Long-term operating performance

To accommodate some of the issues, we applied the Wilcoxon signed-rank to test if the median abnormal performance is equal to zero following the approach in previous studies of Sudarsanam and Qian (2007) and Prezas and Simonyan (2015). First, the test assumes that all values are differ-ent from zero, and second, that no values are equal. The advantage of the Wilcoxon signed-rank test is that it accounts for both sign and magnitude. The differences in returns between sample firms and the applied index are converted to absolute values and ranked from highest to lowest. Then:

𝑾𝑾_𝑻𝑻 =� 𝑽𝑽𝑻𝑻𝒏𝒏𝒓𝒓(𝑬𝑬_{𝒊𝒊,𝑻𝑻})⁺

𝑵𝑵

𝒊𝒊=𝟏𝟏

Where 𝑾𝑾_𝑻𝑻 is the sum of the positive ranks of the absolute ranked values of the abnormal returns.

The test is defined as:

𝒁𝒁𝒘𝒘𝒊𝒊𝑻𝑻𝒘𝒘𝑻𝑻𝒘𝒘𝑻𝑻𝒏𝒏,𝑻𝑻= 𝑾𝑾 − 𝑵𝑵(𝑵𝑵 − 𝟏𝟏)/𝟒𝟒

�(𝑵𝑵(𝑵𝑵+𝟏𝟏)(𝟐𝟐𝑵𝑵+𝟏𝟏)

𝟏𝟏𝟐𝟐 )

This Z-score is converted to a corresponding p-value reported on 10%, 5%, and 1% significance levels.

is favoured by Barber and Lyon (1996) since operating income is perceived as a cleaner meas-ure than earnings.

In addition to operating income, firm performance depends on the productivity of operating assets.

Therefore, the operating income should be scaled, which increase comparability across firms. The predominant measure is return on invested capital (ROIC) where net operating income is scaled by the net operating assets employed by the firm. Unlike nominal accounting measures, ROIC accounts for both the relation between revenue and expenses and the firm’s capital utilization (Petersen, et al., 2017). Though, a common problem in empirical studies is that book value of net operating assets is not reported on a company’s balance sheet in the financial statements. Including ROIC in empiri-cal studies requires comprehensive assessments of each firm’s balance sheet involving subjective categorization of items as either operating or financial assets. A commonly applied substitute of ROIC is the return on assets (ROA) where the book value of assets is used as a proxy for the book value of net operating assets.²⁰ We apply ROA as performance measure expressed by the follow-ing formulas²¹:

𝑨𝑨𝑹𝑹𝑬𝑬 𝑬𝑬𝑬𝑬𝑬𝑬𝑬𝑬_{𝒊𝒊𝝉𝝉} = 𝑬𝑬𝑬𝑬𝑬𝑬𝑬𝑬_{𝒊𝒊𝝉𝝉}

𝟎𝟎.𝟓𝟓 ∗(𝑵𝑵𝑻𝑻𝑻𝑻𝒓𝒓 𝒐𝒐𝑻𝑻𝑻𝑻𝒖𝒖𝒂𝒂 𝑻𝑻𝒇𝒇 𝑻𝑻𝑻𝑻𝑻𝑻𝑻𝑻𝑻𝑻 𝑻𝑻𝒂𝒂𝒂𝒂𝒂𝒂𝑻𝑻𝒂𝒂_{𝝉𝝉−𝟏𝟏}+𝑵𝑵𝑻𝑻𝑻𝑻𝒓𝒓 𝒐𝒐𝑻𝑻𝑻𝑻𝒖𝒖𝒂𝒂 𝑻𝑻𝒇𝒇 𝑻𝑻𝒂𝒂𝒂𝒂𝒂𝒂𝑻𝑻𝒂𝒂_𝝉𝝉) And

