There is always a potential for errors when conducting empirical studies. It is important to be aware of these and minimize this potential as they may affect the reliability of the results of the study. Reliability
38
refers to the consistency of the study and to the fact that any significant result must be repeatable.
Validity refers to the fact that the results obtained meet all of the requirements of the scientific research method. Reliability is a prerequisite for validity, but it is no guarantee for it.
There are two main types of errors that are particularly relevant in a quantitative study like the one conducted in this thesis; errors in data and errors in models.
3.2.1 Errors in data
Errors in data refer to the collection of data; the type of data used and the how this data is obtained.
By ensuring that the data shows what it is intended to, we ensure high reliability and validity. The data used in this thesis is collected from a variety of sources, both primary and secondary. Data from primary sources is information received directly from the original source, like a company’s website, press releases, prospectus, etc. Information from companies must be interpreted with care as it may be somewhat distorted in order to depict the company as being better than it is. However, most of the information gathered for this thesis is numbers, and as financial reporting and prospectuses are subject to laws and regulations regarding the correctness of information we consider reliability to be high.
Data from secondary sources include information from databases such as DataStream, websites of stock exchanges and other financial websites. Gathering data from secondary sources gives low control over reliability, but by using highly reputable financial sites and well-known databases we consider the reliability of information from secondary sources to be high.
The list of IPOs conducted in the period of interest was downloaded from the respective stock exchanges. The selection of IPOs in the final dataset was done manually and is thereby subject to potential human errors. Manual selection of IPOs may lead to wrong IPOs being kept or the exclusion of IPOs that fit the criteria to be kept in the sample. To reduce the potential for human errors, the observations have been reviewed by both authors and information has been controlled through several sources.
After selecting IPOs to be included in the sample, data for each IPO was registered manually in Excel.
This further opens up for potential human errors. When detecting large outliers, the information was
39
checked again to ensure it was correct, in order to reduce the potential of errors. Based on these arguments, we consider the reliability of our dataset to be high.
The time period of the analysis may affect reliability. A longer time period would give us more observations, but then IPOs would be older, and data may not be as relevant anymore. Financial markets are rapidly changing, thus older IPOs may not be representative for today’s market conditions.
A shorter time period would decrease the number of observations, only hurting our reliability. As the sample size is relatively large (89 observations), and observations includes both hot issue and cold issue periods, the time period is considered to provide high reliability and consistency in results.
Validity of the dataset is considered high. There is little room for subjective interpretations as most of the information is numbers. Manual exclusion of IPOs will not reduce validity because the reasons can be explained and can also be found in earlier research done on underpricing. No IPOs have been excluded due to lack of data other than unavailability to prospectuses, thus ensuring validity of the dataset.
3.2.2 Errors in models
When creating a multiple regression model, there is always potential for specification errors. As mentioned earlier the field of underpricing is vast and widespread, thus creating a potential that our model omits some variables that can help explain underpricing. If the model suffers misspecification, the results may be biased and conclusions drawn from them may be incorrect. Models explaining underpricing differ in the factors used to explain the phenomenon. Being restrictive in the number of explanatory variables will decrease reliability, but lack of time and availability to information
necessitates restrictions. Our model builds on previous research and we have included the variables we consider to be most central for explaining underpricing in Scandinavian markets, which is why we consider our model to have high reliability.
The measure of underpricing in our model is very short term. We measure underpricing as the
difference between the closing price at the first day of trading and the offer price. Our model builds on
40
previous research, increasing reliability of the dependent variable and consequently increasing the reliability of our model.
The validity of the model refers to whether or not the study is well designed and provides results that can be generalized to the population in interest. Validity is considered in terms of internal and external validity.
Internal validity refers to whether there exists any cause, or causes, other than the one we have uncovered that also can explain the results. Due to the risk of specification error in our model, we cannot assure that high internal validity is upheld. However, to the best of our knowledge internal validity is considered relatively high as the regressions have explanatory power for underpricing similar to levels obtained in previous studies. In addition, the model includes variables obtainable prior to the listing, which is easily obtainable for retail investors. Due to the fact that previous research has not been able to come up with a unanimous explanation for the phenomenon of underpricing, validity is very hard to achieve.
External validity refers to generalization of the model. Different research has found different explanatory factors, all of which seems specific to market, time period, etc. The dataset as well as codes used for different statistical programs is included in the thesis, which increases the transparency of the study and thereby the external validity. The reader will be able to use other datasets containing the same variables used in the regression models in the thesis and conduct the same analyses that are presented here. The reader will also be able to replicate the results from the regression models by using the same dataset presented here. It is worth noting that the bootstrap method used is
resampling with replacement, which results in unique datasets each time the bootstrap is run. It will thereby not be possible to exactly replicate the estimated numbers from the bootstrap model, but the results and significance levels will be similar. As our study is limited to the Scandinavian markets, it cannot be expected to be generalized to other markets as there will be other characteristics of firms and markets that affect the degree of underpricing. However, we expect that similar results can be found if the study is conducted on the population of interest, i.e. IPOs on the Scandinavian market.
41