6. Discussion
6.2 Interpretations and practical relevance
Page 100 of 177
of SARD and INDSARD are indeed more accurate when using cross-border peers, however, solely until the EBIT margin is applied, where the most optimal combination performs better using a pure Danish peer pool. Similar, Dittmann & Weiner (2005) overall find that European countries obtain more accurate valuation predictions for EV/EBIT once peers are found among European Union member states or within OECD countries. However, as one of only four exceptions, the quoted study’s findings do not hold for Danish targets, as their study suggests that valuation errors are minimized when peers are in fact found using home-country firms. The differences to this thesis’ findings can be related to several factors. While this study investigates SARD and INDSARD with several fundamentals, Dittmann & Weiner (2005) solely considers ROA and Total Assets as selection variables. Furthermore, their sample of Danish targets solely covers the period from 1999 until 2002, i.e. there is no overlap in the time period. An important attribute to the sample applied in this thesis relates to the accounting standard as all firms adapted IFRS in 2005. Thus, dissimilarities in accounting standards are presumably affecting ROA and Total Assets in Dittmann & Weiner's (2005) research, unlike in this thesis, when finding peer cross-border.
Page 101 of 177
literature, which showed discrepancies between some academics, hence, which of the two schools of thought on peer group selection produce the most accurate results. As this thesis’
obtained result favours a fundamental-based peer selection, the underlying reason for achieving it, contrarily to an industry approach, will be examined through a discussion of theory and a practical understanding.
Explanation 1: Industry affiliation in a small market
One apparent explanation to the superiority of SARD compared to industry on the Danish market roots in the consequence of applying a sample and dataset consisting of a limited number of observations. As described in Section 4.4.1, a level-up approach is used under the industry selection criterion, to make sure that the most detailed and informative industry layer is applied where possible in the creation of peer groups. Despite the methodological choice of ‘level up’, it is evident that very few firms can be peered based on the GICS 8-digit level, as only 7.8% of the firms from the dataset can be matched (Figure 4.5 in Section 4). Instead, the majority of peer groups are formed on the GICS 4-digit layer. The categories of classification on this layer are ‘industry groups’, which can be considered to be very broad definitions, e.g. Diversified Financials (4020), Energy (1010), Materials (1510), and Retailing (2550) emphasizing the lack of distinctive industries. However, such broad GICS layers are necessary to use, as there are simply not enough listed firms in Denmark to find comparable firms on the detailed and most informative industry layer (GICS 8-digit). This indicates that the methodological application of the industry criterion is limited on a small market, which can explain the high prediction errors for the industry method achieved in Section 5.2.1. From the performed interviews with practitioners of multiple valuation, it is also evident that a small market as Denmark creates limitations in using an industry classification scheme like GICS. “[..] Denmark gives too small a sample to also divide Denmark into industries, so the sample will be even smaller in the individual industries.
Within the individual industries, there can be larger fluctuations in how a company operates.”, (Nielsen, M.R., 2021, p. 154, Appendix 15). Similar, the reviewed literature from Dittmann & Weiner (2005), finds that for Denmark and other small markets, industry performed less accurately than a selection of peers compared to the actual market. They suggest that in smaller countries firms are exposed to being misclassified in industries, or that firm value does not change across industries. The arguments all lead to the question of whether expanding the peer pool will then
Page 102 of 177
yield better prediction accuracy under an industry selection, and possibly change the relative performance between SARD versus Industry.
In Chapter 5 such expansion is examined as the industry selection criterion is used on an EU peer pool, which in fact leads to an increase in the prediction accuracy compared to a Danish peer pool. However, the relative performance to SARD does not change even when using an EU peer pool, as industry still performs worse than SARD. This finding suggests that SARD is ultimately better than industry in predicting multiple valuation, even on a larger peer pool.
However, one important factor influencing the results when expanding the peer pool is the country-specific difference in subject to industry classification, which will be addressed in Section 6.2.3. Ultimately, the reasoning for the results obtained in this study can be explained through the industry classification scheme applied, in subject to the current economy, which will be addressed in the following section.
