Proposal for Discussion Purposes Only Title: Are we better off articulating Fair Value with Historic Cost? Impact on Earnings Attributes Title: Fair Value Precision Score: a civil trial Note: This paper has benefitted by the helpful comments and suggestions of Trevor Harris, Sharon Katz, Nahum Melumad, James Ohlson, Gaizka Ormazabal, Stephen Penman, Gil Sadka, Gary Finnis (Sproule), Javier Martinez de Olcoz (Morgan Stanley), Columbia PhD Accounting students and the IESE Accounting Department . I greatly appreciate Canoils and the Toronto Stock Exchange for giving me access to their unique databases. I also thank David Elliot, Carrie Nermo and Brian Banderk, from the Alberta Securities Commission, for supplying me with valuable data as well as for their insightful discussions. I gratefully acknowledge that this project has been supported by the Columbia University CIBER. Also, Elie Toubiana provided exceptional research assistance. All remaining errors are mine. Abstract Prior studies compare the value relevance of fair value versus historical cost. However, these studies ignore the cross-sectional variation in the attributes of fair value and historical cost estimates, as well as the accounting articulation between them. The reason is that fair value and historical cost are seldom reported together and firms do not disclose the precision and numerous assumptions of fair value estimates. I have found a unique setting where firms release both, fair value and historical cost estimates. Canadian Oil and Gas (O&G) producers have to disclose the historic cost of the O&G reserves and the probabilistic fair value estimates alongside with the reconciliation of annual changes in this estimates. Hence, I develop a Fair Value Precision Score based on the probabilistic reversals, using the annual reconciliations and industry key performance indicators. I also build a score measure of the historic cost precision using firm-level measures of accruals quality (Dechow and Dichev, 2002; Wysocki, 2009) and the Balance Sheet firm-level measure of conservatism (Beaver and Ryan, 2000). I conclude that portfolios with a higher score, i.e. firms with greater precision of fair value estimates, (or positive changes in the score) are more value relevant and timely than those with lower score. That relevance increases due to the interactive effect of the fair value precision score with fair value and historic estimates. This interactive effect increases in years when the industry uncertainty is lower. Lastly, I show the dynamics of the synergies between both estimates. For the same level of fair value precision, portfolios with higher accruals quality and lower Balance Sheet conservatism have higher value releance. Also, I find that when historic cost or fair value estimates are interacted with the multiplicative effect of the precision scores, those coefficients are statistically significant and the valuation relevance and timeliness increases, suggesting that fair value precision sheds light on the historic cost precision, and vice versa, and that accounting articulation is impounded in the investors’ consensus beliefs. Finally, given a level of historic cost precision, the fair value precision has also an incremental impact on other earnings attributes: for the same historic cost precision, portfolios with higher fair value score have higher predictability and lower contemporaneous earnings-returns relation (Sadka and Sadka, 2008), because investors can predict better the expected earnings persistence and growth with more precise fair value estimates. Likewise, the Balance Sheet conservatism decreases with the fair value score, controlling for the level of 1 Proposal for Discussion Purposes Only accruals quality. This proves that investors understand that the long-term accruals precision is related to the precision of the fair value estimates. I. Introduction Since the approval in 2006 of the FAS No. 157, Fair Value Measurements (FAS 157), we have witnessed a tense debate in the research and investment community on the fair value accounting and the multiple ways to estimate it (the so-called 3-Level fair value hierarchy)1. On the one hand, advocates of fair value reporting argue that it is more relevant to investors than historical cost accounting. On the other, opponents of fair value reporting criticize its lack of reliability due to measurement errors and biases (especially in Level 3), its stock market induced volatility (especially in Levels 1 and 2), and its unintended effects on real activities. As per Nissim and Penman (2008) “the issue of when, rather than how, to apply fair value measurements – as a matter of principle – is unresolved”. Indeed, outstanding literature on fair value takes the existing fair value measurements as given (i.e. the how), and explores the merits of these measurements in different settings (i.e. the when) versus the historic cost alternative. If it is true that current fair value measurements present serious limitations, and historic cost have also shortcomings (Ryan, 2008), an interesting question would be whether we can articulate both estimates and introduce innovations in their measurements to overcome some of their shortcomings. As Einstein warned “we can't solve problems by using the same kind of thinking we used when we created them”. This articulation is not just disclosing both of them without mixing or loosing information, but trying to get something else from the accounting structural interrelation. The obvious challenge is the impossibility of testing empirically new fair value estimation methods that have not been implemented yet. Long time ago, Graham and Dodd2 wrote about the essential nature of the earnings problem, citing Oil and Gas firms. His point is that we need to assess the predictability and other earnings attributes, since they are not the same across firms: Most important of all, the analyst must recognize that the value of a particular kind of data varies greatly with the type of enterprise which is being studied. The five-year record of gross or net earnings of a railroad or a large chain-store enterprise may afford, if not a conclusive, at least a reasonably sound basis for measuring the safety of the senior issues and the attractiveness of the common shares. But the same statistics supplied by one of the smaller oil-producing companies may well prove more deceptive than useful, since they are chiefly the resultant of two factors, viz., price received and 1 Those 3 levels reflect the amount of judgment involved in estimating fair values. Level 1 reflects valuations based on observable inputs consisting of quoted prices in active markets for identical assets or liabilities. Level 2 reflects observable inputs other than quoted prices. Level 3 is the so-called mark-tomodel estimations. 2 As cited by Dechow et al. (2009) 2 Proposal for Discussion Purposes Only production, both of which are likely to be radically different in the future than in the past. (p. 33/34). But despite this old advice, many prior studies compare the value relevance of fair value versus historical cost, as if they were homogeneous concepts with not cross-sectional variation in its application, and also as if they were irreconcilable enemies. Also, these studies do not examine how the precision of fair value measures, the earnings attributes and their mutual interrelation impact on the value relevance and other earnings attributes, namely predictability, accruals quality and earnings-returns relation. The reason is that fair value and historical cost are seldom reported together and firms do not disclose the precision of fair value estimates. I have found a unique setting where firms release both, fair value and historical cost estimates. Canadian Oil and Gas (O&G) producers have to disclose the historic cost of the O&G reserves and the probabilistic fair value estimates alongside with the reconciliation of annual changes in this estimates. In particular, I focus on the Canadian O&G producers for two reasons: first, Canada is the world’s largest O&G public markets by number of issuers (398 vs. 303 combining NYSE and NASDAQ, in 2010), and O&G listings (24 vs 7 in 2009). Second, and more importantly, in 2003 a regulatory change required Canadian issuers3 with O&G activities to break down the quantities (barrels) and values of O&G reserves estimates into at least the two first of these three categories: proved, probable and possible, following explicit numeric cumulative probabilities of recovery and future O&G prices4. Additionally, it required annual disclosures of net reserves reconciliations (in barrels), the single most powerful tool for tracking changes in the value of O&G reserves estimates. Apart from production, additions and disposals, reserves can fluctuate because of economic changes and technical revisions. Changing economic conditions may result in either additions or reductions of reserves as extraction becomes more or less economically viable 5. Technical revisions occur due to estimation procedures, moving reserves from one classification to another, obtaining new technical information (e.g. information coming from production), and poor geological and engineering reserves estimation practices. Technical revisions, after controlling for relevant news, are an important proxy for measurement errors and biases. They provide a measure of the precision of reserve estimates. Anecdotal evidence, such as Royal Dutch/Shell’s big restatement of 20% of proved reserves in 2004 and other companies’ material restatements such as El Paso, Stone Energy, Repsol YPF, etc., suggests that this area is a minefield for investors, increasing litigation costs for reserve related class-action suits (such as $5Bn suit against Shell) and reducing investors’ confidence (Grant T. Olsen et al, 2010). Effective July 18, 2003, the Canadian Securities Administrators (CSA) published “Standards of Disclosure for Oil and Gas Activities” under National Instrument 51-101. Those standards were amended in 2007 and 2010, and are accompanied by the official technical reference, Canadian Oil And Gas Evaluation Handbook, (COGEH). 3 4 Companies are enforced to use expected future (forecasted) prices as valuation inputs, but may optionally disclose another value estimate, using current (spot) prices as valuation inputs. The former refers to the value-in-use approach and the latter is close to the exit value principle. 5 Economic factors not only change revenues and costs, but also affect royalties and income taxes (David Elliot, 2005). 3 Proposal for Discussion Purposes Only Articulation between historic cost and fair value: only value relevance? The accounting articulation between historic cost and fair value estimates, as well as the cross-sectional variation in the precision of both estimates may have an impact on the value relevance of both estimates. If the value relevance is incremental on the interaction between the assessment of the quality of both fair value and historic cost estimates is an empirical question. But before looking at value relevance, it is pertinent to analyze whether the value relevance is the relevant criterion to follow. As in Barth (2000) and Penman (2009) I follow the “basic ownership approach” proposed, among other approaches, by the FASB and IASB6. Holthausen and Watts (2000) criticize that the association criterion has no theory of accounting or standard setting supporting, raising questions regarding the appropriate inferences that can be drawn from this literature and the ability of the literature to inform standard setting. As he criticizes “the incremental association program would end up with a book value of equity that is a transformation of the market value of equity”. That criticism is exacerbated when the fair value information is mixed inappropriately with the historic costs transactions, making the balance sheet uninformative (Penman, 2009), so higher association can be got with lower usefulness or other criteria we may choose, such as accuracy, or even with self-reference market values, or the loosing of relevant information. Also, Penman (2011) states that “the interpretation of observed correlations can be made within a (regression) framework that incorporates all known structural relationships”. From those fair criticisms it is concluded that association is not the goal of accounting, but the outcome after taking care of other needed accounting information attributes: such as reliability, predictability, timely loss recognition, etc. Those attributes are not measures of the same underlying construct, but rather they measure different elements of decision usefulness (Dechow et al. 2009), so the effect of fair value on them does not need to be the same. Also, I incorporate Penman’s advice since I link the value relevance through accounting structural relation between fair value and historic cost estimates in the O&G industry. Notwithstanding, in our setting, since information is not lost (fair value gives incremental not alternative information), even the association test can be a good measure of the information content of the disclosures. Higher association can be got either with incremental information that is independent or dependent (in the sense of information that does or does not update the existing information), so a trade-off is no apparent. Investors can always choose the information they need, so if investors use accounting information as inputs for the equity valuation our conclusions are valid. If investors use information for other purposes, our conclusions are still valid but need to be applied to the new construct. For instance, fair value disclosures can help investors assess the timeliness or predictability of the earnings, etc. In any case, there is not a definitive answer, because even when giving new information, that information can misled investors, and giving bubbles or temporary deviations from fundamentals, give a high association. That’s why I try to capture association, but I have linked it also with reliability and other earnings attributes. So, for instance, in the hypothetical case that I would have had found nonreliable fair value disclosures with higher value relevance, we could use that information 6 See Financial Accounting Standards Board, Preliminary Views, Financial Instruments with Characteristics of Equity (Norwalk, Conn.: FASB, November 2007) and a Discussion Paper from the International Accounting Standards Board with the same title, dated February 2008. 4 Proposal for Discussion Purposes Only to beat the market, concluding that the information can misled the market. The main takeaway is that earnings and fair value are interrelated, and need to be articulated to improve the reliability and relevance of accounting information. Just mixing both information can have unintended consequences. Just disregarding the fair value usefulness is a step back. We need to think in ways of articulating, (not mixing) both disclosures to get better information for the multiple purposes that the financial statement serves. Fair Value Precision Score Despite the potential relevance of the fair value, it goes without saying that the precision can be different across companies and contexts7. Even For the purpose of testing for the articulation between fair value and historic cost estimates, I examine the reliability of fair value estimates and their cross-sectional variation. To that end, I develop a Fair Value Precision Score based on the probabilistic reversals, using the annual reconciliations and industry key performance indicators. I also build a score measure of the historic cost precision using firm-level measures of accruals quality (Dechow and Dichev, 2002; Wysocki, 2009) and the Balance Sheet firm-level measure of conservatism (Beaver and Ryan, 2000). Then, through a two-stage approach, I model the one-period-ahead score and then I regress a la Fama-MacBeth the Market capitalization on Book Value, Earnings, different measures of Fair Value disclosures and interaction effects of these variables with the predicted FV and Historic Cost Precision scores, a firm-level uncertainty composite proxy and a macro (industry) uncertainty proxy, following the spirit of Feltham-Ohlson Model (1995, 1996). Consistent with a rational investor framework, I find that investors give more weight to fair value for companies that have higher FV score and lower firmlevel uncertainty. That weight increases in years when the macro uncertainty is lower. Lastly, I use the same methodology to analyze the changes in FV (with changes in raw returns and market-adjusted returns) related to the changes at levels of information quality. First, I compare the FV changes with the Earnings changes, to study the value relevance of both disclosures, Second, using the annual reserves reconciliation to see the incremental information content of the components (aggregating technical revisions with economic changes, like in the US) versus the net changes, and the former versus the information content of the components given the FV and Historic Cost Scores and the contextual information. I find that the value relevance of FV changes is incremental to that of the historic cost disclosures. Also, the value relevance varies across components, and each component value relevance decreases asymmetrically across firms with the level of the Score. In all these stages I control for information risk (scaled range of the distribution, i.e. probable over proved+ probable) to try to rule out as much as possible the effect of the Even the savvy investor Warren Buffet while stating in his classical Berkshire’s 2008 letter to shareholders, “we