Discussion of assumptions and limitations

In this section, the overall objective is to assess some of the key limitations affecting the overall reliability of the findings obtained in Chapter 5. Hence, by examining the implications and drawbacks of such limitations an understanding of how the findings should be interpreted in relation to the research question can be achieved. First, the implications of standardised selection variables will be outlined, secondly, the applied proxy for growth and its limitation will be examined and, finally, the applied multiples’ limitations in relation to their appropriate application.

6.3.1 Selection variables’ appropriateness

As described in Section 1.3, delimitations are made through a methodological choice of the selection variables and the combinations of these, i.e. the SARD1-5 and INDSARD1-5

Page 113 of 177

methods, respectively. The choice of the five proxies used in this thesis is based on ensuring comparability to the original study by Knudsen et al. (2017). However, a limitation arises from choosing variables in such a standardised manner, as it is not certain that the same fundamentals are optimal selection variables in subject to the applied multiples, peer pools, industry groupings, and size of firms. In fact, it is evident from the performed analysis that the selected variables could be optimized, and possibly affect the relative performance of SARD and INDSARD, ultimately influencing the findings in relation to the stated hypotheses in Section 1.1. In the following section, the standardized selection variables and their combinations will be discussed related to their consequential drawbacks.

Firstly, the analysis performed in Chapter 5 finds evidence suggesting that the choice of relevant selection variable is crucial, as it can affect the relative performance between the SARD and INDSARD selection methods. As an example, the analysis in Section 5.2.2 shows a significant improvement of prediction accuracy for EV/Sales when the EBIT margin is added as a fundamental, but a less significant improvement is seen for EV/EBITDA and EV/EBIT indicating that the standardized selection variables across multiples are not optimal as the relevancy depends on the specific multiples. The improvements from the adding of selection variables do not only vary depending on the multiple but also the underlying peer pools, i.e.

Denmark and EU. The individual variables’ significance of accuracy subject to the multiple and peer pool, is also found to be of relevance. This relevance is examined in the univariate tests in Section 5.6.5, which discovers that none of the three multiples prioritize the selection variables identically. In fact, the rankings and significance of the single valuation fundamentals are very different indicating that the selection variables should be customised, not standardised, for each multiple and peer pool in subject to the selection method.

Secondly, the analysis also finds that the combination of selection variables, in addition to the individual accuracy, is crucial for determining the overall prediction accuracy. As it is evident from Chapter 5, the relative performance of the selection models also depends on how the selection variables are combined as the optimal combination of SARD and INDSARD varies depending on the multiple, peer pool, industry group, and size of the firm. As it is seen in Section 5.6.3.1 and 5.6.3.2 the robustness checks related to errors groupings indicate that the optimal selection variables under SARD and INDSARD vary depending on the industry group and whether the target firm is a Small-, Mid- or Large Cap. Thus, by optimizing the combination

Page 114 of 177

of selection variables to fit these factors, the obtained results could change the overall finding of SARD and INDSARD’s performance in valuation accuracy relative to each other.

Evidently, the limitation in form of the standardised choice of selection variables and their combinations leads to drawbacks that can distort the findings in relation to the stated hypotheses. Whether the relative performances between SARD and INDSARD would have changed if variables were optimized, must be kept in mind when concluding on the results. The underlying proxies and combination are not chosen in relation to their appropriateness to each other and the individual interrelation to the applied multiples, peer pools, industry group, and size thereby suggesting that future research is necessary to make these choices. The considerations of such future research will be addressed in Section 6.4.

6.3.2 Applied proxy for growth

As a part of the research design, the same selection variables applied by Knudsen et al. (2017) are used in this study. However, an exception occurs in relation to the proxy for growth as Knudsen et al. (2017) use a forecasted two-year EPS growth rate while a one-year historic revenue growth is applied in this thesis. As the growth rate is added in the fourth combination of SARD, such difference implies that the interpretation of SARD4, SARD5, INDSARD4, and INDSARD5, respectively, cannot be fully compared across the two studies.

When inspecting the theoretical derivation of the EV/Sales, EV/EBITDA, and EV/EBIT multiples in Section 2.2, it is apparent that future growth in FCFF is an underlying value driver for Enterprise Value. However, it is not possible to observe such growth directly in the market, and it would be comprehensive to perform the calculations of historic FCFF manually for each target and firm in the peer pool. Therefore, the ambition is to find the most appropriate proxy to reflect such growth rate. For this purpose, previous literature utilizes different approaches as some apply revenue growth while other favor earnings growth similar to Knudsen et al. (2017).