Explanation 2: Industry classification schemes becoming obsolete
Another arresting explanation to why an industry peer selection is less accurate than a SARD, roots in the application of the industry classification scheme GICS and the system in itself. The literature favouring an industry approach was published back in 1992 (Alford) and 2000 (Cheng and McNamara). Since then, the research investigating the same topic has shown somewhat discrepant results, with an increasing consensus of favouring fundamentals in the newer literature. A possible explanation to the different results, other than differences in the methodological set-up, could rest upon the rapidly changing nature of the economy and the firms constituting it. Since the origin of the two schools of thought, it is evident that a changing nature of firm’s operations and offerings have taken place, blurring the lines of what constitutes an industry and the definition criteria, which suggest that the existing industry classification schemes can possibly have become obsolete throughout the years. An interesting finding, emphasizing the argument behind this very explanation, is found by Dittmann & Weiner (2005).
Their study shows that the accuracy of a peer selection based on industry deteriorated through the ‘new economy’ boom where firms transitioned to being more service-based, suggesting that SIC industry classification could not separate such service-firms from the ‘old economy’, here characterized by firms from a manufactured-based economy. The changing nature of firms’
business models in relation to industry classification is also an important topic circulating
Page 103 of 177
practitioners’ minds: “[…] there are rarely two companies that are alike, and it gets more and more difficult as companies differentiate more and more and create new markets and more need to think out of the box, and work across industries. It is becoming harder and harder to find an industry ‘pure pace’” (Nielsen, M.R., 2021, p.
147, Appendix 15). It is evident, that practitioners face the exact problem of the difficulties in assigning firms to an appropriate industry, which blurs the definitions and similarities of the firms constituting them.
Even though the GICS classification system has expanded in terms of sectors and sub-sectors since its time of origin, the classification options can be argued not to match the type of firms existing in the current economy, or at least not on the detailed level necessary for it to reflect the three underlying value drivers: profitability, risk and growth. Consequently, the theoretical underpinnings of an industry group reflecting similarities in these three key measures are challenged. As described in Section 2.3.2, it is shown through the SCP-paradigm and Porter’s (2008) theory on ‘the five forces’, that firms within an industry share similarities in profitability, risk, and growth. From practitioners, it is also evident that industry is a proxy which should capture these three measures: “The industry should preferably be involved in capturing the risk. But also, the future growth potential to some extent and also the profitability.”, (Nielsen, M.R., 2021, p. 151, Appendix 15). Returning to Michael E. Porter’s definition of an industry, i.e. a “set of firms having similarities in their products and services”, the economy’s transition to more service-based firms with many different offerings, hence, operating across industries, can disrupt this very definition and challenge the theoretical underpinnings of industry being an appropriate proxy for profitability, risk, and growth. In practice, it is also evident that it is becoming more difficult to classify firms as they operate across different industry groups (Nielsen, M.R., 2021, Appendix 15). Overall, it suggests that the industry representing the underlying three measures, making it a good tool for finding a peer, have become vaguer and could possibly also explain the high percentages errors obtained in the results using industry affiliation in this thesis.
Overall, based on the two schools of thought a peer selection strategy based on fundamentals should be employed, i.e. SARD, rather than industry classification. Industry is understood from a practical perspective to be used as a selection criterion, but not, in the same manner, applied in this thesis. The industry criterion can narrow the comparable pool, but before it can represent similarities in profitability, risk, and growth it must be accurately matched through an industry.
However, with the change of the firms in the ‘new economy,’ it is apparent that the classification
Page 104 of 177
systems are not able to reflect this complexity in the firm’s products and services. If improving the application of the GICS classification in this thesis or use a more detailed industry layer it might change the achieved findings. Overall, it means that industry is not worse than fundamentals in all aspects, but it depends on how it is used. As Nel & le Roux (2015) point out there is no superior school between fundamentals and industry, but many variables are affecting the optimal approach on peer selection, leading to the question of how to optimally combine industry with fundamentals which are discussed in the following section.
6.2.2 Why does SARD
within
industries not consistently outperform SARD?In the analysis within this study, it is examined whether applying SARD within industries yields greater prediction accuracy for Danish targets compared to using fundamentals with no consideration of industries. It was assumed that a combined approach would capture more information upon firm valuation based on the fact that practitioners use industry alongside fundamentals as a key characteristic for the business profile as well as a proxy for risk and growth (Kjærum; Nielsen; Interviewee 3, 2021, Appendix 15). This assumption is reflected in Hypothesis 2, however, it does not consistently find support through the results obtained in Chapter 5. Findings rather suggest that INDSARD is not superior in all combinations to SARD as it depends on selection variables included. The explanation as to why a combination of SARD and industry does not consistently yield more accurate valuation estimates can relate to several factors. First of all, one limitation when applying INDSARD with a Danish peer pool relates to the issue of too few observations per industry. However, when expanding the peer pool, the pattern between INDSARD and SARD stays unchanged, which indicates that the size of the underlying peer pool is not the only explanation. Other explanations relate to the information captured by the applied selection variables relative to what industry affiliation contributes with, and the application of a broad GICS 2-digit industry layer. These explanations will be discussed in the following.
