![Sample our Behavioral Sciences journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/110801/TOC-Subject-Sample-2021-Behavioral-Sciences.jpg"})
Abstract
Empirical studies find that the order imbalance of retail trades can predict future stock returns. The authors investigated the cause of the puzzle using data from the laboratory asset markets in which inexperienced subjects trade in a single asset market (SSW design). The authors found that the retail order imbalance in period t positively predicted returns in period t + 1 in laboratory markets. The existence of return predictability in laboratory markets in which insider information or institutional investors are absent suggests that the predictability is not contingent upon private information or the activities of institutional investors, thus diminishing support of theories relying on these 2 conditions. In addition, the authors found that the return predictability in lab results was stronger and more statistically significant when the subjects were more excited. They tested this novel lab finding in empirical data and confirmed that return predictability is more robust when the market sentiment is higher. The findings suggest that the cause of the return predictability is likely linked to speculative activities.
Notes
1 All p values in the text are cited from the p values from regression models in tables.
2 To facilitate comparisons across experimental markets, one random dividend sequence (8, 60, 28, 8, 60, 8, 0, 28, 0, 60, 28, 60, 0, 8, 8) was drawn for the first market and then used for all subsequent markets.
3 A participant can only join one experiment. Participants were paid a $5 show-up fee and an additional performance-based fee averaging $21.68. Nine participants traded in each market; no participant traded in more than 1 market. Three traders received an initial endowment of $18.00 plus 1 share of the risky asset; 3 traders received $14.40 plus 2 shares; 3 traders received $10.80 plus 3 shares.
4 AOL (2016) ran 24 excitement treatment markets and 31 nonexcitement treatment markets (combined from calmness and fear videos). They found no evidence of bubble difference in nonexcitement markets. Because of the limited sample size, this study groups all markets of nonexcitement into one cohort.
5 Table 2 in AOL (2016) lists the survey on emotional evaluations on each of 6 video clips from the Amazon Mechanical Turk marketplace. The 2 excitement video clips were rated high in both dimensions, valence and intensity. The other 4 video clips were rated either high in only one dimension or neither.
6 The observed ratio in the AOL data is above 50%. To avoid potential bias introduced by the small size of the sample, we set the value to 0.5. Setting the value to the observed ratio makes no notable difference in our results.
7 The order imbalance in a high-volume trading period would better capture the true underlying sentiment in the market, while the order imbalance in a low-volume trading period is susceptible to idiosyncratic errors and thus is less accurate in reflecting the underlying sentiment.
8 Using closing prices instead of average prices yields similar results.
9 Treating return as missing for periods with no trade makes no notable difference in our results.
10 The higher SOIs in the excitement treatment are observed mostly in the early periods (periods 1–9) of the market sessions.
11 In each plot, the observations are first sorted to SOIi,t and then grouped into 20 approximately equal-sized bins. Each dot in a plot represents the mean of SOIi,t and the mean of Reti,t + 1.
12 If SOIi,t in a peak-price period predicts Reti,t + 1, we should observe a drop in order imbalance in the peak-price periods. The data suggest otherwise. SOIi,t increases more in peak-price periods than in non–peak-price periods (not statistically significant). In addition, we do not observe any statistically significant interaction between SOIi,t and a peak-price period dummy variable.