Weather and Intraday Patterns in Stock Returns and Trading Activity

(1)

Weather and intraday patterns in stock returns and trading activity

Shao-Chi Chang

a

, Sheng-Syan Chen

b

, Robin K. Chou

c,*

, Yueh-Hsiang Lin

d

a_{Institute of International Business, College of Management, National Cheng Kung University, Taiwan} b_{Department of Finance, College of Management, National Taiwan University, Taiwan}

c_{Department of Finance, School of Management, National Central University, 300 Jhongda Road, Jhongli, Taiwan} d_{Department of Banking and Finance, Takming University of Science and Technology, Taiwan}

Received 26 September 2007; accepted 5 December 2007 Available online 15 December 2007

Abstract

We examine the relation between weather in New York City and intraday returns and trading patterns of NYSE stocks. While stock returns are found to be generally lower on cloudier days, cloud cover has a significant influence on stock returns only at the market open. There are significantly more seller-initiated trades when there is more cloud cover at the market open, which is consistent with the return results. Cloudy skies are associated with higher volatility and less market depth over the entire trading day. Finally, cloud cover is not significantly correlated with spread measures and turnover ratios. The findings overall suggest that weather has a significant influence on investors’ intraday trading behavior.

JEL classiﬁcation: G12; G14

Keywords: Intraday return; Intraday trading activity; Weather; Sentiment

1. Introduction

Traditional finance theory argues in the efficient markets hypothesis that securities markets are rational and appro-priately reflect economic fundamentals. Yet we often read in the popular press that psychological factors have impor-tant influence on the trading decisions people make in financial markets.1An interesting line of research links psy-chological influences and financial markets. Saunders (1993)argues that weather influences stock returns because it affects the mood of investors. There is ample psycholog-ical evidence that people tend to have a more optimistic evaluation of future prospects when they are in a better mood (Arkes et al., 1988; Wright and Bower, 1992; Bagozzi et al., 1999; Hirshleifer, 2001). Saunders documents strong

evidence that stock returns at the New York Stock Exchange (NYSE) are negatively correlated with cloudi-ness. Hirshleifer and Shumway (2003) in an extension of

Saunders (1993) ﬁnd supporting evidence of a negative relation between cloud cover and equity returns in 26 inter-national stock markets.

Other research shows that the influence of biorhythms on mood affects share pricing.Kamstra et al. (2000) inves-tigate the effect of sleep desynchronosis caused by daylight saving time changes on stock returns. They find that inter-ruptions in sleep patterns have a negative influence on stock returns on the Mondays following daylight saving time changes.Kamstra et al. (2003)also document that sea-sonal affective disorder induced by fewer hours of daylight is predictive of a seasonal variation in equity returns.

Both Kamstra et al. (2000) and Hirshleifer and Shum-way (2003) suggest that it is fruitful to investigate the potential eﬀects of psychological factors on intraday returns, intraday volatility, and transactions volume. We thus extend the literature by examining the inﬂuence of

*

Corresponding author. Tel.: +886 3 4227151; fax: +886 3 4252961. E-mail address:rchou@cc.ncu.edu.tw(R.K. Chou).

1 _{See The Wall Street Journal (October 26, 1992, p. C1) and}

Business-Week (February 13, 1995, p. 84, and May 29, 2000, p. 172).

www.elsevier.com/locate/jbf Journal of Banking & Finance 32 (2008) 1754–1766

(2)

weather on intraday trading behavior. Our research makes at least three valuable contributions to the literature. First, previous examinations of the relation between weather and stock markets use aggregate data and base findings on the daily average of weather variables and daily stock returns. Yet it has been well documented in the market microstruc-ture literamicrostruc-ture that there is a strong seasonality in intraday return patterns (Harris, 1986; Atkins and Dyl, 1990; Stoll and Whaley, 1990; Fabozzi et al., 1995). The intraday data we use allow us to measure the immediate impact of weather on stock returns, thus providing a finer picture that cannot be readily extracted from aggregating daily observations. The advantage of intraday information is particularly valuable if weather influences stock returns more significantly at certain trading hours, say, at market opening periods, and we cannot capture those effects by examining daily data.

Second, intraday data enable us to investigate the effects of weather on daily and intraday trading activities in the market. Most prior studies investigate the weather impact on stock returns only.2 However, the literature suggests that investor sentiment may affect trading activities.Baker and Stein (2004)argue that in the presence of short-sales constraints, traders become more active in an overvalued market. As a result, investor sentiment is positively corre-lated with market liquidity.Mehra and Sah (2002)propose a model demonstrating that mood fluctuation induced by projection bias has an important influence on the volatility of equity prices. Brown (1999) provides strong evidence that both the number of trades and the return volatility of closed-end funds increase with unusual levels of investor sentiment.Lee et al. (2002)find that bullish shifts in senti-ment are negatively correlated with market volatility and positively associated with future excess returns. Therefore, if weather affects the mood and sentiment of investors, it is likely that weather will have important effects on their trad-ing behavior as well. The relation between weather and trading activities may be different for various intraday trad-ing intervals, because tradtrad-ing patterns have been found to vary substantially among different trading intervals (Wood et al., 1985; Foster and Viswanathan, 1993).

Third, intraday data permit a more reliable and efficient estimation of the effect of psychological factors on share prices (Barclay and Litzenberger, 1988; Busse and Green, 2002). The short measurement period reduces the sources of variability that may be attributed to other unrelated extraneous factors. These advantages are important for the interpretation of the effect of psychological factors on stock markets, given that the correlation between mood-related variables and stock returns may be driven by outli-ers and subsamples (Pinegar, 2002).

We examine the relation between local New York City weather and intraday returns and trading patterns of

NYSE stocks. That is, we explore the effects of cloud cover in New York City on intraday returns and a variety of trading variables for NYSE firms, including trading volume, bid-ask spread, quoted depth, return volatility, and order imbalance. We focus on cloud cover in light of the important findings in Saunders (1993) and Hirshle-ifer and Shumway (2003), as well as the psychological evi-dence that sunshine is among the most important weather variables affecting mood.3 Our use of local New York City weather is based on the argument suggested by Saun-ders (1993) that traders physically present in New York City, such as brokers and floor traders, may sometimes affect prices in attempts to exploit their own interests. Because they assemble at the same location daily, a strictly local mood variable has the potential to affect this group to the exclusion of other market participants, who are geographically dispersed. In fact, our test results are biased against finding any effects of the New York City weather. The local traders in New York City are less likely to overlook exploitable opportunities induced by weather effects. Moreover, their number is small relative to the universe of market participants, and the existence of weather effect would show that weather in New York City does have impact on returns and trading activities, despite the dispersed locations of traders across the country.

We find that while stock returns are generally lower on cloudier days, cloud cover has a significant influence on stock returns only during the first 15-min interval of the trading day; its effect becomes insignificant for subsequent trading intervals. The effect is similar for order imbalance in that during only the first 15-min interval of the trading day, there are significantly more seller-initiated trades than buyer-initiated trades on cloudier days. For other trading hours, cloud cover does not affect order imbalance signifi-cantly. Our results on intraday returns and order imbal-ance suggest that when investors are gloomy because of cloudy weather around the market open, they tend to be pessimistic and be less inclined to buy than to sell, resulting in lower stock returns during the opening interval. Our evi-dence supports the argument suggested by Lo and Repin (2002)that investors experiencing significant psychological changes around opening hours reflect their mood in the opening trades, but these transient changes quickly become less important as more information comes to the market during the trading day.

Our findings on the very short-run impact of cloud cover on stock returns and sell-initiated trades suggest that psy-chological factors only have temporary influence on returns. Even though previous studies find that cloud cover seems to affect returns, they also point out that this phe-nomenon does not easily present a profitable opportunity.

2 _{Loughran and Schultz (2004)}_{study weather and trading volume, but}

do not investigate the eﬀect of weather on intraday trading.

3 _{The cloudier the weather is, the worse investor mood will be, and the}

more pessimistic investors become. The psychological evidence can be found in Persinger (1975), Cunningham (1979), and Howarth and Hoﬀman (1984).

(3)

Hirshleifer and Shumway (2003) argue that the weather effect may be profitable only for very low cost traders, but it is not going to be profitable for most investors. Our evidence of the very short-term weather effect suggests that trading on it is less likely to be profitable.