Explanation 1: Predefined peer groups
The initial argument relates to the results being highly affected by the nature of the underlying dataset when solely applying a home-country peer pool for Danish targets in this thesis. It is crucial to bear in mind, that even though 494 observations are present, peer groups are formed on a yearly basis in the period from 2010 until 2019. Thus, the pool of comparable firms only
Page 105 of 177
contains between 53 and 64 observations each year as seen in Appendix 1, Table 1.1.
Consequently, when dividing the observations into further groupings as is the case when applying fundamentals within industries, the market will be limited even further for the peer selection. As seen from Table 4.6 in the methodology Section 4.5, solely seven industry groups are employed including Industrials, Health Care, Consumer Discretionary, Telecom, Materials, Consumer Stables, and IT. However, what is noteworthy relates to the distribution of observations within these industry groupings, as on a yearly basis the majority of industries solely contain between five and seven observations including target firms, with five being the minimum required number to create peer groups of four for a target. Thus, when applying SARD within those industries, the peer selection comes down to a choice between the remaining one and two firms. Some industries are even completely predefined, leaving no impact of the SARD approach at all. Industrials and Health Care are the only industries leaving explanatory power to SARD as they consist of 12 to 20 firms each year. As the findings related to Hypothesis 1 suggests that industry affiliation on a stand-alone basis generates the least accurate prediction accuracy, the Danish target firms belonging to ‘small industries’ with few observations could presumably benefit from peer selection across industries when applying numerous selection variables. The findings of INDSARD not consistently outperforming SARD using a Danish peer pool, reflect that only a part of the industries favours a combination method, i.e. industry barriers partly impact results negatively as there are not enough observations within each year for all industries.
Explanation 2: Interrelation between selection variables and industry affiliation
A second explanation to the findings of this thesis relates to the interrelation between the selection variables applied and industry affiliation. As the expanded peer pool containing EU firms does not change the pattern favouring INDSARD relative to SARD, the findings cannot solely be related to the few observations as such implication is not present for the EU peer pool.
The first three combinations of fundamentals yield improved accuracy by applying industry affiliation, since INDSARD1, INDSARD2, and INDSARD3 are more accurate than SARD1, SARD2, and SARD3 for both EV/Sales, EV/EBITDA, and EV/EBIT. These findings imply that industry is able to capture information which is not reflected in ROE, Net Debt/EBIT, and Size, respectively. Such interpretation corresponds to the theoretical arguments in Section
Page 106 of 177
2.3 in favour of the first school of thought building on the assumption, that a firm’s performance is a function of the external environment it operates in. As discussed in Section 6.1, Alford (1992), Cheng & McNamara (2000), Knudsen et al. (2017), and Bhojraj & Lee (2002) in a combination with fundamentals also find that industry offer explanatory power to the underlying drivers for firm valuation. Based on the conducted interviews such results also correspond to peer selection in practise where industry is perceived as an important measure for operational risk (Kjærum; Nielsen; Interviewee 3, 2021, Appendix 15), which is not otherwise captured. This could potentially indicate that the applied risk proxies, Net Debt/EBIT and Size, are not fully appropriate reflections of firm risk. For instance, this is addressed by Nielsen, M.R. (2021, p. 151, Appendix 15):
“[..] risk becomes difficult to capture in fundamentals. So that is why we prefer to find something within industries because they typically have the same terms with customers and are affected by the same external market such as the regulatory effects that can be in a market. These are the risks that a company in that industry faces that one with the same fundamentals in another industry does not face”
However, findings suggest that once Historic Revenue Growth and EBIT margin are added as selection variables, industry no longer captures further information, as SARD4 and SARD5 in fact yield greater prediction accuracy than the corresponding INDSARD4 and INDSARD5.
