CHAPTER IV. EMPIRICAL FINDINGS
IV.I D ATA
IV.2 O NE - WEEK FILTER
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Chapter IV. Empirical Findings
This chapter presents the major empirical findings for the proposed filters, and reports as well as discusses the results at the end of this chapter.
IV.I Data
I select top 50 largest capitalization stocks of listed companies every year from 1993/1/5 to 2012/12/28, a total of 5225 trading days in Taiwan Stock Exchange Market. If one stock suspends trading, the data that are 10 days before and after the suspension will not be included. The list of top 50 largest capitalization stocks of listed companies is updated on the first trading day of every year. By using top 50 largest capitalization stocks, the return caused by bid-ask bounce can be avoided.
IV.2 One-week filter
Return filter
At the beginning, we begin with examining the results for filters with only lagged return. Table I presents results for the case where profits are
‧
Table, just a few instances of returns are statistically significant. Cooper also uses lagged return filter, and his results show reversals in the loser and winner portfolios. In his paper, although not all the returns of winner filters are statistically significant, those of loser filters are. That is very different from our results.Table II presents the results for the case where profits are calculated two weeks after the strategy-formed-day. It shows that with not too much price decrease in the previous week, the returns turn negative and significant.
Same results can be seen in Table III, presenting the results for the case where profits are calculated three weeks after the strategy-formed-day. Also, Table III shows that with moderate increase in price in the previous week, the returns turn positive and significant three weeks after the strategy-formed-day.
It is puzzling why this takes place just three weeks after the strategy-formed-day.
Results for the cases where profits are calculated four weeks after the strategy-formed-day are collected in Table IV. Almost all of the past instances of significant returns are not significant anymore.
The results may show some evidence for predictability, but it is also not consistently that some instances of results are only significant at sometimes and do not consistent with the previous or following week.
Return and volume filter
Under the lagged return filter space in the table, there are four quadrants
‧
of filters combined with the lagged return and the lagged volume change. The first quadrant is “winner, low-volume” strategy on the upper right; the second is “loser, low-volume” on the upper left; the third is “loser, high-volume” on the lower left; the fourth is “winner, high-volume” on the lower right.
My expectation is when there is high trading volume, the return tends to continue; and when there is low trading volume, the return tends to be reversed. Nevertheless, the results on these four quadrants from Table I to IV are not so significant. Throughout these four tables, the maximum number of significant filters is 30 out of 120, which is very dissimilar to Cooper’s results.
In Table I, there is not any pattern of future returns. Only when price extremely decreases (< -10%) in the previous week, and trading volume decreases or increases no more than 60%, the return tends to be reversed, i.e.
be positive, one week after the strategy-formed-day. That being said, we can still find pattern of future returns, yet with low confidence: when price extremely increases (> 10%) in one week and trading volume changes between -30% to 90%, the return tends to be reversed after one week.
Table II shows that more negatively significant returns can be found in the area of filters with negative lagged return, although the results in Table I shows little pattern. The most obvious pattern can be seen is that with extreme decreases in lagged trading volume, the returns tend to be negative two weeks after the strategy-formed-day.
This pattern persists in Table III and Table IV. In Table III, except the pattern discussed above in Table II, with decreases in lagged price and lagged trading volume at the same time, the returns tend to continue; with increases
‧
that in Table II but is less obvious.To sum up, just some of the results for filters based on the lagged return and the lagged volume change in the previous week are significant with at most 25% of all filters, and cannot be compared to Cooper’s result with more than about 75% of all filters. Moreover, my expectation fails, and Wang’s theory (1994) cannot be realized either.
However, what I have observed much corresponds to Gervais, Kaniel, Mingelgrin (2001), whose empirical evidence shows that a stock experiencing unusually high (low) trading volume tends to appreciate (depreciate) in the future. It is based on the visibility hypothesis claimed by Miller (1997) and Mayshar (1983) which suggests that high trading volume can attract people to buy the stocks, especially when the stock is limited by short selling. Also,
“any shock that attracts the attention of investors towards a given stock should result in a subsequent price increase, as the set of potential buyers then includes a larger fraction of the market, whereas the set of potential sellers is largely restricted to current stockholders.” [Gervais, Kaniel, Mingelgrin (2001, p.878)] The fact that stocks experiencing unusually low trading volume tend to depreciate can be observed in the cases where profits are calculated two or three weeks after the strategy-formed-day obviously.