臺大管理論叢 NTU Management Review VOL.30 NO.3

53 NTU Management Review Vol. 30 No. 3 Dec. 2020 Department. I do not consider seasonal variation in the length of days and seasonal affective disorders as in, for example, Kamstra, Kramer, and Levi (2003), because Thailand is a tropical country. It does not have winter or autumn seasons. Air pressure affects moods because low air pressure leads to a decrease in the partial pressure of oxygen in the blood and body tissues (Sharp and Bernaudin, 2004). Wurtman and Wurtman (1989) explain the effects of cloud cover and rainfall on investors’ moods. The lack of sunlight associated with rainy days causes serotonin levels to fall, which lead to bad moods. The explanation is extended to the ground-visibility effect. Visibility degradation derived from poor sunlight can cause falling serotonin levels and bad moods. Relative humidity is one of the factors influencing the production of small negative ions. Inhaling air containing a large number of these ions reduces the symptoms of investors’ affective disorders (Salib and Sharp, 2002). Temperature affects mood states because heat induces changes in plasma concentrations of neurotransmitters and hormones (McMorris, Swain, Smith, Corbett, Delves, Sale, Harris, and Potter, 2006). Finally, Charry and Hawkinshire (1981) reported that ions created by wind might prompt changes in mood states. I retrieve the SET index data from the SET database and the weather variables from the Thai Meteorological Department database. The stock-return data begins on February 17, 1992, and ends on December 29, 2017 (6,335 trading-day observations), while the weather data began on January 1, 1991, and ended on December 31, 2017 (9,862 calendar- day observations). Following Hirshleifer and Shumway (2003), I calculate the daily weather variables by their average levels from 6.00 to 16.00. I remove seasonality from the weather variables by their averages for each week of the year over the 1991-2017 sample period. Next, the deseasonalized variables are standardized by their averages and standard deviations. Because some observations are missing, I impute the missing cases with zero because zero is the unconditional mean of deseasonalized variables. Table 1 reports the descriptive statistics of the stock return, variance, and weather variables. In Panel 1.1, the Jarque-Bera tests reject the normality hypothesis for all the variables. Because the average annual sample is large at 243.65 days, nonnormality should not affect the statistical inference. The augmented Dickey-Fuller statistics are significant at the 99% confidence level. All the variables are stationary. Therefore, the VAR model is appropriate. Panel 1.2 reports the correlation coefficients of the weather-variable pairs. All correlations, except those for air pressure-ground visibility and air pressure-rainfall pairs, are significant.