Abstract
The US-capital market is the largest in the world. US-American IPOs offer an excellent data sample. On the one hand there exists an extensive amount of IPOs with a large database and the accounting items are usually available over a long time period. Furthermore, US companies typically offer incentives during IPOs by share-based payments which are of further interest in the empirical study for the years after the issue. Therefore, the data of public companies with their primary exchange in the US are included.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
NYSE Euronext (2012).
- 2.
Accruals from working capital data can be collected from 1962 onwards, because COMPUSTAT data prior to 1962 are generally acknowledged to suffer from survivorship bias, see Kothari et al. (1995). Data with missing values are typically eliminated.
- 3.
SFAS no. 95 obligates firms to publish a statement of cash flows for fiscal years ending after July 15, 1988. For some early adopters the report is available from 1987 onwards.
- 4.
Source: CRSP®, Center for Research in Security Prices. Graduate School of Business, University of Chicago. Used with permission. All rights reserved. www.crsp.uchicago.edu. The author acknowledges the data from COMPUSTAT, which were obtained through Wharton Research Data Services (WRDS) and sponsored by University Research Priority Program Finance. The usage of these databases is very common, for example, Cecchini et al. (2012), Ising (2009). The data format in COMPUSTAT is used with flagged STD. STD indicates that the data are the originally reported and not restated figures. If available, PRE_AMENDS are used. These are data prior making amendments to financial statements following a possible SEC investigation. The number of PRE_AMENDS data is rare. The sample is limited to domestic firms (COMPUSTAT mnemonics: popsrc = D and fic = USA) with the domestic accounting standard (DS) and primary issue tag in the US (COMPUSTAT mnemonic: priusa = 1).
- 5.
Pre-IPO data is enclosed in the offering prospectus and is commonly available for three years prior to the IPO where as it can also be hand collected from 10-K forms. See also Armstrong et al. (2009, 20) and Katz (2006, 6).
- 6.
Gunny (2010), p. 877. For a discussion of trimming data instead of winsorizing see Kothari et al. (2005b).
- 7.
For clarity reasons of the thesis, five industries are sufficient. As in prior literature the regulated industries (SIC 4910-4939) and financial institutions (SIC 6000-6999) are omitted as comparable industries due to their special business model.
- 8.
This procedure fosters comparability to prior literature. A main notion of this thesis is to show differences in industries. Therefore, a mixed overall industry sample is not presented.
- 9.
Joos/Zhdanov (2008), p. 446. Similarly, Zucker et al. (1998), pp. 290-291.
- 10.
Joos/Zhdanov (2008), p. 431.
- 11.
Bartov et al. (2002), p. 322.
- 12.
Francis et al. (2012), pp. 261-262; Hsu (2009), p. 265.
- 13.
The calculations were repeated for the first year before the IPO without companies from SIC code 5122 and the results are almost exactly the same.
- 14.
The overall number of 3’168 IPOs includes all IPOs in these years with available data, but the number can be reduced for calculations if, for example, not enough comparable companies in the same SIC and year are available.
- 15.
The tables for Pearson include merely the post-IPO year (year +1) due to space constraints. This year seems most suitable since it includes enough observations and important incentives around the IPO. Including the subsamples would induce opacity in presentation and is left to future research. A Spearman correlation matrix is typically used on ordinal outcomes.
- 16.
Kothari et al. (2005a). The other employed models in the robustness section are the models of Jones, Dechow/Dichev, McNichols, and Healy.
- 17.
This is similar for other accounting items which are also calculated with various models. These models are included as sensitivity checks in the robustness section.
- 18.
Results are presented with adj. R2 (adjusted R-squared) because not the whole population can be included due to a lack of data in COMPUSTAT.
- 19.
It is important to note for the accounting items that the interpretation concerns the discretionary part of each item since the models test the significance of the abnormal discretionary parts and not the whole amount of the items themselves. However, the discretionary part is included in the sum of e.g. accruals of each company (regular accruals plus discretionary accruals equal the reported accruals) and this affects the overall amount.
- 20.
Singer (2007), p. 58. Fedyk et al. (2012) do not evaluate the two subgroups within one industry, but only the overall sample which is insignificant. The outcomes of the overall groups are not exactly comparable because of the slightly different research design and the authors use 30 percent fewer observations.