We also show that cloudy skies are associated with higher volatility and less market depth. The positive rela-tion between cloud cover and market volatility is consistent with the hypothesis that when moods are pessimistic, there is more disagreement in opinion among investors, so stock return is more volatile (Lee et al., 2002; Baker and Stein, 2004). Our ﬁnding of a negative relation between cloud cover and market depth is consistent with the argument that gloomy investors tend to be pessimistic and have less desire to participate in market activities (Loughran and Schultz, 2004; Goetzmann and Zhu, 2005).

We further show that the effects of cloud cover on return volatility and market depth are significant not merely for the opening hours, but also last for the entire trading day. It is well documented in the literature that shocks in volatility are highly persistent (Engle, 1982; Bollerslev, 1986; Nelson, 1991; Bollerslev and Engle, 1993). Moreover, previous studies find that market depth is negatively related to volatility (Bessembinder and Seguin, 1993; Ahn et al., 2001; Goldstein and Kavajecz, 2004). Therefore, the effects of cloud cover on volatility and market depth tend to last longer.

We finally find that cloud cover is not significantly cor-related with spread measures and turnover ratios, suggest-ing that weather is relatively unimportant in explainsuggest-ing these trading variables. Our results are similar for equally weighted or value-weighted stock indices and for individual stocks, and remain robust after controlling for other weather variables (Cooke et al., 2000; Hirshleifer and Shumway, 2003; Loughran and Schultz, 2004; Cao and Wei, 2005), day-of-the-week and month-of-the-year effects (Cross, 1973; Smirlock and Starks, 1986; Thaler, 1987a,b), daylight saving time changes (Kamstra et al., 2000), and intraday seasonality. Our findings overall suggest that weather has a significant influence on investors’ intraday trading behavior – a phenomenon not yet documented in the literature.

The remainder of the paper proceeds as follows. Section

2 develops our hypotheses. Section3 describes the sample and methodology. We present the empirical results in Sec-tion4. The ﬁnal section concludes the paper.

2. Hypothesis development

We develop a variety of hypotheses on the relation between investor sentiment and stock returns and trading activities throughout the entire day and among different intraday trading intervals. We first explore the impact of investor mood on returns. We then discuss the potential effects of investor sentiment on trading variables, such as return volatility, trading volume, bid-ask spread, quoted depth, and order imbalance.

Given the psychological evidence that people tend to be more optimistic as to future prospects when they are in a better mood (Wright and Bower, 1992; Bagozzi et al., 1999), we expect the relation between investor sentiment and stock returns to be positive.Saunders (1993) and Hir-shleifer and Shumway (2003) using daily data show sup-porting evidence for the entire trading day that cloudy weather in New York City negatively influences stock returns at the NYSE by affecting the mood of investors. However, the literature documents that there are signifi-cant intraday return patterns (Harris, 1986; Atkins and Dyl, 1990; Stoll and Whaley, 1990; Fabozzi et al., 1995). It is thus likely that weather may affect stock returns more significantly at certain intraday trading intervals. For trad-ers located physically in New York City, the good mood induced by sunshine may have a more profound influence on their decision-making process at the beginning of trad-ing hours, as investors observtrad-ing the weather and experi-encing psychological changes right around the market open reflect their mood in their opening trades (Lo and Repin, 2002). Nevertheless, as more information arrives in the market during the trading day, the influence of weather on stock returns may diminish quickly. We thus predict a stronger weather effect on stock returns at the market open that may not last for the entire trading day.

There are two sets of competing, but not mutually exclu-sive, arguments predicting the relation between investor sentiment and return volatility. As suggested by Shiller (2003) and Nofsinger (2005), on the one hand, the poorer the social mood is, the more disagreement in opinion among investors. As a consequence of increased diﬀerences in valuation among investors, return volatility increases (Harris and Raviv, 1993; Shalen, 1993; Lee et al., 2002; Baker and Stein, 2004). Investor sentiment is thus expected to have a negative impact on return volatility, implying that cloudy weather is associated with higher return volatil-ity.Brown (1999), Gervais and Odean (2001), and Statman et al. (2006), however, suggest that when the sentiments are bullish, investors may be overconﬁdent and trade more, which causes return volatility to rise. These arguments would indicate a positive relation between investor senti-ment and return volatility, implying that cloudy weather is associated with less return volatility.

Investor heterogeneity is likely to contribute to higher trading activity (Karpoff, 1986; Harris and Raviv, 1993). When sentiments are bullish, investors may become over-confident, overestimate the relative precision of their own private signals, and underestimate the information content embodied in either order flow or equity issues and others’ trading decisions, as they consider others to be less well-informed than they are (Baker and Stein, 2004). Therefore, a market whose pricing is dominated by bullish sentiment levels is unusually liquid. A highly liquid market is usually characterized by high depth and trading volume as well as narrow bid-ask spreads. When investors are in a down mood, they may be pessimistic and have less of a desire to trade and tend to sell rather than buy (Loughran and

(4)

Schultz, 2004; Goetzmann and Zhu, 2005). These argu-ments suggest that investor sentiment is expected to be pos-itively related to the turnover ratio, market depth, and number of buy orders, and negatively related to the bid-ask spread. Therefore, we hypothesize that on cloudier days, there will be lower turnover ratios, market depth, and numbers of buy orders, but higher bid-ask spreads.

Investor sentiment is likely to have differential effects on stock returns over different intraday trading intervals. Since stock price changes are caused by investors’ trading activity, there may also be a different relation between weather and trading variables for various trading intervals. Furthermore, it is well documented that the trading pattern at the market open is substantially different from that for the rest of the day. Trading variables, including volume, spread, depth, and volatility, have all been found to vary substantially among different intraday trading intervals (Wood et al., 1985; Jain and Joh, 1988; McInish and Wood, 1992; Foster and Viswanathan, 1993; Chung et al., 1999). Therefore, it is possible that weather has dif-ferent effects on trading variables in the opening trading interval and in the other trading intervals. As we examine the influence of weather on stock returns and trading vari-ables both throughout the entire day and among different intraday trading intervals, we provide a finer picture of the effects of investor sentiment on intraday returns and trading patterns, not so far documented in any other study.

3. Data and methodology

Our sample includes firms listed on the NYSE, which are covered by the Trade and Quote (TAQ) database and the University of Chicago’s Center for Research in Security Prices (CRSP) database in the period 1994–2004. We elim-inate firms in the financial and utility industries (SIC codes 6000–6999 and 4900–4999). Intraday trading measures such as trade prices, bid-ask quotes, trading volume, and quote size are from TAQ. Shares outstanding and the mar-ket value of the sample firms are from CRSP.

To minimize data errors, we followChordia et al. (2002)

and apply several ﬁlters to the data. First, a trade is excluded if it is out of sequence or has special settlement conditions, because it might then be subject to distinct liquidity considerations. Second, quotes recorded outside the regular trading hours (9:30–16:00) are excluded. Third, observations with negative bid-ask spreads are discarded. Finally, only BBO (best bid or oﬀer)-eligible primary mar-ket (NYSE) quotes are retained.4

The weather data of New York City come from the International Surface Weather Observations (ISWO) data-set provided by the National Climatic Data Center.5 Weather variables are recorded every hour throughout

the entire day. We match the hourly weather variables and the return and trading data by splitting a trading day (9:30–16:00) into eight intervals. The period from 10:00 to 16:00 is divided into six 60-min intervals to coincide with the recorded weather variable. The ﬁrst 30-min (9:30– 10:00) of the trading period is split further into two 15-min intervals (9:30–9:45 and 9:45–10:00) in order to test the weather eﬀect at the market open.