𝑨𝑨𝑹𝑹𝑬𝑬 𝑬𝑬𝑬𝑬𝑬𝑬𝑬𝑬𝑬𝑬𝑬𝑬𝒊𝒊𝝉𝝉 = 𝑬𝑬𝑬𝑬𝑬𝑬𝑬𝑬𝑬𝑬𝑬𝑬_{𝒊𝒊𝝉𝝉}

𝟎𝟎.𝟓𝟓 ∗(𝑵𝑵𝑻𝑻𝑻𝑻𝒓𝒓 𝒐𝒐𝑻𝑻𝑻𝑻𝒖𝒖𝒂𝒂 𝑻𝑻𝒇𝒇 𝑻𝑻𝑻𝑻𝑻𝑻𝑻𝑻𝑻𝑻 𝑻𝑻𝒂𝒂𝒂𝒂𝒂𝒂𝑻𝑻𝒂𝒂_{𝝉𝝉−𝟏𝟏}+𝑵𝑵𝑻𝑻𝑻𝑻𝒓𝒓 𝒐𝒐𝑻𝑻𝑻𝑻𝒖𝒖𝒂𝒂 𝑻𝑻𝒇𝒇 𝑻𝑻𝒂𝒂𝒂𝒂𝒂𝒂𝑻𝑻𝒂𝒂𝝉𝝉)

The average book value of assets is used to accommodate the inconsistency in how the income statement and the balance sheet is outlined. The income statement presents the cumulating net op-erating income over the financial year whereas the balance sheet provides a picture of the firm’s as-sets and liabilities at the end of the financial year. The average of beginning and ending year value of assets is an approximation of the average assets used to generate the net operating income in the corresponding year.

Both formulas presented above are applied to calculate ROA for the sample firm and a portfolio of firms within the same industry used as benchmark. The method of calculating abnormal performance using ROA closely follows the procedure suggested by Barber and Lyon (1996).²² For each firm, the

20 In situations where transactions are announced and completed in different financial years, the use of total assets bias the ROA since earnings from discontinued operations are not included in the numerator, but the related assets are in-cluded in the denominator. As result, the change in ROA EBIT and ROA EBITDA might be overstated for selected firms.

21 EBITDA and EBIT are collected from Capital IQ where other operating costs are defined as expenses that have a close relation to the regular operations. Thus, unusual items such as costs related to restructuring or M&A’s are not in-cluded.

22 This method is also in line with empirical studies on corporate divestments including Daley, Mehrotra, & Sivakumar (1997) and Prezas & Simonyan (2015).

abnormal operating performance referred to as adjusted ROA (AROA) is calculated as the difference between the firm’s realized ROA and the median for all firms included in the benchmark portfolio.

𝑬𝑬𝑨𝑨𝑹𝑹𝑬𝑬𝒊𝒊,𝝉𝝉=𝑨𝑨𝑹𝑹𝑬𝑬𝒊𝒊,𝝉𝝉− 𝑬𝑬𝑨𝑨𝑹𝑹𝑬𝑬𝒋𝒋,𝝉𝝉

This measure of operating performance is called industry adjusted ROA. Subsequently, the change in industry-adjusted ROA is computed as:

𝚫𝚫𝑬𝑬𝑨𝑨𝑹𝑹𝑬𝑬_𝒊𝒊=𝑬𝑬𝑨𝑨𝑹𝑹𝑬𝑬𝒋𝒋,𝒑𝒑𝑻𝑻𝒂𝒂𝑻𝑻− 𝑬𝑬𝑨𝑨𝑹𝑹𝑬𝑬_{𝒑𝒑𝑽𝑽𝒂𝒂}

Changes in AROA is applied rather than absolute levels of AROA. Thereby, changes in the sample firms’ performance are analysed relative to changes in the industry benchmark. The advantage is that change models include the history of a firm’s performance relative to its comparison group’s performance. Barber and Lyon (1996) has demonstrated that test statistics using the change in a sample firms adjusted operating performance are consistently more powerful than level mod-els based on absolute levmod-els of a firm’s adjusted operating performance.