endorse mark-to-market accounting”, at the same time was pointing out in the same letter the measurement errors that happen with long-dated option valuation 7 5 Proposal for Discussion Purposes Only firm underlying uncertainty on the earnings attributes. If I would fail to control for it, the fundamental uncertainty would be related to my FV precision score, so I would not be able to conclude with reasonable certainty that the differential precision of the disclosures are driving the result, but it could be just the differential uncertainty. Although the latter would be a novel way of measuring uncertainty surrounding the firm, it is not the intended main conclusion of the paper. Finally, I use the Fair Value Precision Score to understand the impact on other earnings attributes: namely, accruals quality, conservatism, predictability and earnings-returns relation. Following Sadka and Sadka, (2009) I predict that increasing the fair value precision would enhance the earnings predictability. That increase would in turn cause to reduce the contemporaneous earnings-returns relation. I show that portfolios with a higher score, i.e. firms with greater precision of fair value estimates, (or positive changes in the score) are more value relevant and timely than those with lower score. That relevance increases due to the interactive effect of the fair value precision score with fair value and historic estimates. This interactive effect increases in years when the industry uncertainty is lower. Lastly, I show the dynamics of the synergies between both estimates. For the same level of fair value precision, portfolios with higher accruals quality and lower Balance Sheet conservatism have higher value relevance. Also, I find that when historic cost or fair value estimates are interacted with the multiplicative effect of the precision scores, those coefficients are statistically significant and the valuation relevance and timeliness increases, suggesting that fair value precision sheds light on the historic cost precision, and vice versa, and that accounting articulation is impounded in the investors’ consensus beliefs. Finally, given a level of historic cost precision, the fair value precision has also an incremental impact on other earnings attributes: for the same historic cost precision, portfolios with higher fair value score have lower earnings-returns relation, because investors can assess better the expected earnings persistence, since the more precise fair value estimates enhances the earnings predictability. Similarly, the Balance Sheet conservatism decreases with the fair value score, controlling for the level of accruals quality. Investors understand that the main long term accruals are related to the precision of the fair value estimates. Links to Literature This study is closely related to prior research that investigates the relative dominance of fair value vs. historical costs in the investors’ valuation, as well as the effects of measurement errors and biases. Prior research finds only a weak association between security prices and oil valuation disclosures required by SFAS 69 (e.g., Harris and Ohlson, 1987, 1990; Magliolo, 1986). Three plausible reasons might explain these results: unreliable estimations of reserves quantities, flaws in the valuation model, and model misspecifications (Barth, 1994; Clinch and Magliolo, 1992; Boone, 2002; Kim, 2011). Concerning fair value, some of the studies are Nissim and Penman (2008), Ryan (2008), Bath and Taylor (2008), among others. Lastly, this study is related to valuation and fundamental analysis studies, with a long research literature with sound references as 6 Proposal for Discussion Purposes Only Penman (1989), Ohlson (2009), Nissim and Penman (2001), Nissim, Liu et al. (2002), Nissim (2009) and a long etcetera. The most similar study is performed by Boone (2002). He stated, “in contrast to these firm-specific operating assets, oil and gas assets trade in active markets; these assets have little firm-specific value, and thus their fair values should be accurately estimable”. He finds evidence that suggests that model misspecification, rather than measurement error or time-period idiosyncrasy, most likely explains the weak value relevance of oil and gas present values reported in prior research. Although his assertion that the oil and gas assets have little firm-specific value sounds prima facie reasonable (O&G assets can be easily sold, and the O&G market is worldwide), his conclusion is flawed: we cannot say that fair values can be accurately estimable for two reasons. First, some people tend to think of the O&G market as a global end-user market, but there is evidence that this is a simplification, even more nowadays: big discoveries of shale gas in the US and oil sands in Canada, deep water drilling new techniques, specific geographic shocks such as the Gulf of Mexico moratorium, the increasing spread between the Brent and the WTI references due to the US glut supply, and the decoupling between the gas and oil prices make us think that the market is not moving in the same direction for all the participants, conditional to mix of products drilled, geography focus, technology expertise, etc. Second, O&G companies values reserves, i.e. O&G to be extracted, so the value reference is not related to the O&G liquid markets, but to M&A activities, asset deals or stock market transactions, which are exposed to the same equity and financing shocks of other equities. Both reasons point to the same conclusion: O&G reserves are forwardlooking valuations with a lot of forecasted parameters. Thus, despite the supposedly little firm-specific value we still have big uncertainties on the parameters, and all the firms have different skills that attach different values to this uncertainty. As anecdotal evidence that even in geographically close wells the valuation is uncertain, WSJ reports at January 28, 2011: Exxon expensed hundreds of millions of dollars of reserves because of the write-off of three expensive deepwater wells that turned into dry-holes, off the coast of Brazil, very close to where Petrobras drilled and that resulted into the industry’s biggest oil wells discovered in past decades8. Along the same lines, he hypothesizes that the effect of the variance of the oil and gas prices in the discount rate is a partial cause of the measurement errors. However, although the discount rate is the most sensitive parameter in a DCF approach, he neglects the possibility that, as per Damodaran (2009), the optionality9 embedded in the non producing reserves account for the differences, since the volatility, interests and dividends are the only forecasted parameters in an option. This study is different from Boone (2002) and prior studies in three respects. First, this study directly uses accounting information to capture measurement errors and biases and build a fair value score to measure the fair value quality, and it does not need to rely on 8 Although wells in nearby geographical areas could be exposed to similar political, taxes and economic factors, to be geographically nearby does not mean it is geologically an analog. 9 This is because of the time premium on options embedded in the undeveloped or out-of-the-money reserves 7 Proposal for Discussion Purposes Only strong econometric assumptions about the structure of the errors. Second, this study focuses on a sound alternative to fair value with just one single estimate, and study whether investors can derive the measurement errors in a risk-reward scheme. Finally, this study incorporates a fundamental analysis approach to see whether KPIs help give context to fair value estimations, alleviating the measurement error and biases errors. To my knowledge, this is the first study that tries to capture the measurement errors and biases using fair value with different probability categories and annual technical reconciliations. http://findarticles.com/p/articles/mi_qa5447/is_199910/ai_n21449409/pg_4/?tag=content ;col1 Main takeaways This study has several implications for the literature on fundamental analysis, earnings quality, fair value and asset pricing. First, it builds a tool for looking at the precision of the fair value estimates, looking at probabilistic fair value reversals and industry KPIs, adding a new role of fundamental analysis information, i.e. giving context and articulating the valuation inputs. Second, it links that Fair Value precision score with the earnings and fair value attributes, offering an omitted relevant variable neglected in the literature of earnings quality, especially for long term accruals. Third, it points out the role of the precision of the fair value in increasing the predictability of earnings, and in turn decreasing the contemporaneous earnings-returns relation. That evidence is consistent with Sadka and Sadka (2009), but extends their intuition at the firm-level, in a different setting and also providing different paths for which that relation holds. Fourth, it adds to the fair value debate, bringing some empirical evidence on the potential role of the fair value with probabilistic disclosures and annual reconciliations and the accounting structural articulation with the historical cost estimates. Specifically, this evidence may help to the IASB-FASB10 discussion on fair values with disclosures probabilities and annual reconciliations. Without stretching the empirical evidence, this study’s empirical evidence endorses the articulation of fair value and historic cost in a Fair Value Statement, taking into consideration other relevant criteria of decision, such as costbenefits, plausible standardization, litigation costs, etc. Perhaps this new statement could be firstly introduced in some specific sectors in which it might have a bigger impact, such as real estate companies, mining firms, etc. i.e. for capital intensive industries with long lead time on projects, and/or strongly dependent on commodity prices. Analogously as in the O&G industry, in these industries, the information flowing to the income statement is delayed: either is frequently impacted by "ceiling test" charges related to fluctuations in commodity prices, or it does not take into consideration differences in timing of growth projects, so investors need some track of the unrealized value created. Also, the assets In June 2010 the IASB published Exposure Draft 2010/7, which proposed a “measurement uncertainty analysis disclosure for fair value measurement categorized within Level 3 of the fair value hierarchy (i.e. fair value measurements using significant unobservable inputs). The objective of that disclosure was to provide users of financial statements with information about the measurement uncertainty inherent in fair value measurements categorized within Level 3 of the fair value hierarchy, given that those fair value measurements are more subjective than those derived from observable market prices” 10 8 Proposal for Discussion Purposes Only can be seen as independent each other, so single valuations add up to the total. As a consequence, reported net income and, hence P/E ratios, are of lower utility in the valuation of these companies11. I claim fair value estimates with probability disclosures and annual reconciliations is not a fancy way of masking the current reality through biases or measurement errors, but a sound alternative to give information about value and shed light and improve the earnings and historic cost attributes. By its very nature, value is forward-looking and uncertain. Investors and stakeholders have different horizons and value companies for different purposes, based on different weight on the historic cost and fair value estimates and on information extracted from their accounting articulation. Hence, providing multiple value estimates externally audited and input disclosures, and also articulating these fair value estimates with the other Financial Statements can be the right approach to serve for all users’ purposes (e.g. historical costs contracting approach, value in use, exit value, etc.). The paper proceeds as follows. In Section 2 I discuss the accounting structural articulation of the fair value and the historic cost estimates. Section 3 provides some accounting knowledge on the Canadian O&G industry. The research hypotheses are presented in Section 4. Then, in Section 5, I describe the sample and the methodology. Section 6 discusses the results of the main test and the contextual analysis. Section 7 concludes and suggests future related research. II. Fair Value and Historic Cost. The accounting articulation: It is well known that both fair value and historic costs use forward-looking information. The asymmetry on the treatment between good and bad news is a common place (Basu, 1997). But it is worthwhile to stress that historic cost transactions recognize upfront not only forward-looking negative news (write-offs) but also forward-looking positive news (reversals). Also, although the positive news have a limit, i.e. the level of the previous write-offs, that limit would depend on when exactly the transaction occurred, so if there had been temporary divergences from the true value (as it happened in the real estate bubble), the limit might be set artificially high, and that asymmetry could even vanish. In any case, contingent to the past information, accruals also can accommodate forwardlooking positive news, via reversals. Along the same lines, although unrealized news beyond that limit is not recognized until its realization, the recognition can be manipulated via selective realizations, distorting the accounting via real transactions (Ryan, 2008). As per JP Morgan Analyst Report (April 17, 2008) “EPS and CFPS growth do not tell the whole story as 1) they do not reflect long-term capital efficiency, 2) they are strongly dependent on commodity prices, which makes us reluctant to use it as a primary metric of success and 3) they do not take into account differences in timing of growth projects. In an industry with long lead times on projects, we think a focus on near-term EPS growth might be detrimental to investment decisions and thus to longer-term growth”. Nevertheless, P/E ratios are still monitored by many investors. The major, non-cash expense item most companies bear is the depletion rate per unit of production. Since this charge is directly related to prior years' reserve replacement costs, companies which are able to add reserves more cost-effectively tend to report relatively more income than others, regardless of the commodity price environment. 11 9 Proposal for Discussion Purposes Only The bottom line is that historic cost transactions also need to carry out estimates, and those estimates are not only for negative news. These are the well known accruals. Accruals are defined in the industry jargon as the earnings that have not been realized (i.e. earnings recognized in the current year but whose cash flows are recognized in the future). In the accounting literature, accruals has been preferred to be defined as the mapping between operating cash flows and operating earnings, equating them to working capital movements, mixing accruals from several years. This accrual concept helps to understand the compounded measurement errors and to operationalize the construct, but it does not help to understand the expected accruals quality for a specific earnings period, since it requires having an understanding of the true underlying economics (Dechow et al. 2009). Also this definition is not that operative for industries in which the fixed assets play an important role. For these industries, the link between cash flows and the main accruals (depreciation, impairments, long term provisions) is not as important (because of the timing disconnect) as the assessment of the precision of the components of the earnings. That precision involves estimates of the future cash flows (accrued principle, such as long term provisions), but it also involves already disbursed cash flows, (i.e. depreciation and impairment). The latter are estimates that involve a judgment about the link or allocation of those future cash flows with the revenues, and the amount to be recognized or allocated, based on recoverability (i.e. matching principle, such as depreciation and write- downs). It also involves substantial measurement discretion, and the reversal occurs gradually over many subsequent periods through reduced depreciation or amortization expense. (Nissim and Melumad, 2008). The fair value estimates might give us a signal of the precision with which earnings measures the future cash flows as well as the precision with which earnings are related to the performance of this specific period, not mixing years’ performances. Without fair value estimates, the disclosures for assessing the reasonability of the depreciation, impairments, provisions, etc. are very poor, and although there are some red flags (Nissim and Melumad, 2008) there is not a definitive answer about how to handle those forward-looking estimates. Graham12 already warned about that paradoxical firm behavior of recognizing all the future earnings at once (conservatism), and then stating that they were a transitory component of earnings, leaving those future negative cash flows in an unorthodox limbo. Fair value can help to understand better those impairments and avoiding that an excessive fixation in earnings could make investors get confused13. Since historic cost and fair value estimates are forward looking, those estimates need to be congruent, since they rely on the same projections of future cash flows. So, if the financial statements articulate with the fair value disclosures, and the fair value estimates have enough reliability, fair value and historic cost disclosures may be reciprocally relevant. This approach is in the same line of reasoning of Penman (2009), in which he endorses the proposal that financial statement presentation should exploit the cohesion between financial statements and articulate. The articulation between fair value 12 The Inteligent Investor (page X chapter 12) Although that information is captured through Book Value, following the Ohlson Model logic, a not systematic valuation of the information could induce to think that the forthcoming periods have got a better than expected performance, and that the impairment period was a one-off, potentially misleading the future projections (for instance, future capex and depreciation would not be that persistently small). 