Who’s a major? A novel approach to peer group selection: Empirical evidence from oil and gas companies

This study presents a novel approach to selecting comparable companies in equity valuation. While valuation multiples is probably the most common valuation method in practice, discounted cash flow and residual income valuation models are advocated by academics. A key aspect in valuation by multiples is peer group selection. In this paper, we examine the usefulness of econometric techniques in peer-group selection for the largest companies in the international oil and gas sector. Using Chow tests, we are able to identify firms with similar relationships between valuation multiples and relevant value drivers. These results of our study suggest that analysts and investors should, when carrying out valuations, be careful in selecting the companies that comprise the peer groups. Comparable company selection could be carried out using econometric techniques that select companies on the basis of similarities in the relation between financial information and market valuation, instead of being based purely on analysts’ subjective judgments. Subjects: Corporate Finance; Investment & Securities; Financial Statement Analysis; Gas Industries; Petroleum & Oil Industries

ABOUT THE AUTHORS Frank Asche holds a PhD from the Norwegian School of Economics and Business Administration (1996). His research interests focus on aquaculture and seafood markets, but he has also been doing work in fisheries management and energy economics. Professor Asche has published numerous articles in international journals in economics and leading multidisciplinary journals like Science and PLoS One. He has also undertaken a number of research projects in Norway as well as for international organisations like the FAO, OECD and WTO. Bård Misund holds a PhD from the University of Stavanger. His research interests cover topics on commodity markets and firms, including commodity price behaviour, the spot-forward relationship in futures markets, determinants of commodity firm stock returns, financial statement analysis and valuation of oil and gas firms. Professor Misund has published articles in international journals in accounting, finance and economics. Misund has more than 10 years of industry experience.

PUBLIC INTEREST STATEMENT
Equity valuation is one of the most important applications of finance theory. Survey studies suggest that valuation by multiples is undoubtedly one of the most common methods of equity valuation in practice. Valuation multiples are used to determine share price estimates, as well as for the valuation of initial public offerings, investment bankers' fairness opinions, leveraged buyout transactions, seasoned equity offerings and other merger and acquisition activities. A crucial element of multiples valuation is the selection of an appropriate peer group. Typically, the financial analyst would select peer companies on the basis of subjective judgements. This approach to peer group selection might not be optimal. We propose an alternative approach using econometric analysis. The purpose of our method is to select peer group companies on the basis of the relationship between fundamental information and market valuation. The oil and gas sector is used as a case study.