Afterwards, the median change across all samples and subsamples are found:

𝚫𝚫𝚫𝚫𝑨𝑨𝑹𝑹𝑬𝑬

����������=𝑴𝑴𝒂𝒂𝒂𝒂𝒊𝒊𝑻𝑻𝒏𝒏(𝚫𝚫𝑬𝑬𝑨𝑨𝑹𝑹𝑬𝑬_𝒊𝒊)

The median change in AROA of sample firms is applied in this thesis rather than the mean change of AROA to reduce effects of extreme observations.

In addition to the analysis of AROA, we investigate changes in the unadjusted ROA. The underlying assumption of this analysis is that the post transaction expected performance of the divesting firm would be equal to the pre transaction performance in absent of the transaction. The objective of this analysis is to determine whether abnormal operating performance is caused by decreased perfor-mance of industry comparables or increased perforperfor-mance in the sample firm.

Based on the methodology described above, positive changes in ROA should be the result of in-creased operating performance indicating that corporate divestments create value. However, there are several pitfalls in interpreting changes in ROA (Petersen, et al., 2017). The ROA performance measure has particularly three drawbacks. First, the measure does not account for differences in systematic risk across firms. Even within the same industry, firms might be exposed to different operational risks (Petersen, et al., 2017). Second, total assets reflect all of a firm’s assets including financial and non-operating assets. Consequently, the use of total assets could understate the true productivity of operating assets (Barber & Lyon, 1996). Third, ROA is affected by changes in ac-counting policies or managerial decisions on financial reporting principles. Operating income is an accrual-based measure with high flexibility which managers can over- or understate by increase or decrease discretionary accruals. Thereby, accrual-based performance measures are exposed to the

risk of earnings manipulation (Petersen, et al., 2017). Operating income might be biased if manage-ment manipulates revenue or expense items for personal benefit. Thus, changes in adjusted ROA following a corporate divestment might be explained by earnings management rather than enhanced operating performance.

To increase the robustness of our findings, we include an alternative performance measure. Some of the problems related to accruals-based ROA can be mitigated by applying cash flow-based ROA.

Based on the argument that cash is king, practitioners often favour cash flow-based performance measures (Petersen, et al., 2017). Therefore, we include an additional performance measure defined as cash flow return on assets calculated:

𝑨𝑨𝑹𝑹𝑬𝑬 𝑪𝑪𝑬𝑬𝑺𝑺𝑯𝑯_{𝒊𝒊𝝉𝝉} = 𝑪𝑪𝑻𝑻𝒂𝒂𝒘𝒘 𝒇𝒇𝑽𝑽𝑻𝑻𝒎𝒎 𝑹𝑹𝒑𝒑𝒂𝒂𝑽𝑽𝑻𝑻𝑻𝑻𝒊𝒊𝑻𝑻𝒏𝒏𝒂𝒂_𝝉𝝉

𝟎𝟎.𝟓𝟓 ∗(𝑵𝑵𝑻𝑻𝑻𝑻𝒓𝒓 𝒐𝒐𝑻𝑻𝑻𝑻𝒖𝒖𝒂𝒂 𝑻𝑻𝒇𝒇 𝑻𝑻𝑻𝑻𝑻𝑻𝑻𝑻𝑻𝑻 𝑻𝑻𝒂𝒂𝒂𝒂𝒂𝒂𝑻𝑻𝒂𝒂_{𝝉𝝉−𝟏𝟏}+𝑵𝑵𝑻𝑻𝑻𝑻𝒓𝒓 𝒐𝒐𝑻𝑻𝑻𝑻𝒖𝒖𝒂𝒂 𝑻𝑻𝒇𝒇 𝑻𝑻𝒂𝒂𝒂𝒂𝒂𝒂𝑻𝑻𝒂𝒂_𝝉𝝉)

Whereas EBITDA and EBIT collected in the Capital IQ database do not include costs classified by management as unusual items, cash from operations accounts for all cash flow related costs in-cluded in the net income. Thus, the ROA CASH performance measure is affected by the costs re-lated to corporate restructuring and divestitures.