13 10 Proposal for Discussion Purposes Only disclosures and the financial statements is not as clean, congruent and comprehensive as it could be. Even in the Canadian O&G industry, despite being one of the most advanced regulations in terms of fair value disclosures, the precision of the articulation is far from perfect, and is unequal across firms. But I extend Penman’s reasoning claiming that this still imperfect articulation goes in the right direction and may convey important information. Analogous reasoning can be found in Nissim and Melumad (2008), in which they posit that there is a need for more fundamental and contextual research. The role of key performance indicators (KPIs) and Fair Value disclosures helps linking both advices. As it is known, the O&G industry is very intensive in the use of KPIs, such as reserve replacement ratios, reserve life index, finding and development costs, netbacks, BOEs, etc. The use of industry-specific KPIs for company valuation is also a promising research avenue in the field of fundamental analysis and fair value precision assessment. An excerpt of the report of the SEC Advisory Committee on Improvements to Financial Reporting (2008) confirms the concern of regulators on this issue: “we are also encouraging the private sector to develop key performance indicators (KPIs), on an activity and industry basis, that would capture important aspects of a company's activities that may not be fully reflected in its financial statements or may be nonfinancial measures.” Most academic research analyzes the information content of general valuation multiples and ratios for earnings forecasting and equity valuation. Yet, anecdotal evidence from analyst reports suggests that practitioners use multiples and ratios that are often specific to each industry. These measures try to capture the value drivers of each particular business, linking fair value and historic cost estimates to articulate them and extract further information. Very often, these magnitudes are expressed in physical units rather than monetary units. My aim is to use these industry-specific KPIs to provide incremental context to the fair value measures in addition to the common multiples and ratios, improving the significance and relevance of the fair value disclosure and linking more precisely those estimates with those of the historic cost. As additional evidence of the importance of these KPIs, The Alberta Securities Commission, the Canadian regulator, annually analyzes the technical revisions (KPI with a physical magnitude, i.e. barrels) and conducts reviews of those companies with either big or negative technical revisions. Let’s now examine how the structural accounting relation between historic cost and fair value estimates can impact in each other. First, the impact of fair value on the earnings attributes. If we rely on the fair value estimates and disclosures, we can assess better the expected persistence (predicted sustainability) of the earnings. That predictability can help us to assess better the future expected earnings and the expected persistence of the earnings shocks (Sadka and Sadka, 2008). This predictability is more valuable and acute when the O&G prices have big swings. For instance, under big swings in O&G prices, it is more difficult to assess the reasonability of the depreciation and impairment components, given it depends on our assessment of the future cash flows. Moreover, we would expect that the conservatism impact could diminish, since the asymmetric timeliness of earnings and losses could be corrected by the forward looking fair value estimates. As a consequence, we would expect that if we trust the earnings and we know 11 Proposal for Discussion Purposes Only better their expected persistence, i.e. the predictability of earnings increase, their impact on prices would be lower, decreasing the changes in earnings-returns relation. Second, the impact of historic cost on the assessment of the reliability of fair value estimates. With analogous reasoning, if we trust on the earnings estimates, we could assess the precision of fair value estimates looking at past historic transactions (lifting cost, asset retirement obligations, interest rates, average age of reserves). It is difficult to disentangle the partial effects of both estimates, but the combination of the two helps each other, so if fair value and historic cost estimates articulates and are reliable, they benefit each other, improving earnings and fair value attributes. Fair Value Criticisms and some counterarguments The fair value has received fair criticisms. First of all, it goes without saying that fair value main criticism has been its lack of reliability. In Canada, the fair value disclosures are mandatorily externally audited, and there is an extensive disclosure of the assumptions and annual reconciliation, so whether this criticism is valid is an open question. Also, since historic cost also use estimates, and in many occasions those estimates are not disclosed, the criticism applies to both estimates. The difference is then between an explicit crystal ball versus a black box, but both are forward looking in essence. It has also been said that fair value, when mixed with the financial statements, leaves uninformative the income statement, and makes difficult to redeem the balance sheet changes. On the contrary, under constant premium assumption, we would just need the income statement (Penman, 2009). This argument is appealing for its simplicity, (the so-called balance sheet error cancelling approach), but as it is explicitly admitted, under growth assumptions the balance sheet would not need to have a constant premium (even without growth, since the discount news can exist, changing the balance sheet premium). Adding insult to injury, another criticism states that even in the absence of fair value measurement errors, the information would be irrelevant (Penman, 2009), digging in a similar argument drawn by Holthausen and Watts (2001). But it could be argued that those theoretical assumptions, the income statement would reveal the realized rate of return, so the income statement would be disclosing the expected (required) rate of return and by difference the changes in the future expectations of cash flows and discount rates (Campbell, 1991), becoming informative, since those cash flows and discount rates news give information. Similarly, the creation of value cannot be measured through the historic cost earnings deducted the required rate of return on Book Value. This is not value creations, but value build up. Value is created when the expected payoffs and/or the risk changes, so it could be the case that we could have positive residual earnings while destroying value (for a discussion of EVA see Fernandez, 2008). An approach to mix earnings with fair value with important assets has happened for the European real estate companies. Some of them are trying to clear the mess, breaking down the income statement between direct or realized earnings and indirect earnings, so the Balance Sheet can be constructed, exactly in the same way that R&D can be capitalized using information from the income statement. 12 Proposal for Discussion Purposes Only On the other hand, it can be argued that historic cost estimates might not be relevant enough for the investors in some industries. The historic cost accounting under uncertainty defers the earnings realizations, as if the uncertain earnings and the risk associated would cancel out (in the spirit of the Fed Model), with a conservative and contractual approach in mind (Penman, 2009). But, as accruals try to estimate the portion of the realized profit that cash flows are missing, fair value try to estimate the unrealized profit that historic cost is deferring. Although the Fed model may be a good first-cut for understanding the growth-risk issue at an aggregate level (Thomas and Zhang, 2007), information about deferred (uncertain) earnings may help individual investors to make its own risk-return assessments. Even more when the valuations are externally audited, and they provide a range of outcomes with an annual accountability. Analogously as accruals need to be assessed to understand its quality (in the sense of estimation errors and biases), fair values needs other probabilistic reversals to estimate its precision. Investors ask about this information, so attempts to not disclose it (as the SEC’s decision of not allowing the disclosure of possible and probable reserves) only adds more asymmetry between investors. Along the same lines, in a literature review, Kothari et al (2009) point out that there are two sometimes contradicting purposes of the accounting information, namely, valuation or information focus, and performance measurement and control purposes. They endorse just leaving the balance sheet as a liquidation value statement, with separable and salable assets, and using the income statement as performance measurement and control. This approach seems difficult to appeal even to debtholders. In the absence of pledged collaterals, they are more interested in the firm as a going concern, as it is signaled by the covenants, most of them related to cash flows, dividends pay-outs and leverage thresholds. I will try to show also that the apparent contradiction between the valuation and performance measurement and control roles of the accounting information is at least partially solved when we keep both objectives and try to accomplish them at once. First of all, fair value sheds light on the control objective, and the historic cost estimates help understand the information content of fair value. For instance, very reliable historic cost estimates may not serve the spirit of the contracts. Firstly, in some occasions those estimates hardly measure the indirect performance for long-lead industries. Secondly, some estimates could give no clue about the persistence of this estimate, so the need of strengthening collaterals or retaining performance rewards is missed. Also, since historical cost accounting also involves estimates (accruals), we need to estimate the reliability of these future estimates. It is a myth that earnings are better than fair value estimates based on the absence of liquid markets for many assets and liabilities. On the one hand the underlying assumption is that where there is a market, that valuation is king, implying the strong assumption of efficiency and one value fits all. But value is dependent on the purpose of the acquirer, and second, there is not an accruals market either, so fair value may help unravel these black-box estimates. Hence, externally evaluated fair value estimates with annual reconciliations help us in that task. It makes us clearly see the congruence of the estimates and its reasonability, as well as disclosure the non-captured indirect performance. So information and performance measurement and control should go hand in hand. 13 Proposal for Discussion Purposes Only Also they assert that there should be a limit on recognizing assets whose values depend on future management effort because of moral hazard. But this criticism is disregarding the possibility of keeping and articulating historic cost and fair value statements. Also, they are unintentionally limiting the role of accounting for measuring indirect performance. They only fixate on realized earnings, but they disregard the fact that the fair value also takes into consideration any possible risks, including management risk, because the future cash flows are risk-adjusted. Nobody sells or recognizes future cash flows at face value, so that cautionary statement forgets that the future cash flows need effort to be realized entirely. An alternative approach: Fair Value Statement vs. below-the-line notes It seems that there is not a definitive answer in this topic about which is the best approach (Ryan, 2008, Nissim and Penman, 2008), since the two approaches are not the panacea for any business circumstance and industry. As noticed by Penman (2009) the eclectic approach, trying to compromise between fair value and historic cost is the worst scenario. It has been argued that having both estimates would solve the problem, though it is very costly (Holthausen and Watts, 2001). That the cost of complying with disclosing fair value estimates is cumbersome is unknown. Surely it is more costly now, that there is not a clear articulation with the remaining statements. Anyway, the fair value is already used for some historic cost estimates and impairments. But more importantly, this approach, though apparently putting this irreconcilable issue to bed, is neglecting the importance of the fair value. It is not just about deferring the fair value to the below-the-line information (i.e. notes to the financials). If fair value information is considered valuable and relevant for investors, it is needed to be built a comprehensive fair value statement, which needs to be articulated with the remaining financial statements. If not, even disclosing both historic cost and fair value estimates, the information gets confusing, redundant and perhaps even misleading. So, since recognition implies some reliability, it is still needed to be assured that this new fair value statement disclosure is at least similar to recognition (Bernard and Schipper, 1994). I am sure that even achieving that articulating simplicity is complicated, but possible. III. Earnings Attributes Value Relevance This is the first approach that has been used in the prior literature. Basically value relevance studies are designed to assess how well particular accounting amounts reflect information that is used by investors in valuing the firm’s equity value (Barth et al., 2000). This definition assumes that that the usefulness of accounting information is in its ability to summarize the information used by investors, either they use it efficiently or not (in the sense of HEM). Conservatism I do not use Lev and Nissim (2004) and Penman and Zhang (2000) measures of conservatism. The latter case because it refers to specific assets that are not common in 14 Proposal for Discussion Purposes Only this industry. The former, because the difference of tax reporting vs. financial reporting is not a clear proxy in this industry, because the presence of tax pools makes difficult to compare across companies, since the taxes reported depend on the amount and type of capital investments and the different accelerated amortization rate employed. A depuration of this ratio is still possible, reversing the depreciation and investment mechanism, but the demand of information is very high, not only because of the need of a breakdown of the tax pools information, but also because the tax rates, abatements and incentives vary by provinces and territories (between 10% and 16%)14. See Ryan (2006) approach using O&G shocks in prices as an alternative to return news Accrual quality Other proxies cannot be used for the survival bias that introduces the requirement of having analyst’s information Look at Cormier and Magnan (2002) Predictability and Earnings-Returns Relation As per Sadka and Sadka (2008). Also Driesprong et al. (2007). They found the same than I found for Canada. Shocks in price decrease the volatility, so they are related to decreases in returns! IV. Canadian Oil and Gas Industry O&G firms use historical accounting in their financial statements15. The costs incurred in the discovery and developments of new O&G reserves16 are capitalized following either the full cost method (FC) or the successful efforts method (SE)17. Two problems become immediately apparent. First, the amount of O&G reserves discovered does not show in the balance sheet. So a reader of the financial statements could only find out how much has been invested in exploration activity, but not how efficient these investments have been. Second, the full value of the major asset of the firm, O&G reserves, is not reported in the balance sheet. To overcome these shortcomings, standard-setters require a comprehensive set of disclosures on reserves quantities and values in the footnotes. 14 The fiscal regime is a blend of royalties (10% to 45%) and income taxes. Both Provincial and Federal authorities are involved. For 2011 Federal CIT will be 16.5% and Provincial CIT will be 10% in Alberta and BC, 11.5% in the Northwest Territories, 12% in Manitoba and Saskchewan and 14% in Newfoundland & Labrador. 15 This section borrows from Badia and Duro (2011) 16 O&G Reserves is the main assets of the O&G producers companies. Reserves are estimated remaining quantities of oil and natural gas and related substances anticipated to be recoverable from known accumulations, from a given date forward, based on: analysis of drilling, geological, geophysical and engineering data; the use of established technology; specified economic conditions, which are generally accepted as being reasonable and shall be disclosed. 