Introduction
Equity valuation is one of the most important applications of finance theory. Although academics advocate the use of the discounted cash flow model and its derivative, the residual income valuation model, RIV (Copeland, Koller, & Murrin, 2000;Palepu, Healy, & Bernard, 2000;Penman, 2001), valuation by multiples is undoubtedly one of the most common methods of equity valuation in practice (Asquith, Mikhail, & Au, 2005;Minjina, 2008;Roosenboom, 2007). Survey-based evidence suggests a dominant role for the price-earnings ratio among analysts in determining and evaluation of share prices (Demirakos, Strong, & Walker, 2004). Valuation multiples are also used in the valuation of initial public offerings, investment bankers' fairness opinions, leveraged buyout transactions, seasoned equity offerings and other merger and acquisition activities, M&A (DeAngelo, 1990;Kaplan & Ruback, 1995;Kim & Ritter, 1999).
An adequate process for selecting comparable firms is a necessary prerequisite for valuation by multiples. Typically, comparable companies are selected from the same industry. The underlying assumption is that these firms share the same risk, profitability and accounting methods. This has been the topic of studies by Bhojraj and Lee (2002), Boatsman and Baskin (1981), Alford (1992) and Zarowin (1990). An important conclusion that can be drawn from these studies is that industry membership is an important factor in selecting comparable firms. Hence, we focus on an important industry-oil and gas. The oil and gas industry contain some of the world's largest companies and also has a clear structure for grouping the companies-majors, independents, international. The industry is accordingly well structured to investigate the value relevance of such groups. For the companies, the groupings are important because of the way analysts investigate relative financial performance and therefore for the companies' cost of capital.
A crucial issue in multiples valuation is what criteria one should apply in peer group construction. De Franco, Hope, and Larocque (2015) find that selection criterion varies systematically with analyst' incentives and ability. Moreover, they also find evidence that analysts choose peer groups strategically. Bhojraj and Lee (2002), however, argue that the choice of comparable firms should be a function of the variables that drive cross-sectional variation in a given valuation multiple, independent of industry affiliation. However, evidence suggests that industry affiliation is important when selecting peer groups (Alford, 1992;Boatsman & Baskin, 1981;Tasker, 1998). In the spirit of Bhojraj and Lee (2002), we apply an empirical framework for establishing a relationship between valuation model and financial indicators in a particular industry, the oil and gas sector. Using 46 of the largest oil and gas companies, we investigate whether conventional peer groups (majors, independents, large exploration and production companies) constitute homogenous economic groups (i.e. similar relations between value drivers and valuation multiples). Using Chow tests, we test for differences in the valuation processes statistically by testing for structural shift in the value drivers across companies. Starting with a group consisting of the five super majors 1 (ExxonMobil, BP, Royal Dutch Shell, Total and Chevron), we test which other oil companies belong to this group by testing for a structural shift between this group and a potential super major (the largest among our sample of the 46 largest oil and gas companies). The null hypothesis is that the valuation model for the group of five super majors and the potential new super major is the same. If the hypothesis is rejected, this indicates that the potential super major should not be included in this peer group. This process is carried out for all the companies in the sample (less the original five super majors).
To control for the effects of unobserved variables, we apply panel data techniques, more specifically a fixed effects model. Omission of significant variables may lead to the omitted variables bias. One clear benefit of using a fixed effects model is that we can also capture the companies' cost of capital by including unobservable variables that are fixed for each firm in the sample across time. By also including fixed effects in the time dimension, we can also control for the impact of changes in oil and gas prices. Panel data models are often used in value-relevant studies in the oil and gas sector (Boone, 2002;Misund, Osmundsen, & Sikveland, 2015).
We contribute to the literature by introducing a novel approach to the selection of comparable firms. We apply a Chow test to assess whether the difference between two valuations processes are statistically significant. This contrasts the approach most commonly applied in prior studies, such as Bhojraj and Lee (2002) and Liu, Nissim, and Thomas (2002), focussing on the valuation accuracy without a procedure to assess how different to valuation processes must be to conclude that they belong to different peer groups.
Our study should also be of interest to investors and equity analysts valuing companies using the valuation multiples approach. Our results suggest that financial analysts and investors should select companies for peer groups on the basis of value-relevance of financial information, and not only based on subjective judgements.
The rest of the paper is organised as follows. The Section 2 presents a review of the literature. Section 3 describes the research design. Section 4 presents the data sample, and in Section 5, we present and discuss the results of the analysis. Section 6 concludes.
An emerging literature addresses criteria and the process for peer-group selection. Several studies have investigated the relation between levels of industry classification and homogeneity in firms' financial characteristics such as returns, valuation, risk and profitability. Boatsman and Baskin (1981) choose comparable firms from the same industry on the basis of fundamentals measured as historical earnings growth. This approach results in smaller valuation errors (using multiples) compared to randomly selected firms. Alford's (1992) study highlights the importance of industry in peer-group selection. He selects comparable firms based on fundamentals such as industry affiliation, size, leverage and earnings growth. The author finds that limiting the selection criteria to two-to three-digit SIC codes results in a reduction in valuation errors. The importance of industry-specific multiples is further emphasised by Tasker (1998), who finds a systematic use of industry-specific multiples among investment bankers and analysts in acquisition transactions. Bhojraj and Lee (2002) and Bhojraj, Lee, and Ng (2003) use a regression-based approach for selecting comparable firms independent of industry affiliation. The advantage of this approach that it allows to simultaneously control for the effect of several explanatory variables, and to empirically estimate the appropriate weights to put on each variable. They find that fundamental factors such as profitability, growth, and risk, are strongly associated with the enterprise value-to-sales and price-to-book ratios.
While Bhojraj and Lee (2002) and Bhojraj et al. (2003) advocate the use of objective criteria for selection of peer groups, recent evidence suggests that this is not always the case. De Franco et al.
(2015) examine the selection of peer companies by sell-side equity analysts. They find that analysts on average select peer companies with high valuations and that this effect varies systematically with analysts' incentives and ability. Moreover, their research suggests that analysts choose peers strategically.
In summary, the literature suggests that objective criteria based on valuation similarity should be applied when selecting peer groups. Moreover, prior research also suggests the importance of industry affiliation in peer-group selection.

