3. SENTIMENT ANALYSIS MODEL
3.4 REGRESSION MODEL
construct another standard time window, one month before the issue day. As shown in Table 3-3, therefore, this study forms the different models to test news data between the two different periods. This study also tests our hypothesis under different industries.
At the industry level, the importance of the electronic products sector, which has become the main driver of the Taiwanese economy for a long time is also a key characteristics of IPO firms. In Taiwan, This study thinks the electronics industry has the overwhelming advantage on the cluster effect of media. If one of the electronics supply chain bloom up, the rest of the supply chain would benefit from the trend. Hence, this study also researchs on that the positive sentiment of media coverage about the companies in the electronics industry positively relates to underpricing than those in non-electronics industry.
TABLE 3-3 FRAMEWORK OF THE MODEL
Sample The Pre-IPO Period One Month Prior to The Issue Day
Full Sample
H1-Model (1)、
H1-Model (3)
H1-Model (2) 、H1-Model (4)、
H2-Model (5)
Electronics Industry H1-Model (1) H1-Model (2)、H2-Model (5) Non-Electronics
Industry
H1-Model (1) H1-Model (2)、H2-Model (5)
‧
Equation (1) and (2) is to test hypotheses (H1) which investigate the association between underpricing and the sentiment of media before IPO issue date, and this study uses control variable LN_FIRMSIZE in Equation (1) and (3), but substitute the LN_FIRMSIZE for LN_OFFERSIZE in Equation (2) and (4). There are two reasons to explain why this study uses different model to test H1. First, the LN_FIRMSIZE and LN_OFFERSIZE have a highly positive association, and it is more reasonable to use LN_OFFERSIZE control variable when test the samples of one month prior to the issue day based on the timing of the pre-IPO period. Second, when this study substitutes the LN_FIRMSIZE for LN_OFFERSIZE in Equation (2), the adjusted R-square is higher.
With earlier literature suggesting the positive relationship between media coverage and underpricing is stronger when ex ante uncertainty is greater, and fail to
(2)
‧
whether the IPO underpricing reflects the public media sentiment, one approach is using the independent variable of TONE_RATIO, the other is using the independent variable of TONE. There independent variables TONE_RATIO and TONE are using based approach to measure sentiment extent of the media. This dictionary-based approach may lead to a result that this study doesn’t view each news item as the same weight to measure the sentiment extent about the target firm, however when scoring by calculating the frequency of the words means weighted average method.The difference these two independent variable is TONE_RATIO scaled by the number of the dictionary words to standardize the sentiment score. When scaled by the number of the dictionary words, the weighted average is more apparent. Thus TONE_RATIO emphasizes the different weight score of each news items more than TONE.
𝐈𝐑 = 𝛃𝟎+ 𝛃1𝐍𝐔𝐌_𝐏𝐎𝐒 + 𝛽2𝑵𝑼𝑴_𝑵𝑬𝑮 + 𝛽3𝑰𝑷𝑶_𝑷𝑬𝑹𝑰𝑶𝑫
+ 𝛽4𝑳𝑵_𝑶𝑭𝑭𝑬𝑹𝑺𝑰𝒁𝑬 + +𝛽5𝑺𝑪𝑨𝑳𝑬𝑫_𝑶𝑭𝑭𝑬𝑹 + 𝛽6𝑴𝑲𝑻𝑹𝑬𝑻 + 𝜀 .
Although Equation (5) is to test the hypotheses (H2) and also to find the association between underpricing and the number of positive (negative) news before IPO issue date, but the number of positive or negative news means the media coverage and media visibility of the target firm. There independent variables NUM_POS and NUM_NEG are using dictionary-based approach to classify the news mentioned in section 3.3.1, and there is no concept of weighted average method.
(5)
‧
TONE_RATIO = the score of the frequency count of positive words divided by the positive sentiment dictionary minus the ratio of the frequency count of negative words divided by the negative sentiment dictionary.
TONE = the score calculated as (Positive word-Negative word) divided by (Positive word+ Negative word).
TLHIT = the total words count excluding symbol words in each news item using computer program.
NUM_POS = the number of positive news item about the firm i published during the period from the one month prior to the issue day.
NUM_NEG = the number of negative news item about the firm i published during the period from the one month prior to the issue day.
LN_FIRMSIZE = the natural logarithm of the market value of the IPO firm at the offering and is noted in millions of New Taiwan dollars.
LN_OFFERSIZE = the natural logarithm of the IPO issue size, measured as offer price multiplied by the number of shares offered.
IPO_PERIOD = the natural logarithm of difference in the number of days between the date of listing on Emerging Stock Market and the issue date.
SCALED_OFFER = the dummy variable equals one if revision is greater than one, and zero otherwise.
MKTRET = the market return for 30 trading days prior to issue day, in percent.