In each trading interval, we find the nearest hourly weather observations right before the beginning of that interval. Stock returns and trading variables are calculated for each firm during each trading interval, and then aver-aged across all firms either equally weighted or value weighted by a firm’s market capitalization as of the end of the prior month. We calculate the interval return (RET) to check whether intraday returns are correlated with weather. Trading measures include volatility, market depth, spread, turnover ratio, and order imbalance variables. Two volatility variables are calculated, including price range (RANGE), which is calculated as the interval’s high price minus low price, and standard deviation (RETSTD), which is calculated as the percentage standard deviation of the bid-ask mid-point returns. Market depth (DEPTH) is measured by the average quote size at the best bid and ask prices. Spreads are highly serially correlated and exhi-bit strong intraday patterns. To control for intraday spread autocorrelation and seasonality, we follow Chordia et al. (2002) and Goetzmann and Zhu (2005) by examining the difference between the spread of the interval and the spread of the same interval on the prior trading day. We calculate two first difference spread measures, the percentage effec-tive (DIF_ES) and percentage quoted (DIF_QS) first differ-ence spreads.6For variables related to trading volume, we calculate two turnover ratios, average trading volume per trade (TURNPER) and cumulative trading volume (TURN) in the interval, both scaled by number of firm shares outstanding at the end of the previous month. We calculate two order imbalance ratios. The first is based on trading volume (OISVOL), and is calculated as the trad-ing volume of seller-initiated trades divided by the total trading volume in the interval. The other is based on num-ber of trades (OISNUM), and is calculated as the numnum-ber of seller-initiated trades divided by the total number of trades.7 Finally, to make variables in 15-min and 60-min intervals comparable, we multiply the interval return, price range, and cumulative trading volume in the first two 15-min intervals by 4, and multiply the standard deviations in the first two 15-min intervals by the square root of 4.

Our speciﬁc focus is on whether intraday return and trading activity are related to cloud cover, a factor found

4 _{Chordia et al. (2001)}_{provide a justiﬁcation for using only NYSE}

quotes.

5 _{The same data source is used in}_{Hirshleifer and Shumway (2003) and}

Loughran and Schultz (2004).

6 _{The percentage eﬀective spread is deﬁned as twice the absolute value of}

the diﬀerence between the trading price and the mid-point of the ask and the bid prices, scaled by the mid-point of the ask and the bid prices. The percentage quoted spread is the diﬀerence between the ask price and the bid price scaled by the mid-point of the ask and bid prices.

(5)

to have a significant influence on returns (Saunders, 1993; Hirshleifer and Shumway, 2003). We index cloud cover (CC) from 1 to 4, where 1 indicates a clear sky, 2 indicates scattered clouds, 3 indicates broken clouds, and 4 indicates overcast.8Because all the weather variables are highly sea-sonal, it is important to control for seasonality. To make sure that the results are not driven by seasonal effects, we follow Hirshleifer and Shumway (2003)and deseasonalize each weather variable by subtracting its average value of each calendar week during the sample period from the weather observations at New York City.

We regress stock returns and each trading variable on cloud cover. As the eﬀect of cloud cover may be driven by other adverse weather conditions, we control for other weather variables examined in prior research, including snowiness (Loughran and Schultz, 2004), raininess ( Hir-shleifer and Shumway, 2003), temperature (Cao and Wei, 2005), and wind speed (Cooke et al., 2000). A dummy var-iable for snowiness (Dsnow) is deﬁned as 1 if the data from

ISWO show that it is snowing or has snowed within the last observation period. Raininess (Drain) is deﬁned similarly.

Temperature (TEMP) is measured in Fahrenheit, and wind speed (WIND) is measured by miles per hour. All these weather variables are deseasonalized in a similar way as cloud cover. We also control for day-of-the-week and month-of-the-year effects (Cross, 1973; Smirlock and Starks, 1986; Thaler, 1987a,b), effect of daylight saving time changes in the fall and spring (Kamstra et al., 2000), and the intraday trading session effect. Dummy variables for Monday (DMon), Friday (DFri), January (DJan),

Decem-ber (DDec), and two days of the daylight saving time

changes (one for the ﬁrst Sunday in April, DDLApr, and

another for the last Sunday in October, DDLOct) and

dummy variables for trading intervals are included in the regressions.9

Table 1shows the number of sample firms in each month for 1994–2004. The number of NYSE listed firms in our sample increases initially, then drops slightly afterwards, and finally stabilizes at around 1400. Table 2 shows the summary statistics of weather variables, stock returns, and trading measures. Panel A indicates average cloud cover of 1.78 and average temperature of 56.97 degrees Fahrenheit in New York City. Panels B and C show the equally weighted and value-weighted stock returns and trading variables, respectively. The average interval return (RET) is close to zero. The averages of the equally weighted return standard deviations (RETSTD) and turnover ratios (TURNPER and TURN) are higher than those of the value-weighted variables, indicating that small firms are more volatile and traded more actively. The lag-one differ-ences in quoted and effect spreads (DIF_ES and DIF_QS) are on average close to zero. There are generally more buyer-initiated trades on the market, as the average order imbalance ratio (OISVOL and OISNUM) is less than 0.5.

4. Empirical results

4.1. The impact of cloud cover on intraday returns

Table 3 presents results of stock returns regressed against cloud cover and control variables. Panel A shows the equally weighted regression results. While the results for the entire trading day indicate that stock returns are generally lower on cloudier days, cloud cover has a signif-icantly negative impact on interval returns only during the opening interval. For other intraday trading intervals, cloud cover does not significantly influence stock returns. Our findings indicate that while cloud cover affects stock returns, its influence is only short-term. The evidence is consistent with the hypothesis that investors observing the weather and experiencing psychological changes right

Table 1

Number of sample ﬁrms in each month from 1994 through 2004

Year January February March April May June July August September October November December

1994 1252 1253 1253 1251 1256 1267 1273 1282 1287 1291 1291 1303 1995 1286 1284 1337 1344 1343 1347 1357 1358 1367 1359 1374 1407 1996 1401 1417 1417 1431 1436 1444 1456 1439 1457 1470 1499 1531 1997 1537 1547 1545 1539 1554 1564 1590 1589 1592 1600 1609 1633 1998 1621 1626 1634 1641 1638 1641 1650 1647 1641 1639 1639 1643 1999 1643 1622 1626 1629 1609 1599 1600 1580 1577 1573 1590 1592 2000 1583 1577 1564 1543 1519 1524 1504 1493 1477 1474 1467 1448 2001 1436 1468 1459 1463 1473 1465 1465 1455 1465 1451 1461 1466 2002 1458 1457 1461 1457 1456 1464 1465 1459 1455 1451 1460 1461 2003 1458 1456 1449 1451 1446 1450 1445 1442 1446 1446 1449 1452 2004 1456 1454 1452 1451 1439 1434 1433 1435 1438 1435 1444 1451

Our sample includes firms listed on the NYSE, which are covered by the Trade and Quote (TAQ) database and the University of Chicago’s Center for Research in Stock Prices (CRSP) database. We delete firms in the financial industry (SIC codes 6000–6999) and utility industry (SIC codes 4900–4999).

8 _{The data description indicates that a clear sky represents cloud cover}

of less than 1/8, scattered clouds is cloud cover between 1/8 and 4/8, broken clouds is cloud cover between 5/8 and 7/8, and overcast is cloud cover of more than 7/8. To test the robustness of our results, we also use an alternative measure for each group: 0 for clear sky, 5/16 (the average of 1/8 and 4/8) for scattered clouds, 3/4 (the average of 5/8–7/8) for broken clouds, and 1 for overcast sky. The results are similar.

9 _{The baseline trading interval is the ﬁrst trading interval of each}

(6)

around the market open reflect their mood in the opening trades, and the impact of these changes on returns is tran-sient, vanishing quickly as more information arrives in the market during the trading day (Lo and Repin, 2002). Our results also suggest that the relation between cloud cover and daily returns found bySaunders (1993) and Hirshleifer and Shumway (2003)is likely to be driven by the influence of weather in the opening period. Our evidence on the short-term effect of cloud cover offers one potential expla-nation for the inconsistent evidence in previous research that uses daily average data.10

In Panel A, we also take a closer look at the period of after lunch hours, assuming that most of the traders go out for lunch and observe the weather again. The period of after lunch hours is defined from 13:00 to 13:30. We obtain similar results if it is defined from 12:00 to 12:30 or from 12:30 to 13:00, so we do not report them here.11 Panel A shows that stock returns during the after lunch hours are not significantly related to cloud cover. There are two possible explanations for our findings. First, unlike the market open, where all traders on the NYSE reflect their mood on trading at the same time, traders