17 Under SE firms only capitalize those exploration and development costs that are associated to successful exploration, and expense those associated to unsuccessful projects. Under FC the majority of costs are capitalized. The underlying idea in FC is that all exploration costs are necessary to eventually lead to the development of the company’s total inventory of reserves and resources. 15 Proposal for Discussion Purposes Only The new Canadian rule discussed in this paper is the first serious attempt to modernize and harmonize O&G disclosure requirements to align them with current practices and changes in technology. Effective September 30th 2003, all public Canadian O&G companies are subject to National Instrument 51-101 (hereafter “the Instrument”), a new reserves disclosure regulation passed by the Alberta Securities Commission (ASC). The purpose of the Instrument, as stated by the ASC, is “to enhance the quality, consistency, timeliness and comparability of public disclosure by reporting issuers”. Under the Instrument, firms must distinguish between proved and probable reserves. Optionally, they can also disclose possible reserves. We can find previous distinctions between proved and probable reserves in Canada (before 2003) and with voluntary character in the UK, but they are ambiguous and inconsistent. The Instrument is pioneering in the unequivocal probabilistic definition of reserves taken from the Canadian Oil and Gas Evaluation Handbook (COGEH). Appendix 1 provides a graph and an example to illustrate how the classification is done. Appendix 2 presents an example of reserves value disclosure. According to COGEH, the “best estimate” of the reserves to be recovered should be the P50 estimate, whereas the P90 and P10 definitions correspond to conservative and optimistic estimates, respectively. The wider is the range between P90 and P10 the higher is the degree of uncertainty. In general, uncertainty decreases with time, as more information on a specific well or property becomes available. Specifically, Proved Reserves are those reserves that can be estimated with a high degree of certainty to be recoverable, such that there is at least a 90 percent probability that the quantities actually recovered will equal or exceed the estimated proved reserves (also called P1). Additionally, Proved Reserves have to be broken down between Proved Developed Producing, Proved Developed Nonproducing and Proved Undeveloped. Probable Reserves are those additional reserves that are less certain to be recovered than proved reserves. It is equally likely that the actual remaining quantities recovered will be greater or less than the sum of the estimated proved+ probable reserves (2P). Possible Reserves are those additional reserves that are less certain to be recovered than probable reserves, such that there is at least a 10 percent probability that the quantities actually recovered will equal or exceed the sum of the estimated proved+ probable + possible reserves (3P). The Instrument introduces other disclosures, some of them differing from the old SEC’s requirements. The following are the most significant additions: ο· 18 Use of forecasts of O&G prices and the costs of the firm to value reserves, in addition to the option of disclosing the constant prices case from the prior fiscal year-end18. That optionality was introduced in 2007. Before of 2007, it was also mandatory to disclose the constant case price. The constant price methodology is aligned with the SEC’s (previously last day spot price, and currently a first-day-of-the-month simple average). This constant price case is a good example of how verifiability and comparability sometimes do not go hand in hand, and standard rules for all the firms only makes an illusion of comparability, misleading the information users. Year-end price or average year-price seems to enhance comparability. But this is just only if both companies are the same. Let’s suppose that two companies have the same future production but different extraction patterns. Let’s imagine we know exactly the future prices, and under those O&G prices, both companies turn out to have same NPV, since 16 Proposal for Discussion Purposes Only ο· ο· ο· ο· The future expected cash flows from reserves production are discounted at 5%, 15% and 20% for the forecast price case, besides the usual 0% and 10%. A more comprehensive reconciliation of reserves estimates. Future development costs for the next five years. Breakdown of reserves by major product type. To further guarantee the reliability and comparability of the estimates, the Instrument requires firms to hire independent qualified evaluators19 and to use the COGEH standards to estimate reserves quantities. Reserves Committees, comprised of a majority of independent directors, are encouraged (not required) for the purpose of hiring the external O&G evaluators and supervising their numbers before official approval by the Board of Directors20. Also under National Instrument 52-110 - Audit Committees21 (“NI 52-110”) all the reporting issuers are required to have an audit committee. While that instrument requires for TSX firms a fully independent audit committee comprised of at least three members, this requirement does not exist for TSX-V firms. Since very often a director is a member of both committees (Reserves and Audit), it is reasonable to suspect that when a company is a Venture issuer and does not have a reserves committee it might be because the Audit Committee is not comprised of independent directors, shedding light on the potential quality of both historic cost and fair value estimates. A TSX firm with no Reserves Committee could be signaling lack of expertise in the Board in terms of reserves evaluation. But since there is no specific requirement for those directors in terms of special expertise (although it is advised to bring some experience in the task), it could be alternatively just a signal of lack of interest in the real quality or the external appearance of its reserves disclosures. For the lack of requirement in terms of the committee’s O&G evaluation abilities, the creation of the Reserves Committee for TSX firms is not a clear signal to screen out across them. As Dechow et al. point out “there are the O&G price effect cancels out the time pattern differences. If the O&G prices are different for the forthcoming years, it is not clear why we can presume that both companies are now comparable. So we can say that using the same spot rate is not enhancing comparability, but only perhaps increasing verifiability. In Canada, since those companies are disclosing their O&G price assumptions, and that assessment needs to be externally evaluated, we also have some degree of verifiability. 19 Senior producers (with >100,000 BOE/day production) have the right to apply for an exemption to the requirement that at least 75 percent of their proved reserves be evaluated by an independent qualified reserves evaluator or auditor by using properly qualified in-house evaluators. It is interesting to analyze the senior producers’ signaling role of this decision to investors and the resulting outcome in a separating equilibrium (in a pooling equilibrium we are just raising the level field): on the one hand, it could be argued that external evaluators may signal high-type firms (better quality reporting). Despite in-house evaluators could do a similar and less costly evaluation, the outsourcing signal high-quality, since the quality of the outcome will not be affected (producers still have a clear influence on the external evaluators over the quality of the outcome), but it will become more costly for producers to manage the reserves, in terms of less negotiating power on setting up the fees, so there is less room for management discretion and that room is more costly. On the other hand, senior producers’ size could be relevant for external evaluators, not only in terms of auditing and reserves consulting fee percentage (over their total business), but also in terms of attracting other customers, so this second effect could offset the first high-quality signaling effect. The concluding outcome is an open empirical question. 20 It is needed the signing and approval of the O&G disclosures by two officers and two directors confirming they have reviewed the data and procedures. 21 That regulation came into force on March 2004 17 Proposal for Discussion Purposes Only entities other than auditors that have a similar role in the financial reporting process, and thus may affect earnings quality, although research in this area is limited”. To compute the Fair Value of the Reserves, the cash flows are based on the asset that the entity has at present and must exclude any plans to enhance the assets or its output in the future, but includes expenditure necessary to maintain the current performance of the asset. The cash flows for assets that are under construction and not yet complete (e.g., an oil or gas field that is part-developed) should include the cash flows necessary for their completion and the associated additional cash inflows or reduced cash outflows. Hence, differences between market implied valuations of the reserves and fair values of the reserves are mainly related to different views on: ο· the expected values of the forecasted prices (not only for the forecasted benchmark prices, but also for the adjustment to arrive to firm’s realization prices, i.e. price decks), ο· the quantity and pattern of the extraction, ο· the costs of the firm (efficiency). They are classified as lifting costs -lease operating expenses22 per boe or mcfe produced during a period-, finding and development costs - costs associated with increasing and developing reserves during a particular period, etc. ο· The possibility that the assets to be used have a useful life beyond the forecasted one and technology improvements, ο· the risk or discount rate applied (depends also on interest rates), and ο· the growth beyond the existing projects (M&A, discoveries, etc). This concept also includes current contingent and prospect resources not classified as reserves. Reserves are defined as those resources considered commercially recoverable in a reasonable timeframe under the existing economic and technical conditions (i.e. chance of commerciality23 equals 100%). Unlike reserves, O&G resources may have a chance of development lower than 100% (contingent resources), or a chance of discovery and development both lower than 100% (prospective resources). Why the US is not the right setting Unlike O&G firms in Canada, O&G firms in the US have only been able to report proved reserves, a single point estimate of O&G reserves, using current O&G prices. Proved reserves in the US did not have a clear probabilistic definition and not mandatorily externally audited or evaluated. In addition, technical revisions were not provided in reserves reconciliations. In 2010 the FASB issued Accounting for Extractive Activities – 22 Canada has a concessionary type fiscal regime. Rights to explore, develop or produce oil and gas in a province are obtained by acquiring a oil and gas lease or license from the province or from another party that holds such a license or lease. The owner, usually the crown (Government), retains a royalty interest in the production. In general, royalties are based on a function of productivity and the wellhead price, and each province has its own royalty regime. Crown royalties rates typically range from 10% to 45%. Special and more beneficial tax and royalty regimes apply to oil sands projects (crude bitumen) and Arctic and Atlantic offshore production. Those crown royalties are generally deductible (Ernst and Young, 2010) 23 The chance of commerciality=chance of development x chance of discovery 18 Proposal for Discussion Purposes Only Oil & Gas, following the example of Canadian regulation. However, in the new FASB´s rule the reporting of probable reserves is optional (and almost none firms have disclosed them), the proved reserves have been ambiguously defined and the technical revisions24 reconciliations are not separately disclosed, forgetting an important maxim given by the Canadian regulator: “it is not about being “right”, but about knowing how wrong you might be!”25. Another major difference with Canada is also the SEC enforcement of only the disclosure of the constant price evaluations at a 10% DF (“Standardized Measure of Oil and Gas”, or SMOG), on the basis that it provides a better comparison between companies. NI 51-101, in Canada, adopts a different approach, which provides a broader array of information on which an investor can make a decision. In 2007, NI 51-101 was amended to allow for voluntary, instead of mandatory, disclosure of a constant price case. As per conversations with ASC officers, this was because they found that few people actually used the constant price case for any purpose, due exactly to the lack of comparison across firms and the illusion of comparability that creates. In Canada they require a forecast price case, with the condition that the forecast price should be within the range of major evaluators’ forecasts (which they publish on their websites and the ASC monitor). Required discount factors are 0, 5, 10, 15 and 20%. In fact, in the industry the general sarcastic opinion about the SMOG seems to be that the acronym is appropriate! For all the above reasons the US setting is not an entirely appropriate setting for analyzing the precision of the fair value. Accounting articulation between fair value and historic cost in the O&G industry It was already discussed in the previous section that there exists interrelations between fair value and historic costs estimates. The articulation reasoning can be applied to any industry, but in the O&G industry the link is especially clear, since the accounting rules help to link both concepts. Canadian GAAP (AcG-16) rules that for Full Cost Accounting, the DD&A rate (Depreciation, Depletion and Amortization) needs to be calculated following the unit of production method, i.e. based on the rhythm in which the estimated proved reserves of O&G flows through the production revenue. Also, the proceeds from disposals need to be applied against capital costs with no gains, unless such a disposal would result in a change in the depletion rate of 20% or more. 24 Technical revisions occurs since as an oil and gas field is developed and produced, more information about the mix of oil, gas, water, etc, reservoir pressure, and other relevant data is obtained and used to update the estimates of recoverable reserves. Estimates of reserves are therefore revised over the life of the field. Under NI 51-101 there is a distinction between technical revisions and economic revisions. Canadian’s disclosures do not include under technical revisions the confounding effects of changes in prices and infill drilling, which are disclosed under economic revisions (the SEC’s disclosures mix technical and economic revisions under one single reconciliation line, mixing those different concepts). 25 It is only by recognizing the errors that progress can be made. Only systematic biases can tell us a perseverance in the errors that is suspicious. But a naïve belief in the absence of errors implicit in one single estimate is worse than a bias, since it mistakes the nature of the valuation process, which is stochastic and the true expected value unknown (Pastor and Stambaugh, 2006). 