Such results indicate that once more appropriate proxies for profitability, risk, and growth are applied, industry affiliation does not reflect further information upon market prices. Thus, it is apparent that fundamentals are ultimately better at predicting multiples when growth and profitability in the form of EBIT margin, is applied rather than using a GICS 2-digit code. This suggestion corresponds to Herrmann & Richter's (2003) study as when applying SIC codes, industry classification does not contain superior information to what is already controlled for by the fundamental selection variables once profitability and historic revenue growth are applied.
Explanation 3: Broad industry definitions
Such overall findings, that industry does not improve accuracy once all fundamentals are applied, may also be related to a potential fallacy in this study’s methodological set-up, as industry is reflected by GICS 2-digit codes in INDSARD. Related to the discussion in the previous Section 6.2.1, the GICS codes are not necessarily an optimal indication of industries, thus, the codes could presumably be of even greater misjudgement in INDSARD where the
Page 107 of 177
broadest layer of industry, i.e. GICS 2-digit, are applied. This could potentially also explain to some degree why SARD4 and SARD5 outperform the corresponding INDSARD models.
Practitioners support such interpretation in the interviews as they explain that it is necessary to concretise industries depending on the firm’s specific offerings, position in the value chain, and overall business models (Kjærum; Nielsen; Interviewee 3, 2021, Appendix 15). Dittmann &
Weiner (2005) achieve a similar conclusion, as mentioned in Section 6.2.1 when examining Danish targets from 1999 until 2002, as applying SIC codes yields less accurate predictions for EV/EBIT multiples than simply using the entire market, regardless of whether the market is represented by EU or OECD firms. Their argument rests upon the fact that in smaller countries, firms are more likely to either be misclassified in industries.
To summarize, the inconsistent results of INDSARD’s performance relative to SARD can be explained by several factors. First, for the Danish peer pool, limitations lie within too few observations in each industry, while once the peer pool is expanded such interpretation does not hold. Hence, the explanations can be related to the fundamentals’ appropriateness to serve as proxies for the underlying value drivers, i.e. profitability, risk, and growth. Ultimately, the understanding could, similar to the pure industry approach, be related to the GICS 2-digit codes which do not seem to be appropriate as discussed in the previous section. Similarly, prior research suggests that the performance of peer selection methods is highly dependent on the industry examined (Knudsen et al., 2017; Lie & Lie, 2002; Liu et al., 2002; Nel & le Roux, 2015). Furthermore, when applying the EU peer pool, it is important to bear in mind that explanations cannot with certainty be related to the ‘horse race’ between industry and fundamentals, since cross-border differences also potentially disturb the results. Thus, interpretations related to the underlying peer pools, i.e. Danish versus EU peers, are discussed in the following section.
6.2.3 Why is an EU peer pool better than a home-country?
The final step in the analysis examined whether a smaller, home-country, or a larger, cross-border peer pool is more appropriate when predicting multiple valuation for Danish targets.
Hypothesis 3 assumed that an EU peer pool is preferred due to the limitations of the small, Danish market. The findings in Section 5.4 somewhat support such postulation when considering the optimal combinations of SARD and INDSARD for EV/EBITDA and EV/EBIT, however, for EV/Sales the hypothesis is rejected. Furthermore, less optimal
Page 108 of 177
combinations including SARD and INDSARD with fewer selection variables do not yield improved accuracy from an expansion of the peer pool.
In general, it could be assumed that the more firms to choose from, the greater probability of finding a similar company to a target. As Denmark solely consists of between 53 and 64 listed firms per sample year, both an industry and a fundamental approach encounter challenges in terms of finding similar companies among such few potential peers. In general, it could seem relevant to consider whether a true peer in terms of both future profitability, risk, and growth, even exists in a peer pool only consisting of Danish firms. Similarly, all interviewees in this study unanimously point out that the Danish market is simply too small to identify peers. As discussed in Section 6.2.1, the ‘new economy’ contains a much higher share of specialized firms as opposed to a more manufacturing-led economy in previous times, which can also explain why Denmark no longer withholds enough peers, as pinpointed by a practitioner as well: “The spread between peers is getting bigger and bigger. […] Companies are differentiating more and more, and we need to look broader geographically to find a comparable peer” (Nielsen, M.R., 2021, p. 147, Appendix 15).
When expanding the peer pool from Denmark to European Union member states, not only is industry groupings extended with more firms overcoming issues of predefined peer groups as examined in the previous sections but the spread between fundamentals is impacted as well.