Table 2

Summary statistics

Variable Mean Standard deviation Q1 Median Q3 N

Panel A: weather variables

CC 1.7842 0.9809 1 2 3 21,629

WIND 12.1122 5.8568 8 11 15 21,629

Dsnow 0.0172 0.1302 0 0 0 21,629

Drain 0.0676 0.2511 0 0 0 21,629

TEMP 56.9720 17.4419 44 57 72 21,629

Panel B: equally weighted stock return and trading variables

RET 0.0039 0.3591 0.1239 0.0119 0.1286 21,629 RANGE 0.1835 0.2038 0.0212 0.1388 0.2494 21,629 RETSTD 0.7677 0.2344 0.5887 0.7208 0.8998 21,629 DEPTH 95.8697 55.4325 38.8903 90.7375 146.0107 21,629 DIF_ES 0.0001 0.0390 0.0148 0.0002 0.0145 21,629 DIF_QS 0.0002 0.0342 0.0177 0.0009 0.0166 21,629 TURNPER 0.1143 0.2374 0.0450 0.0741 0.0990 21,629 TURN 1.5148 4.0009 0.6227 0.8544 1.2373 21,629 OISVOL 0.4719 0.0400 0.4453 0.4695 0.4960 21,629 OISNUM 0.4733 0.0375 0.4492 0.4730 0.4967 21,629

Panel C: value-weighted stock return and trading variables

RET 0.0024 0.4608 0.1781 0.0082 0.1806 21,629 RANGE 0.4026 0.4622 0.0706 0.2571 0.5193 21,629 RETSTD 0.6117 0.2165 0.4546 0.5660 0.7193 21,629 DEPTH 148.2820 95.3379 63.0399 120.2533 220.6747 21,629 DIF_ES 0.0001 0.0257 0.0047 0.0002 0.0046 21,629 DIF_QS 0.0001 0.0162 0.0068 0.0003 0.0062 21,629 TURNPER 0.0096 0.0063 0.0040 0.0074 0.0149 21,629 TURN 0.4852 0.1755 0.3545 0.4503 0.5864 21,629 OISVOL 0.4630 0.0515 0.4292 0.4593 0.4925 21,629 OISNUM 0.4705 0.0420 0.4437 0.4690 0.4947 21,629

The sample period is from January 1, 1994, through December 31, 2004. We split the trading day (9:30–16:00) into eight intervals. The period from 10:00 to 16:00 is divided into six 60-min intervals to coincide with the recorded weather variable. The ﬁrst 30-min (9:30–10:00) of the trading period is further split into two 15-min intervals (9:30–9:45 and 9:45–10:00). For each interval, we ﬁnd the nearest hourly weather variables in New York City before the beginning of that interval. The weather variables include (i) CC: cloud cover which ranges from 1 (clear) to 4 (overcast); (ii) WIND: wind speed in miles per hour; (iii) Dsnow: dummy variable for snowiness; (iv) Drain: dummy variable for raininess; (v) TEMP: temperature in Fahrenheit. In Panels B and C, we

calculate stock returns and trading variables for each firm during each trading interval, and then calculate the equally weighted (Panel B) and value-weighted (Panel C) averages across the firms. The variables include (i) RET: percentage return; (ii) RANGE: price range; (iii) RETSTD: percentage standard deviation of the bid-ask mid-point return; (iv) DEPTH: average quote size, defined as the sum of the bid size and ask size, in 100 shares; (v) DIF_ES: difference between the percentage effective spread and that of the last trading day, where effective spread is defined as twice the absolute value of the difference between the trading price and the bid-ask mid-point, scaled by the mid-point; (vi) DIF_QS: difference between the percentage quoted spread and that of the last trading day, where quoted spread is defined as the difference between the ask and the bid prices scaled by the bid-ask mid-point; (vii) TURNPER: average trading volume per trade scaled by shares outstanding at the end of the last month; (viii) TURN: total trading volume scaled by the outstanding shares at the end of the last month; (ix)OISVOL: order imbalance ratio by trading volume calculated as the trading volume of the initiated trades divided by the total trading volume; and (x) OISNUM: order imbalance ratio by number of trades calculated as the number of the seller-initiated trades divided by the total number of trades.

10 _{For example,}_{Loughran and Schultz (2004)}_{study the weather in the}

city of a ﬁrm’s headquarters and its stock returns. They ﬁnd no evidence that cloud cover is related to equity returns.

11_{We also use a 15-min interval to deﬁne the period of after lunch hours.}

That is, we deﬁne the after lunch hours as 12:00–12:15, 12:30–12:45, or 13:00–13:15. The conclusions in this study remain unchanged.

(7)

may go for lunch during diﬀerent time intervals. While they may reﬂect their mood on trading due to weather observed during the lunch hours, the diverse lunch sched-ules are likely to cause the impact of weather to be

insig-niﬁcant. Second, as we argue above, the arrivals of other economically important information during the period of after lunch hours may also reduce the impact of weather on returns.

Table 3

Regressions of intraday return

Variable Trading interval

9:30–16:00 9:30–9:45 9:45–10:00 9:45–16:00 13:00–13:30 15:00–16:00

Panel A: equally weighted index

Intercept 0.0361*** _0.0352** _0.0581*** _0.0420*** _0.0022 _0.0470*** CC 0.0034 0.0316*** _0.0070 _0.0007 _0.0019 _0.0001 WIND 0.0002 0.0007 0.0006 0.0003 0.0001 0.0003 Dsnow 0.0265 0.1351 0.0108 0.0133 0.0041 0.0190 Drain 0.0143 0.0720 0.0002 0.0057 0.0048 0.0339* TEMP 0.0001 0.0023 0.0007 0.0001 0.0000 0.0005 DMon 0.0078 0.0149 0.0345 0.0068 0.0005 0.0348*** DFri 0.0111* 0.0105 0.0084 0.0110** 0.0088** 0.0073 DJan 0.0035 0.0572 0.0139 0.0038 0.0064 0.0074 DDec 0.0304*** 0.1032** 0.0800** 0.0203*** 0.0020 0.0160 DDLApr 0.0369 0.0636 0.2934* 0.0329 0.0023 0.0855 DDLOct 0.0344 0.0877 0.2328 0.0276 0.0093 0.2047***

Interval dummies Yes N/A N/A Yes N/A N/A

Adj. R2(%) 0.66 0.38 0.01 0.67 0.10 0.59

N 21,629 2649 2649 18,980 2708 2708

Panel B: value-weighted index

Intercept 0.0080 0.0030 0.0547*** _0.0314*** _0.0048 _0.0235** CC 0.0024 0.0379*** _0.0178 _0.0030 _0.0075 _0.0016 WIND 0.0001 0.0011 0.0012 0.0000 0.0003 0.0010 Dsnow 0.0179 0.0798 0.0419 0.0103 0.0012 0.0328 Drain 0.0209* 0.1122** 0.0021 0.0080 0.0021 0.0362 TEMP 0.0001 0.0003 0.0013 0.0001 0.0000 0.0011 DMon 0.0089 0.0116 0.0829** 0.0086 0.0005 0.0153 DFri 0.0010 0.0040 0.0158 0.0007 0.0038 0.0100 DJan 0.0077 0.0896* 0.0387 0.0036 0.0107 0.0242 DDec 0.0335*** 0.1879*** 0.0811 0.0120 0.0124 0.0273 DDLApr 0.0615 0.1884 0.3602 0.0428 0.0114 0.1512 DDLOct 0.0858* 0.1751 0.1969 0.0743 0.0129 0.3221***

Adj. R2(%) 0.10 0.81 0.10 0.07 0.16 0.32

N 21,629 2649 2649 18,980 2708 2708

Panel C: individual stock

Intercept 0.0677*** _0.0517*** _0.0545*** _0.0483*** _0.0049* _0.0473*** CC 0.0065 0.0873*** _0.0026 _0.0004 _0.0042 _0.0011 WIND 0.0000 0.0004 0.0008 0.0000 0.0004 0.0003 Dsnow 0.0229 0.0023 0.0347 0.0076 0.0562 0.0636 Drain 0.0069 0.0256 0.0279 0.0032 0.0117 0.0019 TEMP 0.0000 0.0117 0.0001 0.0001 0.0003 0.0005 DMon 0.0125*** 0.0127 0.0343*** 0.0108*** 0.0025 0.0068 DFri 0.0135*** 0.0429*** 0.0216*** 0.0149*** 0.0070*** 0.0100*** DJan 0.0007 0.0577*** 0.0099 0.0061*** 0.0075*** 0.0039 DDec 0.0214*** 0.0586*** 0.0618*** 0.0145*** 0.0029 0.0140*** DDLApr 0.0287** 0.1710 0.2305*** 0.0097 0.0068 0.0835*** DDLOct 0.0453*** 0.0676 0.1651*** 0.0487*** 0.0766*** 0.2142***

The dependent variable is the intraday return. The independent variables include: (i) cloud cover (CC) and weather control variables, including wind speed (WIND), snowiness (Dsnow), raininess (Drain), and temperature (TEMP); (ii) dummy variables of Monday (DMon), Friday (DFri), January (DJan), December