ROA CASH cannot mitigate all problems of earnings management as cash flows can be manipulated through the sale of receivables or cutting research and development costs (Petersen, et al., 2017).

Despite of the drawbacks related to accruals-based performance measures, FASB still perceive ac-crual income as superior to the cash flow statement for measuring a firm’s value creation (Petersen, et al., 2017). Barber and Lyon (1996) has documented that cash-based performance measures are generally less powerful to determine abnormal operating performance around a corporate event compared to accruals-based measures. However, the cash flow-based performance measure can be used to analyse whether changes in adjusted ROA EBIT and EBITA are caused by reversals of pre-transaction accruals.

Benchmark

The methodology described above compares changes in ROA of the sample firm with changes in ROA of a particular benchmark. The objective of using a benchmark is to isolate the change in op-erating performance coming from the corporate divestment by removing the expected performance measured by the benchmark. The choice of an appropriate benchmark involves a trade-off between comparability and data availability which will be discussed below.

The most straight-forward approach would be to use pre-transaction firm performance as a bench-mark for the post-transaction performance. However, this method is perceived as too simple. We seek to compare the performance of each sample firm to a benchmark based on a reference portfolio

of firms in the same industry, i.e., a control group. The assumption behind industry-matching is that some of the variation in operating performance can be explained by an industry benchmark. Thereby, the objective of the industry benchmark is to remove the change in the sample firm’s performance stemming from a general change in the industry. The industry-matching method is often applied by matching firms to other firms with either the same two-digit or four-digit SIC code. Four-digit SIC code matching includes fewer comparable firms that are more closely matched on industry, but where availability of enough comparable firms for all industries is often a problem. To ensure data availability, we apply reference portfolios of firms sharing the same two-digit SIC code. In accord-ance, Barber and Lyon (1996) demonstrates that matching on four-digit SIC codes provides no im-provement in the explanatory power of test statistics compared to two-digit SIC codes.

The reference portfolios are constructed by identifying all listed and privately owned European firms with available financial data in the Capital IQ database having the same two-digit SIC code. A com-mon method to increase comparability is to match firms on country level. However, this implies prob-lems of data availability on enough firms from each country in each industry when constructing ref-erence portfolios on country level. Therefore, the control group of each divesting firm consists of all firms sharing the same SIC industry code within European Developed Markets. Thereby, geograph-ical segmentation of the reference group for each firm matches the geographgeograph-ical screening of the corporate divestments included in the sample.²³

Fama and French (1995) document that small firms have return on equity measures that mean-revert more quickly than similar measures for large firms. We perform a size adjustment to address the concerns expressed by Fama and French (1995), that small firms have lower earnings-to-book-equity ratios. Control firms are size matched based on book value of total assets which should be minimum 50% and maximum 150%. For 10 firms in the total sample, the size requirement was re-moved since no control firms existed that fulfilled the requirements.

Test statistics

The statistical test used to test the median changes in operating performance is the nonparametric Wilcoxon signed-rank test. According to Barber and Lyon (1996), the Wilcoxon non-parametric test is uniformly more powerful than simple parametric t-statistics as no assumptions of normal distribu-tion is required. Thereby, the test statistic is useful even in case of extreme observadistribu-tions where the normality assumption of other test statistics is not fulfilled. Based on the relevant hypotheses de-scribed in Section 5, the Wilcoxon signed-rank is used to determine whether there is statistical sig-nificant AROA.²⁴

23 Please refer to Section 7.2.

24 Please refer to Section 6.2 for further details of the Wilcoxon signed-rank test. The test is performed with the same methodology, however where BHAR is replaced with ROA.