19 Proposal for Discussion Purposes Only Along the same lines, the impairment test is related to fair value reserve through the two step ceiling test. The first step, the recognition states that if the carrying amount is higher than the undiscounted future net cash flows from proved reserves, the amount to be recognized is applied following the second step: the difference between Net Present Value at 10% discount and the carrying amount if applied to losses. What is really striking is that once there is an impairment charge, there is not possibility of reversals. Analogously, the Asset Retirement Obligations needs to be assessed at fair value from the inception, recognizing a liability and increasing PPE. The liability amount is increased by the accretion expenses, and the PPE amount is amortized by the unit of production rate. Finally, I revisit the tax expenses. In all the industries the transfer pricing schemes, tax shelters, etc can make complicated to know the real effect of taxes for valuation purposes, and this industry is not an exception. The main difference with other industries is that the depreciation, depletion and amortization recorded for financial statement purposes is not deductible, rather tax-deductible capital allowances specified in the Income Tax Act are allowed. They are the so-called tax pool. Tax pools are comprised of Canadian Oil and Gas Property Expense (COGPE), Canadian Exploration Expense (CEE), Canadian Development Expense (CDE), Foreign Resource Expense (FRE) and Well Equipment Pool (Class 41, only when-available-for-use). All of them have different discretionary deductible rates (from up to 10% for COGPE to up to 100% for CEE, per year of the unclaimed portion. i.e. on a declining balance basis) and can be carried forward indefinitely. Each year, the tax pools are increased by the cost of the new acquisitions and reduced by the proceeds of the disposition and by deductions claimed from the pool (Ernst and Young, 2010). Trusts work differently. Since they can deduct the distribution to its unitholders they can build a larger balance of tax pools. So basically all these almost up-front deductions make the taxes work for us. This is because the post-tax investment available is higher, i.e. it is as if the Inland Revenue would have given us a tax advance. But that only happens when we have future benefits, and also the impact of those immediate tax deductions would be different based on the different future benefits. So, in order to assess the real beneficial impact of these tax pools, we need to know the future earnings (cash flows) and future capital expenditures. The fair value incorporates these estimations, helping us to understand the ultimate positive effect of those accelerated depreciations. That accounting structural articulation between both estimates, combined with the extant fair value disclosures will help us to assess better the fair value relevance and reliability and the earnings attributes. As an example, the Fair Value estimates help assess the convenience of the capitalization of some exploration costs, as well as the reasonability of the depreciation rate. For instance, we can analyze whether a company with a lot of dry holes cost that have been capitalized under full cost accounting has a reasonable depreciation rate compared with a similar firm following successful method accounting. Interestingly enough, since the DD&A rate under full cost depends on the proved reserves, an acquisition of lower cost than existing in the Balance Sheet reserves can make lower the DD&A expenses, a signal that it is needed to assess the reasonability of 20 Proposal for Discussion Purposes Only those charges (DD&A=Annual Production(BOE) x Full Cost Pool/Total Proved Reserves (BOE)). A more detailed articulation is possible, and we would endorse it, but currently these are just examples of the main accounting articulation using the existing level of disclosures. V. Hypotheses I test 4 hypotheses on the null form related to the impact of the cross-sectional variation of accruals and fair value precision on the value relevance. I also test 3 hypotheses related to the impact of those precision scores (fair value and accruals) in the earnings attributes, due to the articulation between fair value and historic cost estimates. First, based on the relative value relevance of the portfolios, conditional to the differential fair value quality scores, I posit for both levels and changes in the score: H1: Investors give less or same weights to fair value (vs. historic cost measures, i.e. BV and Earnings) for portfolios that have lower information risk, as proxied by the distance between the two fair value estimates scaled by 2P reserves H2: Investors give less or same weights to fair value (vs. historic cost measures) for portfolios that have higher fair value precision score (or positive changes in the score), controlling for information risk Next, based on the relative value relevance of the portfolios, conditional to the differential accrual quality scores (as proxy by BTM for Balance Sheet and asymmetric earnings timeliness for earnings), I develop an additional hypotheses that posit: H3: Investors give less or same weights to historic cost (vs. fair value measures) for portfolios that have higher accruals precision score (or positive changes in the score), controlling for information risk Then, I test the dynamics of the interrelation of both precision scores (fair value and accruals) on the value relevance, with the rationale that the information that both bring to the investors is incremental but also interrelated, due to the articulation already commented: H4a: Investors give same weights to historic cost and fair value measures for portfolios that have higher accruals and fair value precision scores (or positive changes in the interaction between score), controlling for information risk H4b: The value relevance (as proxied by the Adjusted R2) of historic cost and fair value measures for portfolios that have higher accruals and fair value precision scores (or positive changes in the interaction between score) is not increasing, controlling for information risk Lastly, I test the dynamics of the interrelation of both precision scores (fair value and accruals) on other earnings attributes, namely Balance Sheet conservatism and Predictability and Earnings-Returns relation: 21 Proposal for Discussion Purposes Only H5a: Given some information risk and accruals precision, portfolios with higher fair value precision are equally conservative (as measured by asymmetric earnings timeliness (Basu, 1997) and downward bias in BookToMarket (Beaver and Ryan, 2000), i.e. earnings and Balance conservatism) than those with lower fair value precision H5b: Given some information risk, accruals precision and some firm characteristics (controls for potential growth), portfolios with higher fair value precision have equal expected predictability than those with lower fair value precision H5c: Given some information risk, accruals precision and some firm characteristics (controls for potential growth), portfolios with higher fair value precision have equal earnings-return relation than those with lower fair value precision, as a proxy for expected earnings persistence and growth (Sadka and Sadka, 2009) In the following sections I test the hypotheses using data from multiple sources VI. Sample and Methodology Data: Under the Instrument, all reporting issuers in Canada with O&G activities have to file an electronic version of the following forms to their respective securities regulatory authority: Form 51-101F1: Statement of Reserves Data and Other Information Form 51-101F2: Report of Independent Qualified Reserves Evaluator or Auditor Form 51-101F3: Report of Management and Directors These forms are available in the System for Electronic Document Analysis and Retrieval (SEDAR26), the database of the CSA. Many times, these forms are included in the Annual Information Form that TSX O&G firms have to file every year with information on their exploration and production operations. The sources of data I use are CanOils, TSX-CFMRC, TSX Venture Summary Trading Files, Alberta Securities Commission27 Database (ASC), SEDAR, FPost, Bloomberg and Sproule. CanOils28 is the leading database for all the Canadian Oil and Gas E&P companies. It provides in-depth financial & operating performance, oil & gas assets and projects, M&A deals, and equity and debt financings. CanOils covers all oil and gas companies listed on the Toronto Stock Exchange (TSX) and Toronto Venture Exchange (TSX-V). Canoils information is quarterly and annually. From Canoils I get for each firm the reporting currency. I keep it as it is, and only transform it into CAD when necessary 26 This is the equivalent of EDGAR in the US ASC is the equivalent to the SEC in the US for the Oil and Gas Industry 28 I gratefully recognize Canoils for giving me access to its unique database. I also thank Jonathan Moore and Tracey Nabe for their suggestions and insights. 27 22 Proposal for Discussion Purposes Only for comparison purposes29. TSX-CFMRC is a database that provides historical stock market data (price, volume, return, dividends, etc.) for common equities and noncommon equities (preferred shares, rights, warrants, units, etc.) that traded on the Toronto Stock Exchange (TSX), as well as market information (betas, treasury rates, indexes, etc.). TSX Venture Trading Summary Files is a new database supplied by the TMX group, with market information on the TSX-Venture equities. The TSX-V database does not adjust prices by splits, consolidations and dividends. For accessing to that information I use a combination of sources. For splits I use TSX Venture Listed Company Contacts, a TMX Group database that provides monthly outstanding shares, and I combine it with the information on the date of splits from Canoils. For dividends30, I go to the Toronto Stock Exchange website, and I extract from a ftp31 all daily publications (approx. 2,000 files), and through a python algorithm32 I extract the ex-dividend date, currency and amount information for our companies. The FPost Database (Financial Post Corporate Database) contains financial and corporate information. I use it for getting information about the major shareholders I also hand collect all the information from technical revisions disclosures and parameter forecasts from the Annual Information Forms, Annual Reports and Forms 51-101F1 to F3 in SEDAR. Since Canoils does not track the dates of release of the Annual Information Forms and Annual Reports and Quarterly Financial Statements, I build a second python algorithm to recover that information from SEDAR. I also extract the Reserves Committee information for each firm and year from SEDAR. I use a third algorithm to download all the annual information forms and annual reports, and extract the existence33 or not of the Reserves Committee for each firm-year, as well as the creditadjusted risk-free rate used for valuing the Asset Retirement Obligation. Bloomberg is used for data that proxy Oil and Gas Volatility (Oil and Gas VIX34), oil and gas prices 29 Although that translation seems irrelevant, I have to notice that the CAD currency has some potential hedging effect. With energy as a key driver in the Canadian economy, there is a high correlation between energy pricing and the Canadian dollar. During commodity price declines, the Canadian dollar weakens relative to the USD, offering a natural hedge in commodity downturns. During energy down-cycles, this lowers operating costs (in Canadian dollars), while improving sales price (WTI quoted in USD) and acts as a tailwind (Morgan Stanley Analyst Report, May 5 2009) 30 I thank Jill Scullion, from TMX group, for suggesting me this back-door. Most of the TSX-V companies did not report dividends. Trust mandatorily distributes dividends, but since O&G trusts are formed by mature, cash flow generating companies, they are traded at TSX. The remaining corporations at TSX-V, since they are in early stages of exploration of development, are not distributing dividends. 31 http://www.tmx.com/en/listings/products_services/ir_data_solution/venture_market_information.html 32 The python algorithms are available under request 33 I performed a random visual inspection for 50 firms (a 20% substantive test) to check the words used by the companies. Then, I tracked for all the annual information forms and the Form 51-101F3 (when both existed) the combination of words Reserve(s) Committee; Reserve(s) Evaluation Committee; Reserves, Review and Environmental, Health and Safety Committee; Reserves, Safety and Environment Committee; Reserves, Options and Environment, Health and Safety Committee. When the searched gave no result, it was performed again with the word committee. If it did not give any results either, I visually reviewed the form for accuracy (some old pdfs are in a not recognizable reading). Also if any firm discontinued the reserves committee I doubled check it manually (hence, it minimizes errors, since it is common once the committee has been approved it, not to discontinue it). 34 Also called the “fear index”. While volatility technically means unexpected moves up or down, the S&P 500 index option market has become dominated by hedgers who buy index puts when they are concerned 23 Proposal for Discussion Purposes Only index and some market information. Sproule is the monthly information on forecasted O&G prices for one of the major external evaluators, so that information is compared with Bloomberg for looking at forecasted prices and the measurement errors. I gather information from the ASC concerning Reserves Disclosures, which is useful for double checking Canoils information, the name of the evaluators by firm-year, getting track of the changes across years of the companies (for instance, for tracking the legal type of each firm-year: i.e. either trusts and corporations), as well as obtaining the data for reserves restatements. The dates of the restatements are got from SEDAR using the second python algorithm already mentioned. Lastly, I follow each company and link all databases by ticker(s) and name(s) one by one. I check that each information in Canoils match with the information in the remaining databases, and use several combinations to that purpose: I use ticker, name and the ASC database, to follow every company acquisition and M&A activity, changes in names, delistings, going private, amagalmation, legal type changes35, etc. Since it is more fiscally attractive36 for the vendor to sell the share of a company instead of a direct acquisition of the operating assets, instead of assets, the M&A activity in this industry is hectic. There are many ways of getting public, and the multilateral mergers are very common. Also companies graduate from TSX-V to TSX, sometimes changing the tickers. I create a unique ID for a company over time and a unique ID for a firm-year, since there are multiple changes over time. I manually check one by one with all the existing information in all the databases that the firms are financially the same across the years, despite all the changes, and that the information is comparable, and I adjust for any splits and consolidation transactions. I exclude firms that have been acquired, since it would require computing the transaction value and making further assumptions. I corroborate with Canoils that each firm is unique and consistent over the period we have information, and also double check it with SEDAR and the tracking database from ASC. Finally I exclude companies for which I do not find stock prices in the TSX-CFMRC and TSX-V about a potential drop in the stock market, (Whaley, 2008). The figure makes two simplifying assumptions: (a) the rate of return on the S&P 500 over the next 30 days is normally distributed, and (b) the expected rate of return on the S&P 500 over the next 30 days is zero. CBOE began calculating a commodity volatility index in 2008: CBOE Crude Oil Volatility Index (OVX) based on United States Oil Fund, LP(USO) options; 35 An open-ended investment trust is a legal structure that holds income-producing assets and pays income to unit-holders through distributions. Units trade like stocks. Distributions are tax deductible for the trust, eliminating corporate taxes, so it establishes a single level of for distributions to the unitholders. As of October 31st 2006 the Canadian federal government revoked this preferential tax treatment for income trusts, imposing a distribution tax on the income distributed to the unitholders. As a result, the trust would end up paying the same amount of a corporation. The new tax treatment applies for new trusts as of 2007, and its application is deferred until 2011 for trusts that were publicly traded at the date of announcement of the revoke (PriceWaterhouseCoopers, 2008 April). Almost all the trusts were converted into corporations in 2011 (exceptions two foreign trusts that using some loopholes still have a preferential tax treatment). 36 Only one half of the capital gain on the sale of a capital property (such as shares) is included in taxable income. However, the sale of operating assets can give rise to income, 100% of which is included in taxable income and capital gains. Similarly, the reason for so many corporate amalgamations is that no tax consolidation, group relief or profit transfer system applies in Canada. Each corporation computes and pay taxes on a separate legal entity basis, so if there were no amalgamation the interest expense incurred by the acquisition company would not be available to offset the income of the target company (Ernst and Young, 2010) 24 Proposal for Discussion Purposes Only databases37, using it as an additional check that the company is substantially the same over time, and that it can be compared across years. I identify my initial sample using Canoils Database as the master file, and including all junior and senior O&G producers38 for the period 2003-2010 (both years included). I only select public firms quoted in the Toronto Stock Exchange (TSX) and Toronto Stock Exchange Venture (TSX-V). This requirement is asked to avoid companies with little volume because of being cross-listed in other major stock exchange, the majority of them exempted for applying the Canadian GAAP. Since NEX has weaker requirements for the companies to be listed, I avoid using this segment of the Stock Exchange. I also require for each firm-year to use Canadian GAAP, in order to use consistent accounting regulation. I also ask for the firm to have assets and also reserves at the initial and end of the period, since this is the main focus of the paper, and firms without reserves are often very small. I do not exclude companies with no December fiscal year-end. The reason for not excluding them is I focus my study on the precision of the technical revisions, not economic revisions39. Also, I will control for O&G price volatility40. This requirement avoids temporal misspecifications due to different reporting, forecasted price assumptions and different cumulation periods of annual earnings and fair value estimates (results qualitatively hold even using the broader sample). This result in 1,200 firm-years (see table 1) Next, I drop XX observations from firms that are not pure O&G producers (integrated oil, funds and E&P firms with higher than 5% of revenues in non exploration-production activities, such as real estate, drilling, marketing, midstream and refining services). The valuations of these firms might be related with factors other than O&G reserves, potentially confounding my results. I do keep firm-years with legal structure as trusts and companies following Successful Effort Methods, though I will control for them in the methodology to be sure they are not driving the results. Then, I remove from the sample those firm years with no stock price information. A total of XX observations are left with the basic variables I need for the study: market value, book value, liabilities, PP&E (as a proxy for OGA), total assets, net income before extraordinary items, and all the measures of reserves estimates at 10% discount rate. Finally, I eliminate XX firms which changed fiscal years during the sample period and XX firm years with market values per barrel higher than Cdn$80. The latter criterion aims to eliminate firms whose main source of market value is not O&G and other outliers. Harris and Ohlson (1987) apply a threshold of US$40 of imputed value per barrel (note that IV=MV+TL−NOGA and that the average exchange rate for the period of my study was 1.26 Cdn$/US$), consistent with the crude nominal price level of their 37 This master file with all the matching links between databases and the tracking of companies can be sent for validity purposes under request. 