(DDec), and two days of the daylight saving time changes (one for the ﬁrst Sunday in April, DDLApr, and another for the last Sunday in October, DDLOct);

and (iii) dummy variables indicating trading intervals (baseline comparison is to the first trading interval of each regression). A trading day is divided into two 15-min intervals after the market open, and six 60-min intervals for the following trading hours. Each weather variable is deseasonalized by subtracting its average value of each calendar week during the sample period from the weather observations at New York City. Panels A and B show the ordinary least squares regression results for the equally weighted and value-weighted indices, respectively. In Panel C, we perform time-series regressions on each stock and then calculate the cross-sectional averages of regression coefficients. The coefficients on the dummy variables indicating trading intervals are suppressed to save space. ‘‘***_{”, ‘‘}**_{”, and ‘‘}*_{” denote significance at the 1%, 5%, and 10% levels, respectively.}

(8)

Table 4

Regressions of 1-min return for the ﬁrst 30-min after the market open

Trading interval CC WIND Dsnow Drain TEMP Adj. R2(%)

Panel A: equally weighted index

9:30–9:31 0.0078* _0.0003 _0.0236* _0.0003 _0.0000 _0.13 9:31–9:32 0.0042 0.0001 0.0108 0.0030 0.0001 0.22 9:32–9:33 0.0038* _0.0001 _0.0024 _0.0016 _0.0004*** _0.31 9:33–9:34 0.0031* _0.0001 _0.0041 _0.0009 _0.0002** _0.04 9:34–9:35 0.0034* _0.0001 _0.0011 _0.0013 _0.0001 _0.07 9:35–9:36 0.0058*** _0.0000 _0.0089 _0.0042 _0.0001 _0.18 9:36–9:37 0.0051*** _0.0000 _0.0032 _0.0023 _0.0000 _0.28 9:37–9:38 0.0026* _0.0001 _0.0018 _0.0013 _0.0000 _0.12 9:38–9:39 0.0028* _0.0001 _0.0061 _0.0006 _0.0000 _0.18 9:39–9:40 0.0018 0.0000 0.0013 0.0022 0.0001 0.02 9:40–9:41 0.0040*** _0.0001 _0.0049 _0.0048** _0.0001* _0.45 9:41–9:42 0.0036*** _0.0001 _0.0088** _0.0016 _0.0002*** _0.47 9:42–9:43 0.0019 0.0000 0.0078** _0.0002 _0.0000 _0.03 9:43–9:44 0.0020 0.0000 0.0078** _0.0022 _0.0000 _0.13 9:44–9:45 0.0008 0.0001 0.0006 0.0016 0.0000 0.16 9:45–9:46 0.0013 0.0001 0.0036 0.0010 0.0000 0.05 9:46–9:47 0.0009 0.0000 0.0065* _0.0007 _0.0000 _0.04 9:47–9:48 0.0017 0.0001 0.0029 0.0009 0.0000 0.01 9:48–9:49 0.0001 0.0000 0.0021 0.0005 0.0000 0.02 9:49–9:50 0.0011 0.0001 0.0015 0.0007 0.0000 0.05 9:50–9:51 0.0009 0.0001 0.0036 0.0002 0.0001 0.09 9:51–9:52 0.0008 0.0000 0.0007 0.0010 0.0000 0.27 9:52–9:53 0.0001 0.0001 0.0031 0.0021 0.0000 0.02 9:53–9:54 0.0006 0.0000 0.0010 0.0033** _0.0001** _0.22 9:54–9:55 0.0009 0.0000 0.0008 0.0005 0.0000 0.29 9:55–9:56 0.0019* _0.0000 _0.0002 _0.0028* _0.0000 _0.15 9:56–9:57 0.0010 0.0000 0.0023 0.0016 0.0000 0.10 9:57–9:58 0.0008 0.0000 0.0019 0.0026* _0.0001 _0.10 9:58–9:59 0.0005 0.0000 0.0029 0.0010 0.0000 0.12 9:59–10:00 0.0006 0.0000 0.0019 0.0003 0.0000 0.10

Panel B: value-weighted index

9:30–9:31 0.0069* _0.0005 _0.0251* _0.0028 _0.0003 _0.28 9:31–9:32 0.0113*** _0.0002 _0.0080 _0.0050 _0.0001 _0.04 9:32–9:33 0.0027 0.0000 0.0015 0.0021 0.0004** _0.05 9:33–9:34 0.0038* _0.0004** _0.0108 _0.0003 _0.0003** _0.44 9:34–9:35 0.0031* _0.0001 _0.0047 _0.0054 _0.0000 _0.33 9:35–9:36 0.0128*** _0.0000 _0.0009 _0.0076** _0.0000 _0.06 9:36–9:37 0.0035 0.0001 0.0025 0.0060 0.0003** _0.15 9:37–9:38 0.0080** _0.0001 _0.0073 _0.0046 _0.0001 _0.12 9:38–9:39 0.0003 0.0000 0.0057 0.0011 0.0001 0.12 9:39–9:40 0.0043* _0.0001 _0.0089 _0.0004 _0.0001 _0.21 9:40–9:41 0.0060** _0.0000 _0.0056 _0.0073** _0.0000 _0.21 9:41–9:42 0.0047** _0.0001 _0.0121* _0.0025 _0.0001 _0.11 9:42–9:43 0.0016 0.0001 0.0084 0.0020 0.0000 0.21 9:43–9:44 0.0016 0.0001 0.0042 0.0030 0.0001 0.08 9:44–9:45 0.0002 0.0002 0.0039 0.0035 0.0000 0.02 9:45–9:46 0.0036 0.0000 0.0010 0.0040 0.0000 0.01 9:46–9:47 0.0012 0.0002 0.0081 0.0026 0.0000 0.09 9:47–9:48 0.0007 0.0001 0.0031 0.0009 0.0002* _0.05 9:48–9:49 0.0034 0.0001 0.0124* _0.0008 _0.0000 _0.19 9:49–9:50 0.0030 0.0003** _0.0031 _0.0040 _0.0001 _0.18 9:50–9:51 0.0019 0.0000 0.0119* _0.0017 _0.0001 _0.10 9:51–9:52 0.0029 0.0000 0.0044 0.0026 0.0001 0.09 9:52–9:53 0.0005 0.0002* _0.0004 _0.0013 _0.0001 _0.08 9:53–9:54 0.0000 0.0001 0.0047 0.0032 0.0002* _0.04 9:54–9:55 0.0016 0.0000 0.0088 0.0013 0.0001 0.12 9:55–9:56 0.0024 0.0001 0.0089 0.0123*** _0.0000 _0.25 9:56–9:57 0.0009 0.0001 0.0021 0.0025 0.0001 0.03 9:57–9:58 0.0008 0.0002 0.0005 0.0061** _0.0001 _0.09 9:58–9:59 0.0031 0.0000 0.0026 0.0002 0.0000 0.11 9:59–10:00 0.0016 0.0000 0.0042 0.0018 0.0000 0.13

(9)

Panel A also shows detailed regression results of returns on weather control variables, day of the week, cal-endar month, and daylight saving time changes. In gen-eral, weather control variables (i.e., WIND, Dsnow, Drain,

and TEMP) are not signiﬁcantly related to returns throughout the entire day or among various intraday trad-ing intervals, except for Drainin the closing trading period

(15:00–16:00) (marginally significant at the 10% level). In the regression for entire day returns (9:30–16:00), Friday has significantly higher returns, consistent with findings in Cross (1973), Keim and Stambaugh (1984), and Smir-lock and Starks (1986). The Monday effect is concentrated in the period of closing trading. The January effect is weak during our sample period, similar to findings in Maberly and Maris (1991) and Szakmary and Kiefer (2004). Con-trary to Kamstra et al. (2000), we find that the effects of daylight saving time changes in April and October are not significant for the entire day, but are significant in some trading intervals.

Panel B of Table 3 presents the value-weighted regres-sion results of returns. As in Panel A, interval returns are significantly negatively related to cloud cover only during the opening interval, and the relation is weak beyond the first 15-min trading interval. These findings again show that cloud cover affects stock returns negatively, but its influence lasts for only a short period.