- 21.
Note the formulation of results from statistical outcomes in the thesis: On the one hand, results can reject the null hypothesis of no discretionary behavior and hence the findings are “consistent” with the (alternative) hypothesis that discretion exists. The alternative is that results do not reject the (null) hypothesis of no discretion which means that there is no evidence of discretionary behavior.
- 22.
Caylor (2010), p. 93.
- 23.
Fedyk et al. (2012), p. 48. Furthermore, the results are not exactly comparable because they use another revenue model as well as other time intervals and have fewer observations. In the prior version of Singer (2007, 60) the results for the profitable group are the same and the loss group shows negative significance. The comparability issues apply here as well, except concerning the particular model.
- 24.
In the first receivables model, the loss subgroup is significant at the 11% level and the overall sample of the second receivables model is significant at the 12% level.
- 25.
The test for this hypothesis is one-sided.
- 26.
Tests for this hypothesis are one-sided. Results are equally to two sided tests in terms of overall significance, yet stronger.
- 27.
The abnormal R&D value for unprofitable and all firms is even significant at the 0.02% level.
- 28.
The correct term is that due to the results “the null hypothesis of H-R&D 1 is not rejected”, but for reasons of easier and better understating, sometimes the notation “not consistent with the alternative hypothesis” is used in this study.
- 29.
Francis et al. (2012), pp. 261-262; Hsu (2009), pp. 264-265; Sievers/Klobucnik (2012), pp. 1-3.
- 30.
Gunny (2010), pp. 858-859. However, she does not differentiate between industries and notes that SG&A will be reduced in all companies.
- 31.
There are no statistical procedures employed to replace missing data.
- 32.
Teoh et al. (1998c), pp. 175-176.
- 33.
Wongsunwai (2013).
- 34.
Teoh et al. (1998c), pp. 175-176, for example.
- 35.
The cash flow is negatively influenced by other items such as R&D and SG&A, too.
- 36.
Comiskey (1972); Gordon (1972).
- 37.
It is important to note that the rejection or non-rejection of hypotheses is decided as follows: For example, if in the Biotech industry the overall group is significant compared to the established industries, then the null hypotheses is rejected. However, too few observations in the profitable industries lead to no result and the null hypothesis cannot be rejected although growth industries may have 100% significant results. Like in hypothesis set one, this is a conservative approach within this study.
- 38.
The differentiation of industries into growth and profitability is dependent on the years around the IPO and is not the designation for the industry in general.
- 39.
Lévesque et al. (2012); Fedyk et al. (2012).
- 40.
Gunny (2010), p. 857.
- 41.
Lévesque et al. (2012), pp. 47-48.
- 42.
Graham et al. (2005), p. 32.
- 43.
See also Joos/Plesko (2005).
- 44.
Katz (2006); Morsfield/Tan (2006); Wongsunwai (2013), for example.
- 45.
Lévesque et al. (2012), for example.
- 46.
Gunny (2010), for example.
- 47.
Both concentrate on the overall group. Caylor (2010, 83) examines earnings surprises.
- 48.
Wooldridge (2002), pp. 46-57. All presented tests are used selectively on the regressions. There are 13 regression models for the main accounting items which are applied to each industry and year combination. This results in an enormous amount of regressions. Therefore, selective application is appropriate.
- 49.
For example, see Gunny (2010, 867) for accruals and Wooldridge (2002, 300-301) for the general idea. Additionally, the discretionary amounts are winsorized by 1% at the top and bottom.
- 50.
UCLA (2013). Note that the term “residual” does not comply with the final discretionary amounts used in this study due to the matching procedure.
- 51.
Wooldridge (2002), pp. 94-95.
- 52.
Gujarati (2004), pp. 341-349.
- 53.
This is also true for the McNichols model, although it includes several variables that might have been expected to correlate.
- 54.
See Wooldridge (2002, 281-283) for the regression specification-error test (RESET) of Ramsey (1969). In STATA the commands “ovtest” and “linktest” can be used.
- 55.
The subscripts are “i” for the company and “t” for the time period.