This study uses the market data called from TEJ database
3.4.1 DEPENDENT VARIABLE
This study begins our analysis on IPO underpricing, also called IPO firms’ initial returns. Because there is no lock-up period for an IPO in Taiwan for our sample period.
Hence, this study takes the closing price of issue date as the first day market price of
‧
IPO shares. Therefore, firm i’s initial return, IRi, defined as 𝐼𝑅𝑖 =𝐶𝑃𝑖 − 𝑂𝑃𝑖
𝑂𝑃𝑖
Where CPi is firm i’s closing price of the issue day. OPi is the offering price.
3.4.2 INDEPENDENT VARIABLES
TONE_RATIO is the sentimental score that the ratio of the frequency count of positive words divided by the positive sentiment dictionary minus the ratio of the frequency count of negative words divided by the negative sentiment dictionary. Where firm i’s TONE_RATIO is defined as following:
𝑇𝑂𝑁𝐸_𝑅𝐴𝑇𝐼𝑂𝑖𝑗 = 𝑃𝑊𝑖𝑗
𝑃𝑂𝑆_𝐷𝐼𝐶− 𝑁𝑊𝑖𝑗 𝑁𝐸𝐺_𝐷𝐼𝐶
Where PWij (NWij) is the frequency count of positive (negative) words in the news item j of the firm i.
POS_DIC (NEG_DIC) is total the number of positive (negative) sentiment dictionary mentioned in Section 3.3.1 and shown in Appendix 2 (3). TONE_RATIO indicates positive tone of every news item, TONE_RATIO means the frequency difference between the positive and negative word scaled by the sentiment dictionary.
TONE is the sentimental score based on a frequency count of the number of positive and negative words, using the dictionary of this research. TONE is calculated as the count of positive words minus the count of negative words, divided by the sum of positive and negative word counts so the maximum and minimum values of TONE are 1 and –1, respectively.
𝑇𝑂𝑁𝐸𝑖𝑗 = 𝑃𝑊𝑖𝑗 − 𝑁𝑊𝑖𝑗 𝑃𝑊𝑖𝑗 + 𝑁𝑊𝑖𝑗
Where PWij (NWij) is the frequency count of positive (negative) words in the news item j of the firm i.
NUM_POS is the number of positive news item about the firm i published during the period from the one month prior to the issue day.
‧
NUM_NEG is the number of negative news item about the firm i published during the period from the one month prior to the issue day.
3.4.3 CONTROL VARIABLES
This study uses some variables to measure of firm- and offer-specific characteristics, as follows:
TLHIT is the total words count excluding symbol words in each news item using computer program. TLHIT is included in order to help evaluate the length of each financial news article. This variable is used to control for length of each news item.
LN_FIRMSIZE is the natural logarithm of the market value of the IPO firm at the offering and is noted in millions of New Taiwan dollars. It has been shown that larger IPOs show less underpricing (Beatty and Ritter 1986), because larger IPOs have lower uncertainty.
LN_OFFERSIZE is the natural logarithm of the IPO issue size in thousands of New Taiwan dollars, measured as offer price multiplied by the number of shares offered.
And the empirical analysis indicates that smaller offerings, have substantially higher average initial return. (Beatty and Ritter 1986). Hanley (1993) hypothesizes that offer size is inversely related to the change in offer price. Lowry and Schwert (2004) argue that investment bankers set the initial price range of riskier issues lower to minimize the chance of the issue being worth less than its projected value. In conjunction with notion that larger IPOs are easier to value (Prezas et al. 2000). Lowry and Schwert’s (2004) argument implies a negative correlation between issue size and price update, and initial return.
IPO_PERIOD is the natural logarithm of difference in the number of days between the date of listing on Emerging Stock Market and the issue date.
SCALED_OFFER equals one if equals one if the offer price is set above the average of initial filing price range, and zero otherwise. SCALED_OFFER variable,
‧
available through the road show for an IPO, underwriters tend to revise the offer price.When SCALED_OFFER equals one, it reflects strong demand for the issues are associated with greater underpricing. Liu et al. (2009) use price revision as a good proxy for the investor demand reported through the book building process. Liu et.al (2009) define investors’ demand is quiet relatively to the offer price. With the revelation of private information, when the investors’ demand is high, the offer price is revised upwards from the midpoint of the initial filing range. Whereas, when demand is low, the offer price is revised downwards. This is first pointed out by Hanley (1993) who noted Benveniste and Spindt’s (1989) model of book building predicts partial adjustment to private information.
MKTRET is the market return for 30 trading days prior to issue day, in percent.
This variable is used to control for market conditions at the time of IPO. This study uses the market data called from TEJ database. Prior research has shown that hot markets are associated with higher underpricing than cold markets (Ritter 1984). Logue (1973) first examines whether past market returns can predict future underpricing and finds a positive relationship between pre-IPO market return and IPO first-day return.