We also examine individual stocks traded on the NYSE. We perform time-series regressions on each stock and then calculate the cross-sectional averages of regres-sion coefficients. As shown in Panel C of Table 3, the impact of cloud cover on returns is again significantly negative at the market open only and its effect diminishes quickly.

We further closely examine finer intervals during the first 30-min after the market open.Table 4reports the min-ute-by-minute results for the equally weighted index, the value-weighted index, and individual stocks, where only the coefficients of weather variables are presented in order

Table 4 (continued)

Trading interval CC WIND Dsnow Drain TEMP Adj. R

2

(%) Panel C: individual stock

9:30–9:31 0.0127** _0.0002 _0.0622 _0.0270** _0.0000 9:31–9:32 0.0055* _0.0009 _0.4140** _0.0328 _0.0000 9:32–9:33 0.0015 0.0003 0.0571 0.0166 0.0001 9:33–9:34 0.0037*** _0.0002** _0.0024 _0.0012 _0.0002* 9:34–9:35 0.0027*** _0.0001 _0.0024 _0.0009 _0.0000 9:35–9:36 0.0026** _0.0001 _0.0116 _0.0021 _0.0001 9:36–9:37 0.0043*** _0.0000 _0.0419*** _0.0015 _0.0001* 9:37–9:38 0.0017*** _0.0000 _0.0023 _0.0007 _0.0000 9:38–9:39 0.0027*** _0.0000 _0.0003 _0.0009 _0.0000 9:39–9:40 0.0002 0.0000 0.0089 0.0156*** _0.0001 9:40–9:41 0.0020*** _0.0001 _0.0099 _0.0039 _0.0002* 9:41–9:42 0.0035*** _0.0000 _0.0029 _0.0029 _0.0002* 9:42–9:43 0.0008 0.0000 0.0019 0.0003 0.0000 9:43–9:44 0.0014 0.0000 0.0043 0.0040*** _0.0000 9:44–9:45 0.0001 0.0000 0.0051* _0.0013** _0.0000 9:45–9:46 0.0005 0.0001 0.0050 0.0007 0.0000 9:46–9:47 0.0009 0.0000 0.0004 0.0004 0.0000 9:47–9:48 0.0018* _0.0000 _0.0088* _0.0003 _0.0000 9:48–9:49 0.0006 0.0000 0.0122* _0.0006 _0.0000 9:49–9:50 0.0004 0.0001* _0.0061 _0.0002 _0.0000 9:50–9:51 0.0002 0.0001 0.0023 0.0007 0.0000 9:51–9:52 0.0004 0.0000 0.0013 0.0001 0.0000 9:52–9:53 0.0006 0.0000 0.0048 0.0012 0.0000 9:53–9:54 0.0006 0.0000 0.0001 0.0027* _0.0001* 9:54–9:55 0.0003 0.0000 0.0017 0.0022 0.0000 9:55–9:56 0.0010 0.0000 0.0007 0.0008 0.0000 9:56–9:57 0.0006 0.0001 0.0004 0.0006 0.0000 9:57–9:58 0.0005 0.0001 0.0008 0.0009 0.0000 9:58–9:59 0.0005 0.0000 0.0026* _0.0005 _0.0000 9:59–10:00 0.0002 0.0001* _0.0034* _0.0006 _0.0000

The dependent variable is the 1-min return during the ﬁrst 30-min after the market open. The independent variables include: (i) cloud cover (CC) and weather control variables, including wind speed (WIND), snowiness (Dsnow), raininess (Drain), and temperature (TEMP); and (ii) dummy variables for

Monday, Friday, January, December, and two days of the daylight saving time changes (one for the first Sunday in April and another for the last Sunday in October). Each weather variable is deseasonalized by subtracting its average value of each calendar week during the sample period from the weather observations at New York City. Panels A and B show the ordinary least squares regression results for the equally weighted and value-weighted indices, respectively. In Panel C, we perform time-series regressions on each stock and then calculate the cross-sectional averages of regression coefficients. Only the coefficients on the weather variables are reported to save space. ‘‘***_{”, ‘‘}**_{”, and ‘‘}*_{” denote significance at the 1%, 5%, and 10% levels, respectively.}

(10)

Table 5

Regressions of intraday trading variables

9:30–16:00 9:30–9:45 9:45–10:00 9:45–16:00 13:00–13:30 15:00–16:00

Panel A: order imbalance A.1. OISVOL CC 1.0308 9.8264*** _2.5062 _0.2643 _0.3697 _2.6384 WIND 0.0867 0.0582 0.2135 0.1085 0.2790 0.0581 Dsnow 0.5817 27.9887* 1.9126 2.4247 7.3977 7.8677 Drain 0.6149 9.5004 4.0033 1.9421 0.5873 0.3759 TEMP 0.0873 0.0295 0.0664 0.0934 0.0664 0.0900 Adj. R2_(%) _0.84 _0.61 _0.40 _0.04 _0.37 _0.55 N 21,629 2649 2649 18,980 2708 2708 A.2. OISNUM CC 1.5213 5.6041*** _2.8808 _0.9530 _0.0042 _0.8497 WIND 0.2523 0.1925 0.2372 0.2624 0.3580* _0.3651* Dsnow 27.7189 9.2940 47.1395 30.3050 5.9344 4.1798 Drain 0.5894 10.8757* 5.9128 0.6577 0.9503 0.2275 TEMP 0.0461 0.2154 0.1391 0.0231 0.0577 0.0067 Adj. R2_(%) _0.86 _0.25 _0.10 _0.82 _0.89 _0.51 N 21,629 2649 2649 18,980 2708 2708 Panel B: volatility B.1. RANGE CC 41.5129*** _101.8320*** _102.7749*** _31.0077*** _4.0105 _17.1529* WIND 4.7245* _6.4450* _7.9596** _3.9237* _0.7373 _4.0634* Dsnow 1.3105 63.2427 47.4596 6.1053 3.6136 17.3783 Drain 0.2284 41.4344 41.5998 5.4820 17.6588** 12.6192 TEMP 1.7883* _3.6724* _4.1818* _1.2527* _0.1550 _1.1310 Adj. R2(%) 61.46 2.50 3.87 56.98 0.44 1.36 N 21,629 2649 2649 18,980 2708 2708 B.2. RETSTD CC 20.7130*** _41.1196*** _37.1572*** _17.5029*** _3.9088 _8.5619 WIND 3.5749* _4.4862* _4.6719* _3.4459* _1.6966 _3.7978* Dsnow 9.3146 61.3954 24.5280 3.6640 16.2277 2.3708 Drain 6.3615 1.9146 11.0022 7.3212 4.6876 1.6365 TEMP 1.3901* _1.7319* _2.7816* _1.3381* _0.7297 _0.8965 Adj. R2(%) 35.21 1.55 3.06 31.93 1.68 1.61 N 21,629 2649 2649 18,980 2708 2708

Panel C: market depth

CC 16.9306*** _9.3644*** _14.3265*** _17.9674*** _18.8466*** _24.5570*** WIND 0.7610* _0.3603 _0.5970** _0.8161* _0.8898* _1.0608** Dsnow 1.8575 18.6680* 22.0430* 4.1905 15.5020 6.4137 Drain 13.8222* 0.9981 1.3128 15.7483* 32.2327** 23.5949** TEMP 0.0478 0.3551 0.4730* _0.0944 _0.3863 _0.1578 Adj. R2_(%) _4.27 _1.14 _1.17 _2.39 _0.81 _1.07 N 21,629 2649 2649 18,980 2708 2708 Panel D: spread D.1. DIF_ES CC 0.2035 0.0841 0.1430 0.2418 0.129 0.1661 WIND 0.0131 0.0520 0.0686 0.0076 0.0591 0.0271 Dsnow 1.1326 1.8003 3.1079 1.0465 0.8878 0.4743 Drain 0.3244 0.3203 1.5118 0.3301 0.7661 1.0230 TEMP 0.0155 0.0145 0.0066 0.0157 0.0180 0.0194 Adj. R2_(%) _0.17 _0.17 _0.26 _0.25 _0.25 _0.48 N 21,629 2649 2649 18,980 2708 2708 D.2. DIF_QS CC 0.1809 0.1870 0.0607 0.1624 0.1438 0.3636 WIND 0.0308 0.0551 0.0983* _0.0274 _0.0797 _0.0601 Dsnow 1.4142 5.9790* 3.9497 0.8949 0.6690 0.8103 Drain 0.7388 0.7445 1.3934 0.7395 0.9306 1.6900* TEMP 0.0158 0.0020 0.0111 0.0188 0.0108 0.0183 Adj. R2(%) 0.81 0.12 0.50 1.12 1.12 2.04 N 21,629 2649 2649 18,980 2708 2708

(11)

to save space.12We find that during the first 12-min after the market open, the impact of cloud cover on stock returns is, in general, significantly negative. This impact then diminishes quickly. The effects of other weather vari-ables on returns are generally insignificant. The results in

Table 4provide further support for our evidence inTable 3. Cloud cover has a significant influence on stock returns only during the first few minutes after the market open, and its impact diminishes gradually as more information comes to the market throughout the day.