- 56.
If ROA is included as an additional regressor in the Jones and modified Jones models, the specification improves only marginally, see Kothari et al. (2005a, 186).
- 57.
The model uses the change in reported revenues rather than the change in cash revenues (ΔSales- ΔRec) to avoid biased estimates of discretion for growth firms; see also Stubben (2006), p. 34.
- 58.
Roychowdhury (2006), p. 352; Stubben (2006), p. 14; Stubben (2010), p. 700. The model uses the change in reported revenues instead of the change in cash revenues to avoid biased estimates of discretion for growth firms; see Stubben (2006), p. 34.
- 59.
Using past revenue instead of future revenue means that R&D is used according to present information instead of information that is not available to managers when deciding about these expenses. Furthermore, there is no possibility to use Tobin’s Q in pre-IPO years as one model in Fedyk et al. (2012, 40). The corresponding authors mention no diverging results of the model with sales growth instead of Tobin’s Q.
- 60.
Note that “change in sales” is deflated by total assets while “sales growth” is deflated by sales.
- 61.
Roychowdhury (2006), p. 345.
- 62.
The results are still biased if the aforementioned hypotheses hold and sales are managed in several years around the issue.
- 63.
Singer (2007), p. 24.
- 64.
Anderson et al. (2003), pp. 48-49.
- 65.
For an exemption see Cohen et al. (2009, 1-2).
- 66.
If R&D and advertising are non-existent in COMPUSTAT, both are presumed as zero.
- 67.
Dechow et al. (1998); Roychowdhury (2006), p. 345.
- 68.
Since both approaches of mandatory SG&A as well as mandatory R&D, SG&A, and advertising are used, the first model is also used as model three but with a different data basis. Equally, the second and fourth models only differ by the approach of data prerequisites.
- 69.
Roychowdhury (2006), p. 345. See also Dechow et al. (1998), but differently in Aaker/Gjesdal (2010, 11).
- 70.
Dechow/Dichev (2002), pp. 40-41; Pae (2005), p. 9.
- 71.
Aaker/Gjesdal (2010), pp. 8-10. Although it outperforms other models in this category it is only used in the robustness section due to a loss of observations when using lagged CfO.
- 72.
Roychowdhury (2006), p. 346.
- 73.
For the sake of clarity numerical tables are omitted.
- 74.
It is important to note that the interpretation concerns the abnormal part of accounting items since the models estimate results for the significance of the discretionary parts. However, the discretionary part is also included in the sum of the actual amount (i.e., regular accruals plus discretionary accruals equal the reported accruals) and this affects the overall amount.
- 75.
Singer (2007), p. 58. Fedyk et al. (2012) do not evaluate the two subgroups within one industry, but only within the overall sample which is insignificant. The outcomes of the overall groups are not exactly comparable because of the slightly different research design and Fedyk et al. (2012) use 30% fewer observations. The different design includes, for example, average total assets instead of total assets as a deflator which decreases amounts.
- 76.
The Jones and modified Jones models display positive significance at the 17% level for loss firms, respectively.
- 77.
Since the accounts receivable models are mainly the same throughout the study, they are only reported separately if they diverge.
- 78.
Fedyk et al. (2012), p. 48.
- 79.
In the first receivables model the loss subgroup is significant at the 11% level and the overall sample of the second receivables model is significant at the 12% level.
- 80.
This results from the control variable of sales growth from two years.
- 81.
Discretionary expenses are calculated if R&D, SG&A, and advertising data are available.
Author information
Authors and Affiliations
Rights and permissions
Copyright information
© 2014 Springer Fachmedien Wiesbaden
About this chapter
Cite this chapter
Ising, P. (2014). Empirical Study. In: Earnings Accruals and Real Activities Management around Initial Public Offerings. Springer Gabler, Wiesbaden. https://doi.org/10.1007/978-3-658-03794-9_6
Download citation
DOI: https://doi.org/10.1007/978-3-658-03794-9_6
Publisher Name: Springer Gabler, Wiesbaden
Print ISBN: 978-3-658-03793-2
Online ISBN: 978-3-658-03794-9
eBook Packages: Business and EconomicsEconomics and Finance (R0)