In theory, earnings growth could be interpreted as the most appropriate proxy as increasing earnings will lead to higher free cash flows, hence, it serves as an underlying value driver for FCFF growth. In addition, literature in general favor forecasted earnings growth rather than historic. However, as declared in the methodological setup for this study, it has not been possible to extract forecasted earnings for Danish target firms due to limited analyst coverage. As an alternative, historic earnings growth could have been applied, however, performing such calculations are impacted by transitory costs which lead to rather volatile patterns for the Danish

Page 115 of 177

sample affecting its correlation to enterprise value. Hence, rather than excluding the growth selection variable, revenue growth is applied instead. Despite the lack of direct theoretical connection to FCFF, several studies achieve improved accuracy when applying revenue growth as the proxy including Nel & le Roux (2015), Serra & Fávero (2018), and Herrmann & Richter (2003). Similar to the forecasted earnings, forecasted revenue growth is not available in the sample years either, thus, historic figures are applied. It could be argued that historic numbers do not affect the future based on the Random Walk theory in relation to the Law of One Price as stated in Section 2.2, however, investors seem to some degree consider the past performance when evaluating firms in practice (Berk & DeMarzo, 2017). This also corresponds to the points made by interviewees in this study in relation to the application of listed multiples: “[…] So we made a regression based on historical growth. Even though the valuation is only about the future, the historical figures also have something to say about verifying expectations” (Interviewee 3, 2021, Appendix 15).

Furthermore, the reported figures applied in this study is solely based on one-year growth as longer periods could not be used without losing an even greater part of the initial population of Danish targets. Hence, the applied growth rates can be influenced by fluctuations with single years’ performance protruding.

Ultimately, the limitations of applying a historic growth rate of revenue should be considered when interpreting SARD4, SARD5, INDSARD4, and INDSARD5, as the findings cannot be directly compared to Knudsen et al. (2017). Furthermore, prediction accuracy could potentially be improved if a future earnings growth variable is used as a proxy for growth in FCFF, while the relative performance of the selection methods containing the growth variable could also have shown a different pattern.

6.3.3 Applied multiples

In relation to the multiples applied in this thesis, both delimitations and limitations are present. The methodological delimitation of only using EV-based multiples in the analysis leads to a twofold consequence: (1) it limits the comparability to previous studies, and (2) it limits the possibility of examining whether the obtained results remain robust across multiple types. Furthermore, limitations arise from the multiples not being applied appropriately in relation to their individual applicability, possibly influencing the results’ reliability. In the following, the implication from such limitations will be described in context to the multiples used in this study.

Page 116 of 177

As is evident from the findings in Chapter 5, EV/Sales obtains high prediction errors compared to the two earnings-multiples. In practice, the appropriate application of an EV/Sales multiple depends on a firms’ business characteristics and operations. Certain businesses such as venture capital firms or firms with contracted revenue, are often characterized by high growth opportunities, but will often experience low or even negative earnings. Valuing such firms based on their earnings would therefore not be representative of their true value nor potential.

Contrarily, using an EV/Sales multiple can better capture its future value if it is reflected in its revenue turnover. As the methodological setup in this study does not consider the individual multiples’ application related to each target, it is possible, that the obtained results can be distorted. As described in Section 4.1.3, all observations with negative earnings are excluded from the datasets, to where EV/Sales would in fact be an appropriate valuation multiple. In contrast, the remaining observations all have positive earnings to where the EBITDA and EBIT multiples could be considered more appropriate than Sales. The lack of appropriate application of EV/Sales can be one underlying reason for why the multiple yields such high prediction errors, thus, if applied appropriately the results could possibly have changed.

Similarly, for EV/EBITDA and EV/EBIT the appropriateness depends on the individual firms’ characteristics. EBITDA is not affected by depreciation and amortization and is therefore used for firms where Capex is not an important factor in estimating the firms’ value or future potential, contrarily, including these non-cash expenses can distort the view of firm value. For instance, an EV/EBITDA multiple would be appropriate in valuing certain industries, e.g.

service-based firms, which are characterized by having economic value in intangible assets such as human capital. Contrarily, EV/EBIT is used in practice when Capex plays a significant role in the firm’s business, e.g. asset-heavy industries. Practitioners, however, are to a larger extend applying an EV/EBITA multiple when Capex should be considered, since it leaves goodwill out, not to disturb the comparability between firms (Rosenbaum & Pearl, 2009; Nielsen;

Kjærum, Appendix 15). Ultimately, the choice of appropriate multiples depends on the individual firm and what measure mirrors the FCFF best, i.e. the normal cash generation. As the study does not distinguish between firms in relation to the appropriateness of the multiples’

application, a drawback from this limitation can impact the findings related to the stated hypotheses.

Page 117 of 177