads: The Tooth Fairy, Santa Claus, Social Security. The implied failure of social security would be a loss of anticipated welfare for many people. The prospect of such a loss could prompt people to take risks they might not otherwise take. Some might even go so far as to open an E*TRADE account.

²² The prevalence of bias inducing brokerage television commercials increased after 1995 (Barber, Elsbach, and Odean, 2002).

²² Speech at National Press Club, May 4, 1999, http://www.sec.gov/news/speeches/spch274.htm.

²⁴ Kraus and Stoll (1972), Holthausen, Leftwich, and Mayers (1987), Laplante and Muscarella (1997), and Beebower and Priest (1980) use closing prices either before or following a transaction to estimate effective spreads and market impact. See Keim and Madhavan (1998) for a review of different approaches to calculating transactions costs.

²⁵ Had we estimated spreads by dividing transaction prices by closing prices, net returns would be calculated as:

.

²⁶ The return on T-bills is from Stocks, Bonds, Bills, and Inflation, 1997 Yearbook (Ibbotson Associates, Chicago, IL 1997).Google Scholar

²⁷ The construction of these portfolios is discussed in detail in Fama and French (1993). We thank Kenneth French for providing us with these data.

²⁸ Lyon, Barber, and Tsai (1999) document that intercept tests using the three-factor model are well specified in random samples and samples of large or small firms. Thus, the Fama-French intercept tests employed here account well for the small stock tilt of individual investors.