Research design
It is possible to derive expressions for valuation multiples using traditional finance theory. The point of departure is Gordon's Dividend Discount Model (DDM). Bhojraj and Lee re-expresses the DDM model in terms of the PB ratio (based on the work of Feltham & Ohlson, 1995).
where P* t is the present value of expected dividends, B t is the book value of equity, r e is the cost of equity capital, and ROE is return on equity. This equation shows that a firm's price-to-book ratio is a function of its expected ROEs, its cost of capital, and its future growth rate in book value (B t+i-1 /B t ).
Equation (1) demonstrates the theoretical link between a valuation multiple and its value drivers. Ideally, this model should be at the centre stage of any selection of peer groups. Companies with similar structural relationships between value-drivers and valuation should be grouped together. The idea is that companies in the same peer group should be characterised by similar relation between valuation and value drivers. If they are not, we should be able to find a structural break in the valuation model.
However, there are some concerns about the appropriateness of model in Equation (1), especially for oil and gas companies. The price-to-book ratio is not a common multiple for valuing oil and gas companies. Doubts have been raised about the usefulness of historical cost measures for oil and gas companies (Financial Accounting Standards Board (FASB), 1982). The reasons that have been put forth are factors relating to the nature of oil and gas exploration and production activities (Wright & Gallun, 2005), choice of competing methods for accounting for oil and gas exploration activities (Bryant, 2003), 2 and the existence so-called "legacy assets" which are oil and gas producing assets that are completely depreciated, but still generate cash flows (Antill & Arnott, 2002). Hence, analysts and investors in the oil and gas sector use an alternative valuation multiple, enterprise value-to-reserves, the EV/R ratio.
Another problem with Equation (1) is that it potentially excludes additional explanatory variables that can affect the magnitude of the EV/R ratio, across companies and over time. Omission of explanatory variables that affect the left-hand side variable in a regression may result in the omitted variables bias, negatively impacting the inference we can make from the models. Typically, a set of control variables are included, and which act as proxies for unobserved explanatory variables. However, the selection of appropriate control variables is a very challenging task for the researcher and may not be successful. In fact, prior studies suggest that key performance ratios in the oil and gas sector such as the reserves replacement ratio is not significantly associated with valuation multiples (Osmundsen, Asche, Misund, & Mohn, 2006;Osmundsen, Mohn, Misund, & Asche, 2007) or returns (Kumar Bhaskaran & Sukumaran, 2016). An alternative to using explicit control variables is to apply panel data techniques, such as fixed effects. The benefit of using a fixed effects model is that the latter technique is designed to capture the impact on the left-hand side variable from unobserved variables. We therefore use the following empirical model where FE i and FE t represents time and firm fixed effects, respectively. The left hand side variable, EV/R, is enterprise value divided by the total amount of oil and gas reserves, measured in oil equivalents. Instead of using the return on equity, ROE, we use earnings before interest, taxes, depreciation and amortisation (EBITDA) divided by the amount of oil and gas reserves. This is in principle similar to value-relevance studies, where the accounting variables are often scaled by the amount of oil and gas reserves (Misund, Asche, & Osmundsen, 2008).
The type of specification in Equation (2) assumes that the relationship is stable, i.e., the estimated parameters are constant over the sample. This implication allows us to test for structural shifts in the relation between valuation and value drivers.
If there are two different peer groups, there will be two different parametric specifications of the relationship between value-drivers and valuation in the sample: Peer group 1:

Peer group 2:
If the coefficients in the two equations are statistically different from each other, this provides evidence for a structural break in the econometric modelling of multiples valuation (see e.g. Chow, 1960). Hence, structural break tests can be applied to examine whether the valuation process changes when extending the group of peers. We test for structural breaks using the dummy variable approach (Gujarati, 1970a(Gujarati, , 1970b, which allows us to run a single regression instead of two, which would be the case for a Chow test (Chow, 1960). Gujarati asserts that the dummy variable method is preferable to the Chow test for several reasons. First, running only a single regression can substantially abridge the analyses. Second, the single regression can be used to test a variety of hypotheses. Third, the Chow test does not explicitly indicate which coefficient, intercept or slope is different. Fourth, pooling increases the degrees of freedom and may improve the relative precision of the estimated parameters.
Using the dummy variable approach and allowing for a structural change, Equations (3) and (4) can be combined and written as follows: where C2 is a dummy variable that is zero for company 1 and 0 for company 2. The variable 4 it represents the error term. We test for structural break in the model by testing for joint significance of the interaction terms using a Wald test. That is, one tests if the hypothesis that the interaction terms are jointly significantly equal to 0 (i.e. H 0 : β 2 = γ 1 = γ 2 = … = 0) can be rejected at a specific level of significance. If the null hypothesis is rejected, then the results provide evidence for a structural break in the econometric modelling of valuation.

