![Sample our Economics, Finance,Business & Industry journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5931/TOC-Subject-Sample-2021-Economics-Finance-Business-Industry.jpg"})
Abstract
The New York Stock Exchange extended its trading hours by 30 min in 1974 and in 1985; the first extension resulting in a delayed close and the second in an early open. We find a shift in volume to the new period after each extension. Additionally, there is a larger increase in volume after the 1985 extension than after the 1974 extension. We argue that the second effect is explained by the first. The extension at the end of the day allows some investors to postpone their trades, which results in occasional information cancellation or discovery; this mutes the effect of the extension on volume. In contrast, the extension at the start of the day allows some investors to accelerate trades, which precludes information cancellation or discovery and its negative effect on volume. This explanation suggests that the effect of an extension on volume depends, at least in part, on its timing.
Notes
1. Effective from September 1999, the London Stock Exchange extended its trading hours by opening an hour early. Other markets that have extended their trading hours include Brussels (by 45 min, in October 1999), Paris (30 min, in April 2000), and Milan (20 min, in June 2003). The Frankfurt Stock Exchange extended its afternoon trading session in 2000 but reversed this decision in November 2003 because it did not generate sufficient volume to justify the costs of keeping trading desks open.
2. Trading in the major US markets currently ends at 6:30 p.m. The NYSE added two crossing trading sessions in June 1991, with Session 1 running from 4:15 to 5:00 p.m. EST and Session 2 from 4:15 to 5:15 p.m. In June 2004, Session 2 was extended by an hour, and two additional off-hours crossing sessions, Sessions 3 and 4 were added (both sessions run from 4:00 to 6:30 p.m.). Further extensions of trading hours are being considered by the major exchanges in the face of heightened competition from electronic trading networks.
3. For simplicity, this argument assumes 30-min decision intervals. Note that there is no such trade-off after the early open. This is because acceleration of informed trades to the new period after the early open allows the informed to capitalize on their information early (which precludes information discovery risk) while also trading in a thick market.
4. Our hypothesis predicts that weekly volume on the Tokyo Stock Exchange should increase with Saturday trading, owing to reduced information discovery or cancellation, while weekly return variability should be unaffected (due to unchanged information production).
5. The ISSM data start in 1983 and do not cover the period surrounding the late close.
6. Lo and Wang Citation(2000) show that different de-trending methods yield significantly different results when applied to weekly turnover. Given Lo and Wang's results and the added complication that we are dealing with intraday turnover, we choose the intuitively appealing approach of studying relative turnover.
7. It is not reasonable to disaggregate return variance in the 3:00–4:00 p.m. interval because, unlike turnover, hourly variance depends not just on the sum of half-hourly return variances but also on the correlation between the returns.
8. To the extent that liquidity trading is characterized by temporary price pressure (e.g. Harris and Gurel Citation1986), our analysis of noise trading simultaneously addresses the effects of a possible increase in liquidity trading.
9. It appears unlikely that the increased turnover associated with the early open is driven by lower transaction costs. Using ISSM data, we find the mean of the relative spread over the year before the early open is 2.75% and that over the year after the early open is 2.78% (the relative spread is the raw spread scaled by the mean of the bid and ask prices). A comparison of medians yields similar conclusions.
10. We do not carry out a similar exercise around the late close, since the new period does not overlap with the operating hours of any significant international market.
11. We use the mid-point of the bid and ask quotes to avoid bid-ask bounce. The opening price is the first mid-quote within 15 min of the open. If there are no quotes for a stock within 15 min, we set the opening price to missing.
12. These are computed as the variance of two-day (weekly) returns divided by twice (five times) the variance of daily returns. The two-day returns are overlapping, whereas the weekly returns are non-overlapping and calculated from Wednesday to Wednesday. Chordia, Roll, and Subrahmanyam Citation(2001) show that liquidity is low at the beginning of the week and high at the end of the week, with Wednesdays being relatively normal.
13. To preserve tractability, the model does not feature the trade off faced by informed traders near the end of the day after the late close. As discussed above, this assumption seems reasonable.
14. We start with 10 lags of the dependent variable and remove the insignificant lags to obtain a more parsimonious specification. However, we retain the third lag in the model estimated for turnover around the early open. This does not affect our conclusions but facilitates comparability with the late close results.
15. Alternate benchmark periods (e.g. 10:00 a.m. to 2:00 p.m. for the late close and 11:30 a.m. to 4:00 p.m. for the early open) deliver similar results.
16. Our results are not influenced by our inclusion of volume associated with the call auction, which is usually the first trade of the day. Using ISSM data around the early open, we exclude the first trade of the day and our conclusions are identical.
17. We are unable to analyze the effects of the extensions in Europe because these markets do not have equal-length extensions at the open and the close and we lack intraday European data.
18. Information will induce trade if it generates changes in reservation prices that are different across traders. Changes in traders’ reservation prices can be different either because traders receive different private signals (heterogeneous information) or because they interpret public information (homogeneous information) differently.
19. We assume these probabilities for numerical convenience. Any non-zero probabilities will suffice. The assumption of one category of good or bad news is made for numerical convenience. Our results continue to hold if the different categories of information and their probabilities of occurrence are symmetric between good and bad (e.g. there could be three levels of good news and three of bad news).