This is seen already in the descriptive statistics in Section 5.1 as ROE, Size, and EBIT margin obtained lower spread in EU compared to Denmark when examining the IQR. When the peer pool is smaller, leaps in fundamentals impact prediction errors to a larger extent as firms might be peered despite significant differences in selection variables when applying the SARD
Page 109 of 177
approach. An illustrative example is provided in Table 6.1 showing how peer selection for Bang
& Olufsen A/S would appear if solely six firms are present in the peer pool. As the ranking of the firms is conducted relative to the entire sample, the peer group would contain firms with large dissimilarities in ROE relative to Bang & Olufsen A/S. With an expanded peer pool, it can be assumed that such leaps in firms’ fundamentals will diminish. The complication of this methodological fallacy of SARD’s ranking will be discussed further in Section 6.4 in conjunction with suggested future research, while it is solely an interpretation of disadvantages of smaller peer pools in this section.
A greater number of economic fundamentals is however not unconditionally an advantage as it is crucial for prediction accuracy of SARD, that fundamentals are fully comparable across the matrix of ranks, i.e. biased estimates occur if the selection variables do not represent indistinguishable economic characteristics. Thus, the fact that the finding’s support for Hypothesis 3 depends on the multiple and the combination of fundamentals in SARD and INDSARD, indicates that the informative power of the selection variables is not identical across countries. Such assumption is supported by the univariate tests performed in Section 5.6.4 as the ranking of selection variables varies dependent on the peer pool being Danish or from EU.
This finding corresponds to Serra & Fávero's (2018) results examining Brazilian target firms using a home-country and a US peer pool, as they suggest that the explanatory power of fundamentals depends on the specific country. Hence, when the expansion of peer pools is conducted across borders new challenges arise. Ultimately, it is a trade-off between obtaining a larger peer pool with a greater probability of finding comparable fundamentals, while on the other hand cross-border differences mitigate the similarity between peers. The optimal peer pool expansion would occur if it was pure on the home-country market, i.e. profiting from a higher number of peers available for the selection methods to choose from without disruption from cross-border differences. Such setup is the case in the original SARD paper when Knudsen et al. (2017) perform peer group selection on the US market, where no country borders are disrupting findings as the peer pool is one regulatory unity with presumably less cultural differences than across European countries. Hence, this could to some degree account for the generally lower prediction errors in Knudsen et al.'s (2017) study relative to the errors in this thesis.
Page 110 of 177
Potential interpretations as to why cross-border differences arise relate to the overall external environment surrounding firms in different markets. Such dissimilarities can be related to a handful of factors, including for instance corporate tax rates, legislation and regulatory conditions within each EU country, salary, and hiring conditions, the culture and trends affecting firm circumstances, etc. Accounting differences should though not be an influential factor on the accuracy of the SARD approach since all firms in the EU peer pool apply IFRS.
Even as companies have a great deal of flexibility within the standard such as capitalization of R&D costs or classification of leases (Petersen et al., 2017), these differences should not affect the EU peer pool any further than the Danish. However, the value relevance of accounting figures could presumably still serve as an explanation to some degree as firms can seek to exploit such accounting flexibility in order to optimize local regulations as e.g. tax rates. Potentially, firms operating in countries with higher burdens of taxation could to a greater extend seek to expense R&D costs rather than undertaking a capitalization in order to minimize the tax accounts. Nevertheless, the main explanations related to the cross-country differences could be assumed to lie within a range of external factors affecting a firm’s risk as well as the prospects for profitability- and growth potential. The specific interpretation of such differences among countries is outside the scope of this thesis. However, the findings indicate that the optimal peer pool lies somewhere in between a pure home-country and a broad EU peer pool, which will be addressed in Section 6.4 related to future research. It should also be kept in mind, that especially for large Danish firms it might even be perceived as an advantage applying cross-border targets due to a large degree of internationalization.
6.2.4 Implications of applying SARD in practice
In general, it is understood from this study that SARD is more accurate than selecting peers based on an industry approach. However, from a practical perspective, it can be argued that SARD is not a preferred tool, as the relatively high percentage errors imply that it is not an accurate multiple predictor. As the market prices are ultimately a reflection of investors’
aggregated expectation of firm value, the high level of prediction errors implies dissimilarities between investors’ choice and prioritization of selection criteria to the those applied in this thesis. From the discussion above, it is evident that a practical approach to selecting peers is built on the same underlying idea as presented in this study, i.e. a fundamental- and industry-based approach, though it is much more prone to subjectivity and less standardized than the