38 Oil and gas activities are defined in the part 1.1 of NI 51-101 as those related to exploration, development, and production of hydrocarbons. This definition excludes transporting, refining or marketing of oil and gas, as well as activities related to the extraction of other natural resources. 39 My results are robust to excluding not December fiscal year-end firms 40 It will be performed through the historical annual standard deviation of prices (annualized standard deviation based on the , and through the VIX index, which tracks 25 Proposal for Discussion Purposes Only study period. My final sample contains 1,200 firm-year observations, from 2003, year in which the Instrument became effective, to 2010. The dependent variable, market value (MVE), is calculated taking the stock price and the outstanding shares three days after the filing of the Annual Report or NI 51-101 forms in SEDAR, whichever is filed later, to ensure that all the information is available to investors. For the time-series, in order to have enough statistical power, I require to have at least 4 years of information for each company. I find XXX firms with this information (XXX firm-years). Hence, the results may be biased because of survivorship bias, in the sense that the companies that do not have a longer stream can be either younger (XXX firm years starting at 2008), or more or less successful (XXX firms-years have been liquidated, acquired or going private. I have XXX firms acquired). The direction of the bias (if any) is difficult to predict, though it seems to be skewed towards more successful and mature firms. Methodology: First of all, I build a composite Fair Value Precision Score41 in order to assess the main criticism of fair value opponents, namely, its lack of reliability due to measurement errors and biases. This score helps examine the precision of the fair value estimations, in order to get a sense of the biases and measurement errors of the fair value estimates. Then, I perform cross-sectional and longitudinal firm-level tests to analyze the impact of the fair value precision on the value relevance of the fair value and historic cost estimates, following the Ohlson Model (1995, 1996). The value relevant impact of the fair value precision is a consequence of the information risk, which Riedl and Serafeim (2011) defines as “the uncertainty regarding valuation parameters for an underlying asset”. Also, as in Francis, Olsson, and Schipper (2008), I define precision of information as the precision (and lack of uncertainty) of a measure with respect to a valuation relevant construct. In our case, the reliability and precision of the measurement of the probabilistic distribution of the barrels associated to the O&G reserves. Finally I carry out sensitivity analysis Fair Value Measurement Errors: Introduction The Fair Value Precision Score have been constructed using the reversals for both proved (1P) and proved plus probable (2P) reserves, as measured by the technical revisions. I 41 This score is building in the fruitful accounting avenue of other documented practical scores, such as bankruptcy prediction models producing Z-scores, O-scores, M-scores, Q-scores, and S-scores (Altman, Ohlson, Penman…). All of them have been developed with a product oriented approach in mind, and use financial statement information, most of the times from the statements or notes closely related to the statements (such as LIFO reserve). Analogously, I want to use KPIs, and other footnotes to shed light on the quality of the fair value probabilistic estimations. This fair value score is difficult to estimate, since most of the times firms only disclose a single point estimate, and we have little information on the inputs for the information. That’s why the importance of the unique setting I am studying to shed light on the quality of the fair value. 26 Proposal for Discussion Purposes Only also use some KPIs as a double check of the reliability of the disclosures, given that because of the limited number of years we have it is possible to manage even the technical revisions42. In this study the precision of the fair value is analogous to the accruals literature (discretionary or not), in which over or underestimated accruals are captured through future reversals. Similarly, fair value estimates leave a trail: reversals in the probabilistic estimates beyond the expected changes or big bath fair value accounting that will disclose increases in the reserves estimates without a clear link to the finding and development costs. Most approaches to measurement error in the O&G accounting literature rely on imposing structure on the statistical properties of the errors, as reflected by the model specified for calculating the regression. Although that approach is sound econometrically wise, when there are many variables introduced, the number of assumptions increases dramatically, being difficult to assess the ex-ante assumptions without a piece of evidence. Hence, those are joint tests of the uncounted assumptions and the model, leaving us the impression that the results are difficult to follow or just a minor empirical validation, given the uncounted ex-ante priors of the designer. Boone (2002) is a meritorious example, in which the author has to posit 15 covariances in order to be able to analyze whether there are measurement errors. Also, though he is eager to point out that the market is just the benchmark, not the error-free true values, most of the heroic assumptions imply that the market perfectly understand all the sources of measurement errors and its 15 covariances, which is at least challenging without assuming market efficiency. For instance, not differentiating between positive shocks and negative shocks and not taking into consideration the conservatism may invalid the first covariance, i.e. measurement errors and market values have a negative covariance. Also, the covariance between cash flow and discount rates shocks is not discussed. So, even a minor flaw in his long chain of assumptions or any empirical counter fact would have a domino effect in the whole structure. On the contrary, as I will elaborate further below, I just try to let the numbers speak alone with much weaker assumptions: technical revisions43 will reflect along the years all the sources of error and biases, because reserves estimates adjusted by production should ultimately converge to the true O&G reserves. Fair Value Measurement Errors: Technical Revisions Reserves are estimates made44using uncertain and limited information. Hence, it may be affected by measurement errors, biases and trends, as drawn in figure 1. As per the figure 1, for 2P reserves the trend (in the absence of skewness) will be nil, and for the proved reserves it will be positive. There are displacement bias, (shifts in the entire distribution) 42 Although we refer to management of disclosures, we do not intend to say flagrant fraud, but rather measurement errors or biases, either intentional or not. Fraud can be even challenging to prove with external information 43 This approach is based on Elliott and Robinson, 20XX 44 This section borrows heavily from Elliott and Robinson, 20XX. COGEH and Determination of Oil and Gas Reserves. The COGEH is the main authoritative references in this field as per the NI 51-101 27 Proposal for Discussion Purposes Only and variability bias (changing the variability of the distribution, under or overestimating artificially the uncertainty of the estimates). Technical revisions is a category required to be reported as part of the reconciliation of previous and current financial year-end estimates. A technical revision is a change in reserves estimates (in barrels) in properties owned by the firm at the start of the reconciliation period as the result of new technical information (including from production). It does not include factors that require capital expenditure, such as infill drilling, or the consequences of royalty or ownership changes, which fall into the reconciliation categories of extensions and improved recovery, and economic revisions, respectively.45 As it was said in section III, while US O&G firms provide only a point estimate of the reserves (proved), Canadian companies disclose at least two point estimates, supplying also an approximate measure of their O&G Reserves uncertainty. Managing this uncertainty46 is extremely important in the O&G industry, for investors, debt suppliers, and even for the industry players, given the large number of M&A transactions. The above discussion is for individual wells. However, the aggregation of not perfectly correlated wells will cause that the measurement errors, if random, be cancelled out, compensating across wells and over time. Similarly, if the biases are unintended and random, the effect will be zero. However, that is not usually the case, so since the reserves will be flowing through the cash flows, the biases will ultimately be reflected in the technical revisions. Finally, the trend for proved reserves, given that the sum of them is arithmetic (not probabilistic) will increase. The 2P reserves trend would be nil in the absence of skewness, since the aggregation would cause not effect. In the presence of positive skewness there would be a positive trend, which would be ultimately equal to the difference between the median (2P) and the mean. 45 This explanation borrows from the ASC 2009 report “On one hand, oil reserves are nothing more that another type of company’s inventories, but unlike the inventories that can be precisely calculated, oil reserves are uncertain. Oil and gas reserves represent the cumulative production of a field until it is completely depleted. Production depends mainly on the volume in place (net pay and area), the geology of the reservoir (porosity, permeability), the physics (engineering) of the fluids (pressure, temperature, saturation, density and viscosity), the development scheme (wells producers and injectors), and the economics (cost and price). The geological uncertainty adds to the economic uncertainties.” (Roman Kremer, 2005). 46 28 Proposal for Discussion Purposes Only Figure 1. Source: COGEH and Elliot and Robinson (2005) Figure 2. Source: COGEH and Elliot and Robinson (2005) But that trend is not visible if not corrected for production, that’s why it is important to define technical revisions over the opening balances Reserves Category Expected Outcome A review of technical revisions over time provides a measure of the quality of reserves estimates. Provided that the firms have employed appropriate evaluation methodologies, the technical revisions generally expected on various reported reserves categories are as follows Reserves Category 29 Entity Level (wells) Reported Level (firms) Proposal for Discussion Purposes Only Proved Positive reserves revisions should occur in significantly more of the entities than negative revisions 2P Close to zero: positive reserves (proved+probable) revisions should equal negative reserves revisions 3P Negative reserves revisions should (proved+probable+ occur in significantly more of the possible) entities than positive revisions Negative reserves revisions should seldom occur at this level Only minor positive or minor negative revisions should occur at this level Positive reserves revisions should seldom occur at this level The above results are a consequence of the the law of large numbers and the Central Limit Theory (CLT)47. The reason is that over time the O&G ultimate reserves distribution will reduce its variance, since news will come and uncertainty will vanish. As per the COGEH, “the uncertainty in reserves estimates is a function of the quantity and quality of the data available, which depends largely on the extent of depletion of an accumulation”. Basically, this approach is just reflecting a statistical fact. If a company states that he has more than 90% probability of having at least the number of barrels disclosed in the proved reserves, it means that the probability of having a negative technical revision is at a maximum 10%. Also, the probability of having a big negative technical revision is much lower than 10%. On top of that, given that the ASC mandates not to aggregate properties using probabilistic aggregations, but simple arithmetic aggregation, the probability of estimates different than the median are changed. Specifically, the proved reserves, i.e. P90, would have a much higher probability, around 98% (see example below extracted from CSA Staff Notice 51-327). As a conclusion, the possibility of having big negative technical revisions should be very low. That possibility, under ergodic central limit theorem assumptions and taking into consideration the low probability of high negative technical revisions, needs to be almost remote over time48. 47 Assuming independent and identical distributions (iid) and controlling for the number of wells companies have cross-sectionally as well as controlling for the geography (since the latter can introduce perfect correlations, as in the Gulf of Mexico, etc). We also need to track this information over time with stationary and ergodic CLT assumptions (we do not want to introduce additional uncertainty to the process so that it explodes over time. Hence, stationarity is needed to analyze data over time) 48 Also our measure of technical revisions for proved reserves takes into consideration the differential uncertainty over time and cross-sectionally, and control also for number of wells and geography, as a proxy of the level of approximation to the Central Limit Theorem due to aggregation. Under arithmetic aggregation the mean should not be affected. That takes into consideration that the expected value is a good approximation of the best estimate, given a repeated outcome, either through a number of wells or over time. Hence, the impact on the median is contingent to the proximity to the mean. As per the ASC the level of skewness is not known, but it is probably not great in the Western Canadian Sedimentary Basin where few fields are likely to have significantly skewed distributions. After conversations with the ASC they pointed out that the potential skewness in unconventional or frontier resources is material. Hence, it is more likely to have positive technical revisions for 2P reserves in firms which a higher proportion of those resources. We control for the proportion of those resources in the reserves through a proxy of the acreage of undeveloped resources. The level of uncertainty does not need to impact those 2P technical revisions, since we are aggregating the effect over a number of wells. 