4.2. The impact of cloud cover on trading variables

Table 5 shows the eﬀects of cloud cover and other weather variables on trading activities, where again only the coeﬃcients of weather variables are presented in order to save space. These are value-weighted regression results.

The results for the equally weighted index and individual stocks are qualitatively similar, so we do not report them here.

Panel A shows the impact of weather on order imbal-ance. In the ﬁrst 15-min interval, there are generally signif-icantly more seller-initiated trades than buyer-initiated trades when the sky is cloudier. In other trading periods (including the period of after lunch hours), however, order imbalance is not found to be signiﬁcantly correlated with cloud cover. This evidence is consistent with that presented in Table 3. As investors are in a worse mood around the market open, they tend to be pessimistic and less inclined to buy rather than to sell, resulting in lower stock returns during the opening interval.

The inﬂuences of cloud cover on return volatility and market depth are shown in Panels B and C, respectively. Panel B indicates that price range and return standard devi-ation both increase signiﬁcantly with cloud cover. Our results suggest that investor sentiment is negatively associ-ated with market volatility, consistent with the hypothesis that when investors are in a poor mood, there is more

Table 5 (continued)

9:30–16:00 9:30–9:45 9:45–10:00 9:45–16:00 13:00–13:30 15:00–16:00 Panel E: turnover E.1. TURNPER CC 0.3980 0.3704 0.4328 0.4213 0.6030 0.5962 WIND 0.0560* _0.0338 _0.0510 _0.0592* _0.0404 _0.0847* Dsnow 0.4928 0.0372 0.4525 0.5529 0.9886 0.7818 Drain 0.6421* 0.1500 0.2213 0.7658* 0.8592* 0.9205 TEMP 0.0269* _0.0027 _0.0044 _0.0325* _0.0188 _0.0314* Adj. R2_(%) _4.00 _1.58 _1.30 _2.04 _1.54 _1.41 N 21,629 2649 2649 18,980 2708 2708 E.2. TURN CC 5.7850 11.4418 14.2848 4.9393 0.7377 17.8101 WIND 1.4368 1.8492 2.6181* _1.3808 _0.2326 _1.9011* Dsnow 2.5802 17.7259 40.6775 7.2683 1.5141 34.3965 Drain 2.7337 14.6084 19.4639 3.5338 2.1745 10.7800 TEMP 0.0680 0.1842 0.5182 0.1025 0.0312 1.0211* Adj. R2(%) 49.94 4.21 8.44 53.48 5.66 3.22 N 21,629 2649 2649 18,980 2708 2708

The dependent variables are the value-weighted average trading variables. Two order imbalance ratios, one by trading volume (OISVOL) and one by number of trades (OISNUM), are defined. Order imbalance by trading volume is calculated as the trading volume of seller-initiated trades divided by the total trading volume. Order imbalance by number of trades is calculated as the number of seller-initiated trades divided by the total number of trades. Two volatility variables, price range (RANGE) and returns standard deviation (RETSTD), are examined. Price range is the interval high price minus the interval low price during the indicated interval. Return standard deviation is the square root of the sum of the squared tick return of the bid-ask midpoint. Market depth (DEPTH) is defined as sum of the bid size and ask size, in 100 shares. The first differences in effective spreads (DIF_ES) and quoted spreads (DIF_QS) are defined as the differences between the spread of the interval and that of the same interval of the previous trading day, to control for the well-known intraday seasonality and correlation in spreads. The percentage effective spread is defined as twice the absolute value of the difference between the trading price and the mid-point of the ask and the bid prices, scaled by the mid-point of the ask and the bid prices. The percentage quoted spread is the difference between the ask price and the bid price scaled by the mid-point of the ask and bid prices. Turnover per trade (TURNPER) is defined as the average trading volume per trade scaled by the outstanding shares at the end of last month. Cumulative turnover (TURN) is the total trading volume in the interval scaled by the outstanding shares at the end of the last month. The independent variables include: (i) cloud cover (CC) and weather control variables, including wind speed (WIND), snowiness (Dsnow), raininess (Drain), and temperature (TEMP); (ii) dummy variables for Monday, Friday,

January, December, and two days of the daylight saving time changes (one for the first Sunday in April and another for the last Sunday in October); and (iii) dummy variables indicating trading intervals (baseline comparison is to the first trading interval of each regression). A trading day is divided into two 15-min intervals after the market open, and six 60-min intervals for the following trading hours. Each weather variable is deseasonalized by subtracting its average value of each calendar week during the sample period from the weather observations at New York City. The coefficients on dummy variables indicating month, day of the week, daylight saving time changes, and trading intervals are suppressed to save space. All the regression coefficients other than those for DEPTH are multiplied by 1000. ‘‘***_{”, ‘‘}**_{”, and ‘‘}*_{”denote significance at the 1%, 5%, and 10% levels, respectively.}

12_{Our results are qualitatively similar if we use the two-, three-, or 5-min}

(12)

disagreement in opinion among them, and hence return volatility increases (Lee et al., 2002; Baker and Stein, 2004). Panel C shows that market depth drops significantly as cloud cover increases. This evidence supports the hypothesis that cloudy weather makes investors pessimistic and reduces their inclination to trade in the market (Loughran and Schultz, 2004; Goetzmann and Zhu, 2005). Panels B and C also show that cloud cover influences return volatility and market depth not just in the first two 15-min periods, but also over the entire trading day. Previous studies document that shocks in volatility are highly persistent (Engle, 1982; Bollerslev, 1986; Nelson, 1991; Bollerslev and Engle, 1993). Moreover, market depth is found to be negatively related to volatility in the litera-ture (Bessembinder and Seguin, 1993; Ahn et al., 2001; Goldstein and Kavajecz, 2004). Therefore, the effects of cloud cover on volatility and market depth last longer.

Panels D and E of Table 5 show the impact of cloud cover on spreads and turnover ratios, respectively. Neither variable is significantly related to cloud cover. When we combine these observations with the findings in Panels A, B, and C, clearly cloud cover has a more profound influ-ence on order imbalance, return volatility, and market depth, than on spreads and turnover ratios.

4.3. Robustness checks

To check the robustness of our results, we examine results for different partitions of intraday trading intervals. We split a trading day into seven intervals, a 30-min first interval followed by 6-h-long intervals. We also try thirteen equal-length intervals of 30-min each. Results for these alternative definitions of intraday intervals are similar, and our conclusions remain the same. We also use a 6-h average for weather observations before the beginning of an interval to define the weather variables, and the results are again very similar to those reported. Finally, to acknowledge the non-synchronous trading problem for illiquid stocks, we repeat all tests excluding 10% of the least liquid firms, and obtain similar findings.

5. Concluding remarks

Our examination of the impact of weather contributes to the literature by its focus on intraday return and trading activity for NYSE firms. We find that greater cloud cover induces a significantly negative intraday return only in the first 15-min period of the trading day. We also find more seller-initiated trades during the opening period when the skies are cloudier. Weather affects stock returns and order imbalance only around the market open, and it diminishes in importance quickly as more information arrives in the market throughout the day.

We further show that spreads and turnover ratios are not significantly related to cloud cover, but it has a signif-icantly positive effect on return volatility and a significantly negative effect on market depth. These effects occur not

only in the opening hours, but also over the entire trading day. Our findings overall suggest that weather influences intraday trading behavior because it affects investor mood.

Acknowledgements

The authors wish to thank Dosoung Choi, William Goe-tzmann, Frank C. Jen, Cheng-few Lee, Tim Loughran, and especially an anonymous referee for helpful comments and suggestions. Seminar participants at the 2007 FMA Annual Meeting provided many valuable comments. This paper was reviewed and accepted while Prof. Giorgio Szego was the Managing Editor of The Journal of Banking and Fi-nance and by the past Editorial Board. Yueh-Hsiang Lin gratefully acknowledges ﬁnancial support from the Na-tional Science Council in Taiwan (NSC96-2416-H147-012).