Data
The sample consists of oil and gas companies for the 1992-2013 period drawn from John S. Herold Company's (JS Herold) oil and gas financial database. 3 The Herold database consists of more than 500 publicly traded energy companies. From this universe, we select the 50 of the largest oil and gas companies that report both financials and supplementary information in accordance with the U.S. (2) Securities and Exchange Commission's (SEC) regulation. 4 For four of these companies, we had fewer than 3 years of data, and these were excluded from the final sample of 46 firms. The descriptive statistics are presented in Table 1. Table 2 presents IHS Herold's classification of the largest North American and international oil and gas companies. We use Herold's selection of oil and gas majors as our benchmark sample. The aim of the analysis is to examine whether we can expand this initial group of companies by adding additional firms if they are significantly similar.
Since the data covers a time period of more than 20 years, both autocorrelation and heteroskedasticity may be present in the data, negatively affecting the inference we are able to make from the results. We therefore test for heteroskedasticity using the Breusch-Pagan test (Breusch & Pagan, 1979) and serial correlation using the Breusch-Godfrey test (Breusch, 1978;Godfrey, 1978). If we find evidence of either serial correlation or heteroskedasticity, or both, we need to adjust the standard errors before calculating the t-values and p-values from the regression. Heteroskedasticity can be corrected for using the White (1980) approach and serial correlation can be corrected using the Arrelano method for fixed effects models (Arellano, 1987).

Table 1. Descriptive statistics
Notes: EV/R is the enterprise-to-total oil and gas reserves ratio and EBITDA is Earnings before interest, taxes, depreciation, and amortisation (million USD), scaled by the amount of oil and gas reserves (in millions of barrels of oil equivalent).

Results and discussion
The analysis is carried out as follows. First, we produce an empirical model of the relationship between price-to-book and its value drivers for a subset of five Super Major oil companies. All other companies will be compared to this particular group. Second, we introduce firms classified as international majors, one by one. Chow test is used to investigate whether the new company has a significantly different relationship between valuation and financial indicators than the five original super majors. Finally, we investigate whether firms classified as United States and Canadian E&Ps can be included in the super major peer group. We do this by repeating the second step with United States & Canadian E&Ps instead if the international large companies.

Part 1: The relationship between price-book and financial indicators for oil super majors
First, we carry out tests to see whether we should use a pooled OLS or a fixed effects model (pooling test) and whether a fixed effects or a random effects model is appropriate (Hausman test: Hausman, 1978). The tests conclude that a fixed effects model is the most appropriate for our data (Table 3). Secondly, we test for heteroskedasticity and serial correlation in the residuals from the empirical estimation of the model in Equation (5) using the initial subsample of oil and gas super majors. We cannot find evidence of neither heteroskedasticity, nor serial correlation (Table 4) and we do not need to correct our standard errors. Finally, we estimate the model in Equation (5) and the results are presented in Table 5.
The coefficient on the profitability variable is significant (Table 5), which provides evidence that EBITDA is a relevant profitability measure for the oil and gas majors. Moreover, the difference in the two adjusted R 2 measures suggest that the fixed effects, both for time and individuals, capture the effects from unobserved variables.   Next, we include new companies to the Super Major group, one by one, using an extended sample. Significance of the joint interaction terms indicates that this new company belongs in the Super Major group.
Our results suggest that several of the oil and gas firms (e.g. ENI and MOL) belonging to peer groups other than "oil majors", are more closely associated with the super majors, than with other firms in the group they have been added to. The implication of our study is that the oil major peer group could benefit from adding other companies, such as ENI. Arguably, a larger peer group would improve the accuracy of the multiples valuation method.  Part 2: North American large exploration and production companies Similar to the analysis of integrated companies (Part 1), we also find that several of the companies that are typically characterised as North American large E&Ps are more closely related to oil majors than to the other companies in their peer group.
In summary, our results suggest that the approach used in the current study can be used to in the selection of companies to be included in peer groups for the purpose of equity valuation using multiples. The results should be of interest to investors and equity analysts covering the oil and gas sector, as well as other industries.

Conclusion
The Chow test for structural shift is a methodology that can be used to identify peer groups that have similar structures in their valuation process. Applying the test to 46 oil and gas companies, we find that several companies, both among the largest international integrated companies, as well as among the largest North American E&Ps have a similar structure in their valuation process to the oil super majors. Moreover, our findings suggest that investors, taking into account profitability and several unobserved factors, value several of the largest E&Ps in the same way as they value majors, suggesting the pricing of the latter securities are perhaps more efficient than several of the international integrated companies.
We do not find that other groups of firms have a structurally similar valuation process. This means that comparison of firms in groups such as independents and internationals are likely to result in large variation in the companies' perceived performance since the measures show the differences in the valuation process rather than the differences in economic performance.