30 Proposal for Discussion Purposes Only Obviously, the probability of having negative technical revisions, although low, can differ across companies in accordance with their different fundamental process, i.e. their own underlying uncertainty. That’s why we would rank the companies, giving each one a percentile, based on the technical revision percentage adjusted for the expected value. The expected value will be the industry adjusted median49. In the sensitivity analysis I also use the industry-size adjusted median and other expected models. Statistical theory indicates that, because they are aggregations of a number of individual property estimates, greater variance of technical revisions may be expected for firms with smaller volumes of reserves than for firms with large reserves. Firms whose technical revisions are outliers beyond the usual pattern will have a lower percentile, indicating a lower precision in their estimates. As anecdotic evidence of the usefulness of this approach to capture abnormal patterns, as per conversations with ASC officers, this is the approach the ASC uses to select firms for further review of the evaluation reports as part of their continuous disclosure review program. In any case it is important to admit from the inception that this approach, though letting the numbers speak alone and just basing its assumptions in the well defined law of large number, it is not perfect. It just give a probabilistic suspect of lack of precision, but the impossibility of knowing the real underlying fundamental process makes this approach not deterministic. In fact, critics of the evidence presented in the managing of earnings literature (and by extension, fair value) claim there is no enough scientific evidence in those charges, with some exceptions, i.e. specific cases brought by the SEC, specifically major frauds. That criticism equates validity with the absence of doubt. But that strong evidence is not needed to make good decisions. As an analogy and that’s the reason for the title, in criminal cases, the prosecution has to present proof beyond a reasonable doubt. On the contrary, fair value managing evidence is closer to that of the civil trial, in which the plaintiff just needs a preponderance of the evidence. Analogously, although we would not whistleblower the SEC with the evidence at hand, at least we would like to 49 I also carry out a sensitivity analysis using the mean with analogous qualitative results (not tabulated) 31 Proposal for Discussion Purposes Only have this information to have a better probability of not investing in companies under reasonable suspect of lack of reliability. As per Bertrand Russell: “all exact science is dominated by the idea of approximation”. Fair Value Precision Score: composite proxy of Fair Value Quality Concerning the measurement error and bias criticism of the Fair Value, in a first stage, I use industry aggregate distributions to look for companies whose technical revisions are outliers versus the industry (peer) group. I perform this analysis looking at the usual pattern for each type of product (light & medium oil, heavy oil and associated and non associated natural gas). Similarly, I analyze the time series behavior to examine trends and variance in individual evaluator’s and individual firm’s estimates. Although the best approach would be to use the ultimate reserves as a benchmark, since it is not confused by the accumulated production effect (see figure 3), that information is challenging to obtain externally. This approach would not cause any issue on the assessment of the direction of the trend (see example). Instead, since we do not have that information, the changes have to be assessed looking at the technical revisions over the initial reserves balance, i.e. deducting the effect of the production. Year Initial 1 2 3 4 5 6 7 8 Production As of Dec. 31 Proved Proved+Probable Proved+Probable+Possible Yearly Cumulative Ultimate Remaining Ultimate Remaining Ultimate Remaining Remaining Ps 0 0 20 20 100 100 220 220 120 27.8 27.8 50 22.2 100 72.2 173 145.2 73 20.8 48.6 70 21.4 100 51.4 152 103.4 52 15.6 64.2 82 17.8 100 35.8 137 72.8 37 11.7 75.9 90 14.1 100 24.1 114 38.1 14 8.8 84.7 95 10.3 100 15.3 110 25.3 10 6.6 91.3 98 6.7 100 8.7 105 13.7 5 4.9 96.2 99 2.8 100 3.8 102 5.8 2 3.8 100 100 0 100 0 100 0 0 Figure 4. Source: Elliot and Robinson (2005) I build the Fair Value Precision Score in two alternative ways. First, I calculate the score for each firm and year (cross-sectional approach). Second, I use the technical revisions percentage average for each firm across years (time-series approach) and then use a cross-sectional time-series industry mean50. Below I detail the first approach. 1) I calculate for each of the main types of product (light-oil, heavy-oil and natural gas) and for each firm and year their 1P (proved reserves) technical revisions, as a percentage of 2P reserves (proved+probable reserves). The 2P scalar allows me to control not only for size but also for the variability, since for the same proved reserves, higher variability would involve a higher 2P reserves. Analogously, I calculate the 2P technical revisions as a percentage of 2P reserves. Now it is not 50 This assumes time invariant parameters, i.e. persistence in the precision of the fair value estimates over time 32 Proposal for Discussion Purposes Only need to adjust for variability as we explained above, since the expected values of the technical revisions needs to be zero over time and across wells51. The rationale for measuring the technical revisions not only for 1P estimates but also for 2P reserves is the following. First, concerning the 2P estimates, different firms may be more or less aggressive in the application of the fair value, either because of lack of technical expertise or because of biases. Second, even when two firms arrive to the same fair value 2P estimate, this is the first moment or expected value, so it may seem plausible that the same two firms have different probabilistic distributions in mind for the same expected amount, in terms of the distribution variance (second moment). And the fair value implementation seems to be more challenging when there are industry (or macrowide) inputs that are subject to uncertainty. 51 I will perform sensitivity analysis based on size (number of wells). Since reserves are aggregations of a number of individual property estimates, greater variance of technical revisions may be expected for firms with smaller number of wells (so generally, the more concentrated the production, the higher the uncertainty and risk). Also I will perform sensitivity analysis for age (different production rates), and geography, because of different skewness. I can improve also the variability proxy using Probable/2P adjusted measurements 33 Proposal for Discussion Purposes Only 2) Then I industry-adjust the results, subtracting the industry median52 (without taking the firm in consideration). The industry median helps understand which would be the expected technical revision, as well as control for macro-shocks that affect most of the participants53. For 1P, the positive technical revisions are expected54, so negative revisions or zero revisions are a signal of less precise estimates. For 2P zero revisions are expected, but since negative revisions are less desired and there is likely positive skewness I will use the same measure55. Then, I rank the firms and calculate quartiles for each type of product and for both 1P and 2P reserves. NOTA: ES absurdo calcular industry adjusted y luego ranking, pues a todos les quito lo mismo, preservando el ranking (tiene sentido si hay mas industrias). Por eso debería ser size adjusted (coger 4 deciles) 3) I calculate the scores weighing for the amount of reserves of each type for each firm and year πππππππ‘ (1π) = ππ»π,π,π‘(1π) ∗ π·ππππππΏπ,ππ‘ (1π) + ππ»π,π,π‘(1π) ∗ π·ππππππ»π,ππ‘ (1π) + ππ»π,π,π‘ (1π) ∗ π·πππππππΊ,ππ‘ (1π) πππππππ‘ (2π) = ππ»π,π,π‘(2π) ∗ π·ππππππΏπ,ππ‘ (2π) + ππ»π,π,π‘(2π) ∗ π·ππππππ»π,ππ‘ (2π) + ππ»π,π,π‘(2π) ∗ π·πππππππΊ,ππ‘ (2π) 52 I use the median for avoiding outliers. I also use the mean with similar results For instance, the Gulf of Mexico moratorium. 54 I am giving a premium to companies that could be engaged in big-bath activities. I expect that I will catch them up over time. In any case, the time-series Score will address this specific issue 55 An alternative would be to skewness-adjust the industry median and use the absolute value of the difference, weighing similarly the positive and negative differences 53 34 Proposal for Discussion Purposes Only 1P≡Proved reserves 2P≡Proved plus probable reserves ωLO,i,t (1P)≡ percentage of light-oil proved reserves (1P, in barrels) over the total proved reserves (in barrels) for firm i at the beginning of year t ωHO,i,t(1P)≡ percentage of heavy-oil proved reserves (1P, in barrels) over the total proved reserves (in barrels) for firm i at the beginning of year t ωNG,i,t(1P)≡ percentage of natural gas proved reserves (1P, in barrels) over the total proved reserves (in barrels) for firm i at the beginning of year t QuartileLO,i,t(1P) ≡ for the light oil type, firm i and year t, this is the quartile of industryadjusted 1P (proved reserves) technical revisions, as a percentage of 2P reserves (proved+probable reserves) NET OR GROSS?? 4) I build a composite score. I calculate the weights using a base case (2/3,1/3). Then I perform sensitivity analysis: (1/3,2/3) (0.5,0.5) and principal components. The base case rationale is based on the importance of proved reserves for measuring the DD&A rate and the impairments, as well as for the clarity of the negative technical revisions signal in this case (then I calculate quartiles): πΉππππππππ‘ (π_π ) = 2 1 ∗ πππππππ‘ (1π) + ∗ πππππππ‘ (1π) 3 3 The second approach (FV Precision Score time-series) follows the same steps, but in the step 2 I previously average across years for each firm. This approach helps to account for the possibility of one-off negative technical revisions that are compensated by future positive revisions with similar magnitude. Then I subtract the industry median calculated similarly. πΉπππππππ (π‘_π ) = 2 1 ∗ ππππππ (1π) + ∗ ππππππ (1π) 3 3 Reinforcing the FV Precision Score: KPIs as a Double-Check It could be possible that a company do not disclose or admit wrongdoing in the precision of their fair value estimates, so the absence of technical revision could mislead us, causing our first proxy not to capture that behavior in the short term. I use industryspecific KPIs (Key Performance Indicators) to alleviate the problem of the information risk of some firms, using them as double-check, and to give context to the fair value, enhancing its value relevance. Hence, I also analyze other cross-checking measures to compare with the industry groups to extract the right outliers along several dimensions that capture the biases and measurement errors (as in Grant T. Olsen et al, 2010). These checks are performed to reduce the possibility that the firms might be not disclosing the true technical revisions, and also to assess the congruence of the articulation between fair value inputs and historical cost past transactions: 35 Proposal for Discussion Purposes Only ο· ο· ο· ο· ο· ο· ο· ο· 56 Signal 1: Year-over-year percentual changes on Finding and Development Costs56 (including and excluding changes in Future Development Capital Costs. We don’t include Acquisition costs since it is less plausible they have been biased. Also, excluding changes in future development can signal biases with more precision) (in barrels, to exclude price effects). Rationale: if a company invest less per reserves added it can be possible, but it could signal overstatements of reserves. Ranking to look for quartiles. If -10% it may sign you are overstating. So more positive better percentile, less likely they have distorted the reserves data. Signal 2: Extension and discoveries (MMBOE, 2P, gross) over number of successful development and exploration gross wells drilled. Rationale: too many extensions and discoveries in very little wells can signal either a big discovery or some overstatements. So very big number, say 2Mn barrels/well implies lower quartile. (note: I have to change the sign to do the reading easier) Signal 3: Developed O&G Reserves per well (MBOE, net) on Net Producible O&G Wells (note: we don’t have this information, so I can use Net O&G wells, so it can alternatively signal undeveloped wells, a source of bias). Large volumes per well may signal overstatements Signal 4: Acquisition costs ($/P1, $/P2) on Total P1 (or P2) reserves acquired (versus industry). Finding Dev. And Acquisition. Usually very low costs sounds either overstatements (or alternatively high abandonment costs) Signal 5: Reserves Life Index (years) over Net Producible Wells (note: we don’t have this information). A high index combined with high amounts would involve they are not applying appropriately decline curves. I can compare it with industry mean, or a weighted average (weighting based on number of net producible wells). I would do a quantile rank on Life Index and on number of well (for both lower means higher percentile). I multiply both ranks, and calculate quantiles based on the resulting measure. Signal 6: Undeveloped Reserves (P1 and P2) on Company ranks (threshold: median). More risk to overstatement Signal 7: Future Dev. Costs per BOE on actual F&D actual costs per BOE Signal 8: Projected Reserve Inventory Life Index for Undeveloped 2P Reserves on Undeveloped Reserves (note: we don’t have this information, we just have 1P). MMBOE (more than 5 years sounds unreasonable given To comply with NI 51-101 standards, companies are required to present their finding and development (F&D) costs by following 2 possible methods: Method 1 (based on Proved (1P) Reserves) is: Exploration and development costs and changes in estimated future development costs relating to the finding and development of Proved Reserves during the most recent financial year divided by additions to Proved Reserves during the same period. Method 2 (based on Proved & Probable (2P) Reserves) is: Exploration and development costs and changes in estimated future development costs relating to the finding and development of Proved & Probable Reserves during the most recent financial year divided by additions to Proved & Probable Reserves during the same period. 36 Proposal for Discussion Purposes Only ο· ο· ο· ο· COGEH indications). Inventory Life=Future Devl Costs/Actual Costs. Number of years to develop these reserves given the actual funding of development costs Signal 9: Look at Reserve Replacement Ratios average57 (RRR average) and current RRR or rank companies (possibility of managing earnings) Signal 10: Future operating cost and future revenue costs (per BOE) over actual operating costs and revenues (per BOE) Signal 11: Future Abandonment ratio over ARO Signal 12: I estimate next year production vs actual production. See the correlation with technical revisions, if not correlated it seems they are telling us a lie ππ π²π·π°πΊππππππ = ∑ π ∗π ππ π€ π=π I then calculate the quartiles. I also analyze the correlation between both Scores and between the FVScore and the individual components of the KPIScore. Determinants of FV Precision Score (c-s approach) (sensitivity analysis) I have built a FV Precision Score based on realized technical revisions, as a proxy of the measurement errors. The following step is to analyze which the determinants of that precision or lack of are. These determinants involve an expected model of the future (one-year-ahead) FV Precision Score, i.e. in turn of the one-year-ahead industry-adjusted technical revisions. The main determinants of FV Precision Score can be grouped into three components. First, factors that proxy for uncertainty and/or trend-skewnes. Higher reserves uncertainty may cause higher technical revisions. Also, the trend-skewness is included to take into consideration that the 2P is the median, not the expected value, so higher positive reserves skewness (positive trend in the 1P and 2P reserves estimates) may cause higher positive technical revisions. Second, factors that proxy for the measurement expertise and /or potential biases. These factors help understand the possible incentives that may cause intended biases, and the lack of measurement expertise, which may cause unintended measurement errors and/or biases. The innate uncertainty and/or skewness can be measured through the following proxies: ο· Reserves Volatility: since the underlying reserve distribution is unknown, I use the ratio Probable/2P (both in BOE) as a proxy of the volatility of the distribution58. This metric underpins a firm’s ability to grow production in the middle term, since reserves are wasting assets and are depleted over time 58 As explained in Badia and Duro (2011) this proxy if we assume a log-normal is very closely to the lognormal standard deviation. As it is explained in Badia and Duro, it is assumed in the industry that the reserves follow a shifted truncated log-normal distribution, so the aforementioned proxy is noisy. Anyway, a combination of this proxy and that of the skewness help correct the main stumbling blocks associated with the different shifted and truncated points. 57 37 Proposal for Discussion Purposes Only ο· ο· ο· ο· 59 Skewness: The resources are classified as conventional (light and heavy oil, natural gas, etc.) and unconventional (shale gas, coalbed methane, oil sands, etc). "Continuous” or "unconventional" oil accumulation means that the oil resource is dispersed throughout a geologic formation rather than existing as discrete, localized occurrences, such as those in conventional accumulations. Unconventional resources usually have much higher positive skewness, i.e. the possibility of more positive extreme results. This differentil skewness with respect to conventional resources is due to the unknown extraction rates of these new discoveries, the longer timing of the projects and the lack of precise analogs. The unconventional resources has been newly developed in the last years, thanks to new exploration, drilling and technologies, so they still have a lot of questions marks in terms of environmental, political and technical impact. The issue is that not all those unconventional resources are classified as reserves, so we cannot use it as proxy for technical revisions, which only apply to reserves. Analogously, the undeveloped reserves are supposed to have more uncertainty, and in turn more skewness, than the already developed reserves. The uncertainty is supposedly already accounted for, since they are broken down between 1P and 2P, but the skewness is not dealt with. Hence, the ratio of undeveloped resources over the total reserves can be a valid proxy for both higher potential skewness and higher possibility that you have unconventional resources in the reserves. Production rate: the production is the main source of technical revisions, since it helps correct for the future estimates. So to account for the crosssectional variation in the speed and amount of production I use the ratio of production/producing reserves and producing reserves/total reserves59. Age: as I explained above new fields are more skewed. Also the production rate usually follows a log-normal or negatively exponential pattern. Hence, age, proxy as accumulated depreciation over depreciation and also as life reserve index (producing reserves/production), is an indirect measures of skewness as well as the production rate Size: ceteribus paribus, as per the central limit theorem, higher number of wells decrease uncertainty. Also, the law of averages states that if we repeatedly take samples of the same type of uncertain number, the average of the samples will converge to a single result: the true average of the uncertain number. Hence, for big companies, we would expect less negative technical revisions for two reasons. First, as we explained, provided that the entities are not perfectly correlated, as the number of aggregated entities increases, the 1P and 2P estimates move to the left, increasing the probability of positive technical revisions. Second, higher number of wells (either by firm or over time) make more plausible that the estimate converge to the expected value. It can be argued that the ratio may signal the amount of news or information firms have, so the uncertainty would indeed diminish. Also, the inverse of this ratio may signal future needs of organic capital expenditures, i.e. future debt financings, so the sign of this proxy is an open empirical question. 