References

Ahn, H.J., Bae, K.H., Chan, K., 2001. Limit orders, depth, and volatility: evidence from the Stock Exchange of Hong Kong. Journal of Finance 56, 767–788.

Arkes, H., Herren, L.T., Isen, A.M., 1988. The role of potential loss in the inﬂuence of aﬀect on risk-taking behavior. Organizational Behavior and Human Decision Making Process 42, 181–193.

Atkins, A.B., Dyl, E.A., 1990. Price reversals, bid-ask spreads and market eﬃciency. Journal of Financial and Quantitative Analysis 25, 535–547. Bagozzi, R., Gopinath, M., Nyer, P., 1999. The role of emotions in

marketing. Journal of Academy of Marketing Science 27, 184–206. Baker, M., Stein, C.J., 2004. Market liquidity as a sentiment indicator.

Journal of Financial Markets 7, 271–299.

Barclay, M., Litzenberger, R.H., 1988. Announcement eﬀects of new equity issues and the use of intraday prices. Journal of Financial Economics 21, 71–100.

Bessembinder, H., Seguin, P.J., 1993. Price volatility, trading volume, and market depth: evidence from futures markets. Journal of Financial and Quantitative Analysis 28, 21–39.

Bollerslev, T., 1986. Generalized autoregressive conditional heteroscedas-ticity. Journal of Econometrics 31, 307–327.

Bollerslev, T., Engle, R.F., 1993. Common persistence in conditional variances. Econometrica 61, 167–186.

Brown, G.W., 1999. Volatility, sentiment, and noise traders. Financial Analysts Journal 55, 82–90.

Busse, J.A., Green, T.C., 2002. Market eﬃciency in real time. Journal of Financial Economics 65, 415–437.

Cao, M., Wei, J., 2005. Stock market returns: a temperature anomaly. Journal of Banking and Finance 29, 1559–1573.

Chordia, T., Roll, R., Subrahmanyam, A., 2001. Market liquidity and trading activity. Journal of Finance 56, 501–530.

Chordia, T., Roll, R., Subrahmanyam, A., 2002. Order imbalance, liquidity, and market returns. Journal of Financial Economics 65, 111–130. Chung, K., Van Ness, B.F., Van Ness, R.A., 1999. Limit orders and the

bid-ask spread. Journal of Financial Economics 53, 255–287. Cooke, L.J., Rose, M.S., Becker, W.J., 2000. Chinook winds and migraine

headache. Neurology 54, 302–307.

Cross, F., 1973. The behavior of stock prices on Fridays and Mondays. Financial Analysts Journal 29, 67–69.

Cunningham, M.R., 1979. Weather, mood and helping behavior: quasi-experiments with the sunshine Samaritan. Journal of Personality and Social Psychology 37, 1947–1956.

Engle, R.F., 1982. Autoregressive conditional heteroscedasticity with estimates of the variance of UK inﬂation. Econometrica 50, 987–1008. Fabozzi, F.J., Ma, C.K., Chittenden, W.T., Pace, R.D., 1995. Predicting intraday price reversals. Journal of Portfolio Management 21, 42–53.

(13)

Foster, F.D., Viswanathan, S., 1993. Variations in trading volume, return volatility, and trading costs: evidence on recent price formation models. Journal of Finance 48, 187–211.

Gervais, S., Odean, T., 2001. Learning to be overconﬁdent. Review of Financial Studies 14, 1–27.

Goetzmann, W.N., Zhu, N., 2005. Rain or shine: where is the weather eﬀect? European Financial Management 11, 559–578.

Goldstein, M.A., Kavajecz, K.A., 2004. Trading strategies during circuit breakers and extreme market movements. Journal of Financial Markets 7, 301–333.

Harris, L., 1986. A transaction data study of weekly and intradaily patterns in stock returns. Journal of Financial Economics 16, 99–117. Harris, M., Raviv, A., 1993. Diﬀerences of opinion make a horse race.

Review of Financial Studies 6, 473–506.

Hirshleifer, D., 2001. Investor psychology and asset pricing. Journal of Finance 56, 1533–1597.

Hirshleifer, D., Shumway, T., 2003. Good day sunshine: stock returns and the weather. Journal of Finance 58, 1009–1032.

Howarth, E., Hoﬀman, M.S., 1984. A multidimensional approach to the relationship between mood and weather. British Journal of Psychology 75, 15–23.

Jain, P.C., Joh, G.H., 1988. The dependence between hourly prices and trading volume. Journal of Financial and Quantitative Analysis 23, 269–283.

Kamstra, M.J., Kramer, L.A., Levi, M.D., 2000. Losing sleep at the market: the daylight saving anomaly. American Economic Review 90, 1005–1011.

Kamstra, M.J., Kramer, L.A., Levi, M.D., 2003. Winter blues: a sad stock market cycle. American Economic Review 93, 324–343.

Karpoﬀ, J., 1986. A theory of trading volume. Journal of Finance 41, 1069–1088.

Keim, D.B., Stambaugh, R.F., 1984. A further investigation of the weekend eﬀect in stock returns. Journal of Finance 39, 819–837. Lee, M.C., Ready, M., 1991. Inferring trade direction from intraday data.

Journal of Finance 46, 733–748.

Lee, W.Y., Jiang, C.X., Indro, D.C., 2002. Stock market volatility, excess returns, and the role of investor sentiment. Journal of Banking and Finance 26, 2277–2299.

Lo, A.W., Repin, D.V., 2002. The psychophysiology of real-time ﬁnancial risk processing. Journal of Cognitive Neuroscience 14, 323–339. Loughran, T., Schultz, P., 2004. Weather, stock returns, and the impact of

localized trading behavior. Journal of Financial and Quantitative Analysis 39, 343–364.

Maberly, E.D., Maris, B.A., 1991. The January eﬀect, arbitrage oppor-tunities and derivative securities: has anything changed? Journal of Futures Markets 11, 253–257.

McInish, T., Wood, R.A., 1992. An analysis of intraday patterns in bid/ ask spreads for NYSE stocks. Journal of Finance 47, 753–764. Mehra, R., Sah, R., 2002. Mood, projection bias and equity market

volatility. Journal of Economic Dynamics and Control 26, 869– 887.

Nelson, D.B., 1991. Conditional heteroscedasticity in asset returns: a new approach. Econometrica 59, 347–370.

Nofsinger, J.R., 2005. Social mood and ﬁnancial economics. Journal of Behavioral Finance 6, 144–160.

Persinger, M.S., 1975. Lag responses in mood reports to changes in the weather matrix. International Journal of Biometeorology 19, 108– 114.

Pinegar, J.M., 2002. Losing sleep at the market: comment. American Economic Review 92, 1251–1256.

Saunders, E.M., 1993. Stock prices and Wall Street weather. American Economic Review 83, 1337–1345.

Shalen, C., 1993. Volume, volatility, and the dispersion of beliefs. Review of Financial Studies 6, 405–434.

Shiller, R.J., 2003. From eﬃcient market theory to behavioral ﬁnance. Journal of Economic Perspectives 17, 83–104.

Smirlock, M., Starks, L., 1986. Day-of-the-week and intraday eﬀects in stock returns. Journal of Financial Economics 17, 197–210.

Statman, M., Thorley, S., Vorkink, K., 2006. Investor overconﬁdence and trading volume. Review of Financial Studies 19, 1531–1565. Stoll, H., Whaley, R., 1990. Stock market structure and volatility. Review

of Financial Studies 3, 37–71.

Szakmary, A.C., Kiefer, D.B., 2004. The disappearing January/turn of the year eﬀect: evidence from stock index futures and cash markets. Journal of Futures Markets 24, 755–784.

Thaler, R.H., 1987a. Anomalies: the January eﬀect. Journal of Economic Perspectives 1, 197–201.

Thaler, R.H., 1987b. Anomalies: seasonal movements in security prices II: weekend, holiday, turn of the month, and intraday eﬀects. Journal of Economic Perspectives 1, 169–177.

Wood, R.A., McInish, T.H., Ord, J.K., 1985. An investigation of transaction data for NYSE stocks. Journal of Finance 40, 723–740.

Wright, W.F., Bower, G.H., 1992. Mood eﬀects on subjective probability assessment. Organizational Behavior and Human Decision Process 52, 276–291.