38 Proposal for Discussion Purposes Only ο· Geography: the differential uncertainty due to geography is supposedly embedded in the distribution. But the geography can affect via production rate and skewness. Though supply shocks such as the Gulf of Mexico moratorium do not affect the reserves quantity, it does affect the NPV of reserves via a production delay, hence decreasing the production rate. Similarly, the skewness in the frontier areas is different than that of the Canadian onshore fields. I proxy it with a Dummy, G, which equals 1 for more than industry median percentage of reserves in Canada and US and 0 otherwise. Canada and US are chosen since the legal institutions are enforcement mechanism are known to be stronger and they are closely geographically connected markets (cite Laporta, Australia). The measurement expertise and/or potential biases factors have been addressed in past studies. The main factors are: ο· ο· ο· 60 M&A transactions: Melumad and Nissim (2008) states that “companies are more likely to overstate earnings when they raise capital (e.g., Teoh et al., 1998b) or engage in M&A transactions (e.g., Erickson and Wang, 1999). Thus, the likelihood that earnings have been inflated is higher in periods preceding such activities”. As in Livnat et al. (2004) I also will use FIN (cash proceeds from issuance of common or preferred shares in the current and following year if available, scaled by beginning of year’s market value of equity) and FINQ (is equal 1 if the free cash flows is negative). Both, FIN and FINQ proxy for incentive to manage reserves, since it assures an easier access to the public equity and debt markets when there is cash needs. Auditors: a change of (in-house or independent) evaluators may indicate disagreement regarding accounting policies, and therefore higher likelihood that earnings have been overstated (DeFond and Subramanyam, 1998). I will use the evaluator the company choose (big four or others) and the change in evaluators as proxies60. Leverage: Reserves are used as a basis for securing loans (collaterals) and as part of the financing covenants. Additionally, the performance of the company in terms of new reserves discovered gives a signal of the risk borne by the financial institution. As is explained in the COGEH, “the loan life is usually limited to no Senior producers (with >100,000 BOE/day production) have the right to apply for an exemption to the requirement that at least 75 percent of their proved reserves be evaluated by an independent qualified reserves evaluator or auditor by using properly qualified in-house evaluators. It is interesting to analyze the senior producers’ signaling role of this decision to investors and the resulting outcome in a separating equilibrium (in a pooling equilibrium we are just raising the level field): on the one hand, it could be argued that external evaluators may signal high-type firms (better quality reporting). Despite in-house evaluators could do a similar and less costly evaluation, the outsourcing signal high-quality, since the quality of the outcome will not be affected (producers still have a clear influence on the external evaluators over the quality of the outcome), but it will become more costly for producers to manage the reserves, in terms of less negotiating power on setting up the fees, so there is less room for management discretion and that room is more costly. On the other hand, senior producers’ size could be relevant for external evaluators, not only in terms of auditing and reserves consulting fee percentage (over their total business), but also in terms of attracting other customers, so this second effect could offset the first high-quality signaling effect. The concluding outcome is an open empirical question. 39 Proposal for Discussion Purposes Only more than the period required to recover half the reserves (the reserves’ half-life), with a maximum term established by the financial institution’s lending policy. The loan amount is expressed as a percentage of the discounted value of production (NPV 10%), most of the times before taxes, unless a large portion of the company’s total cash flows is dedicated to loan repayments.” Hence, high levels of or changes in leverage ratios may indicate high likelihood of violating debt covenants and accordingly strong incentives to overstate earnings (e.g., Sweeney, 1994). Hence, I will include leverage as proxy. ο· Reserves Committee: comprised of a majority of independent directors, are encouraged (not required) for the purpose of hiring the external O&G evaluators and supervising their numbers before official approval by the Board of Directors. Also under National Instrument 52-110 - Audit Committees (“NI 52-110”) all the reporting issuers are required to have an audit committee. While that instrument requires for TSX firms a fully independent audit committee comprised of at least three members, this requirement does not exist for TSX-V firms. Since very often a director is a member of both committees (Reserves and Audit), it is reasonable to suspect that when a company is a Venture issuer and does not have a reserves committee it might be because the Audit Committee is not comprised of independent directors, shedding light on the potential quality of both historic cost and fair value estimates. A TSX firm with no Reserves Committee could be signaling lack of expertise in the Board in terms of reserves evaluation. But since there is no specific requirement for those directors in terms of special expertise (although it is advised to bring some experience in the task), it could be alternatively just a signal of lack of interest in the real quality or the external appearance of its reserves disclosures. For the lack of requirement in terms of the committee’s O&G evaluation abilities, the creation of the Reserves Committee for TSX firms is not a clear signal to screen out across them. As Dechow et al. point out “there are entities other than auditors that have a similar role in the financial reporting process, and thus may affect earnings quality, although research in this area is limited”. I proxy for the presence of reserves quality with the Dummy variable R, which equals 1 when there is not a reserves committee and 0 otherwise. ο· Stock and earnings performance: Past earnings performance can be a proxy for bias, since reserves estimates also play a role on some accruals estimates. Reserves impact on depletion costs through the two-ceiling accounting principle for impairment purposes under the Full Cost Accounting and through the depletion rate. Also, under Successful Efforts, classifying a field as reserves would allow the capitalization of the subsequent costs of drilling and development of the field (until then, the drilling costs are considered to be exploratory costs and are expensed). I will proxy for this with a Dummy variable, P, which equals 1 with below the median ROA and/or stock performance and 0 otherwise. ο· KPI Score: as I discussed above, the KPIs can be a leading indicator of the future technical revisions, due to intentional or unintentional unrealistic fair value estimates, compared with the past historic cost realizations. I will also interact the inverse of KPI Score with the FV Score to look for the effect of contrarian signals. 40 Proposal for Discussion Purposes Only ο· Reserve replacement ratio, which proxy for growth prospects. High growth companies are more likely to engage in earnings management, because missing a benchmark for a growth firm may be associated with a much stronger (negative) market reaction (Ertimur et al., 2003). For both 1P and 2P. After explaining the main factors that help explain the one-year-ahead Fair Value Precision Score, I combine them one by one in the following regression model (I can use multinomial model with quartiles or a linear regression using next year % of technical revisions as dependent variable). In a first stage, I regress separately the one-year-ahead FV Precision Score on those current factors. ππ πππππππ‘ππππ,π‘+1 (π_π ) = πΌ + π½1 ππππππ‘ππππ‘π¦π,π‘ + π½2 ππππ πΈπππππ π,π‘ + π½3 πΉπππππππ,π‘ 1 + π½4 πΎππΌππππππ,π‘ + π½5 πΉπππππππ,π‘ ∗ + ππ,π‘ πΎππΌππππππ,π‘ πΉπππππππ,π‘+1 (π_π ) = πΌ + π½1 βππππππ‘ππππ‘π¦π,π‘ + π½2 βππππ πΈπππππ π,π‘ + π½3 βπΉπππππππ,π‘ 1 + π½4 βπΎππΌππππππ,π‘ + π½5 βπΉπππππππ,π‘ ∗ + ππ,π‘ βπΎππΌππππππ,π‘ The coefficient of the FV Score variable gives information about the expected behavior of the technical revisions over time. If β3 is positive it means that the precision is a persistent quality of the firms (either a momentum effect or a potential smoothing of technical revisions). On the contrary, a negative β3 may signal that the precision is mean reversal (reversals). Then, I account for two new variables, news, which is the sum of the company residuals from the regressions over the actual value, a variable that proxy for news61, and pred62, which is the fitted values obtained with all the regressors over the actual values, a variable that proxy for explanatory factors. Obviously the sum of both variables equal one. Each year, I rank the predicted variable and assign firms-years to quartiles (an alternative would be to use as dependent variable the one-year-ahead industry adjusted technical revisions, without assigning them to quartiles). These variables will play a role in the consequences of the FV precision, namely the effect of the predicted values in the value-relevance and on the future abnormal returns. 61 I use as proxy of news the residuals, since by definition news is uncorrelated with information from previous periods. 62 This variable can be broken down by each regressor. First I can estimate the parameters and the partial effect of each of them, i.e. the parameters (uncertainty, biases, etc) and then I compute the dependent fitted value just using the calculated parameters (which rule out the other partial regressors’ effects). The variable is computed using the dependent fitted values over the actual variables. 41 Proposal for Discussion Purposes Only Lastly, KPI Score could also add value to predicting the one-year-ahead FV Score when used in conjunction with the contemporaneous FV Score for subsets of stocks. Hence, I examine the performance of levels and changes of KPI Scores within each category of stocks partitioned based on FV Scores. I use a two-way analysis, where I independently sort firms by their FV Score as well as by their KPI Score, and report the time-series mean one-year-ahead FV Score. Relation FV Precision Score and FV Restatements Firms have to disclose the technical revisions on an annual basis. Although technical revisions are supposed to be a measure of the precision of the reserves disclosed, there is a possibility that they are mistaken, either intentionally or not. Some of those mistakes are disclosed over time in the so-called reserves restatements. I am trying to capture those potential mistakes with the KPI Score, so it is relevant to ask whether the reserves restatements are related to KPI Scores and FV Precision Score. The reserves restatements can be seen as a robustness test, since they are an external indicator of the precision of the Fair Value and of the reliability of the disclosed technical revisions. There are some reserves restatements at SEDAR database that do not affect the quantities reported, but sometimes only include additional documentation or disclosures. In order to capture the genuine FV reserves restatements, I have compared the information on the Canoils database, which does not retroactively correct for restatements, and the ASC database, which do correct for reserves restatements. Then, for those specific restatement cases, I have got the date of restatement from SEDAR. I create a logit model with the dependent variable equaling one when there is a restatement, and 0 otherwise π ππ π‘ππ‘πππππ‘π,π‘+1 (π_π ) = πΌ + π½1 ππππππ‘ππππ‘π¦π,π‘ + π½2 ππππ πΈπππππ π,π‘ + π½3 πΉπππππππ,π‘ 1 + π½4 πΎππΌππππππ,π‘ + π½5 πΉπππππππ,π‘ ∗ + ππ,π‘ πΎππΌππππππ,π‘ First Consequence of the precision: Value Relevance Many studies performed in the O&G industry (Magliolo (1986), Harris and Ohlson (1987), Shaw and Wier (1993), etc.) rely on estimates of an "imputed value" regression model63. As Boone (2002) explains that is a restricted form of the standard balance sheet valuation model. This specification restricts the coefficients on liabilities and non-oiland-gas assets to 1 and the intercept to be constant across firms. Since none of those 63 This model imputes the value of oil and gas assets as total market value of equity less the book value of non-oil-and-gas assets and liabilities. The imputed value is then regressed on the accounting measure of oil and gas assets (either historical cost or present value). 42 Proposal for Discussion Purposes Only studies give thorough details of the computations to arrive to the Non O&G assets (that is not an easy task), it is difficult to assess whether the conclusions of Boone are correct because of differences while choosing the non O&G assets and liabilities or because of the claimed model misspecifications. Breaking down total assets (TA) into O&G assets (OGA) and non-O&G assets (NOGA) in the basic balance sheet identity, I express OGA as a function of owners’ equity (OE), total liabilities (TL) and NOGA: OGA=OE+TL-NOGA All variables are scaled by reserves quantities64. I proxy the market value of OGA with the aggregate –i.e. proved plus probable– disclosed estimation of O&G reserves (PVOG). Next, I substitute the market value of equity (MVE) for OE. Finally, in the case of TL and NOGA, I proxy market values with book values. After introducing market values, I isolate MVE on the left-hand side, obtaining the following regression model: MVEit =∝ +β1 πππΊπ΄ππ‘ + π½2 ππΏππ‘ + π½3 ππππΊππ‘ + πππ‘ (1) As mentioned, all the values in expressions (1) are scaled by units of Barrels Oil Equivalent65 (BOE) of proved plus probable reserves. Barrels equivalents of reserves have been often used as a deflator in previous O&G research (e.g., Magliolo, 1986; Harris and Ohlson, 1987, Badia and Duro, 2011). It provides a natural deflator that allows a meaningful economic interpretation of the variables and mitigates the scale effects (Barth and Kallapur, 1995; Easton, 1998; Brown et al., 1999). I also run a returns specification to further alleviate heterogeneity and scale effects. Barth and Clinch (2009) carry out simulations on the relation between equity market value and equity book value and earnings in the context of the Ohlson model (1995). They show that share deflated and undeflated specifications generally perform best, regardless of the type of scale effect, confirming that size differences across firms in and of themselves need not reflect scale effects that lead to incorrect inferences. So the conclusion is that it is not possible to know whether there are scale effects, and the scalars introduce additional issues, biasing the inferences in an unknown direction. In any case, since I will also run the Ohlson model66, I will use no deflators and number of shares as a deflator for sensitivity purposes. 64 This section borrows from Badia and Duro (2011) A Barrel of Oil Equivalent (BOE) is a measure approximately equivalent to 1,000 ft.3 of Oil and a 6,000 ft.3 of Natural Gas. This measure is based on the approximate equivalent heating value of oil and gas. Berry et al (1998) conclude that the revenue-based conversion is not superior to the energy-based conversion. The revenue-based conversion in their period is approximately 12 and in my sample period is 10.5 (Using WTI and Henry Hub: 1 barrel of oil values 10.5 barrels of natural gas. This ranges from 6 in 2003 to 18 in 2010). Hence, although as they acknowledge, there is a risk of time idiosyncrasy in their results (1989-1993), the context is similar. I perform my results with both estimates with qualitatively similar results 66 Barth et al. results are not confined to Ohlson model, but this is a direct application 65 43 Proposal for Discussion Purposes Only As I already mentioned before, this specification follows the standard balance sheet valuation model. In fact the right hand side is a breakdown of the equity book value. This approach is convenient when the goal of the study is to analyze just the fair value features67 without further information. But since I will be comparing historic costs with Fair Value, I need a more comprehensive measure of the historic cost estimates. As Penman (2009) reminds us there is also an income statement. For this reason, I will follow Ohlson model (1995, 1996, 1999, 2000) for the model specifications, as in Berry and Wright (2001). The Ohlson model represents firm value as a linear function of equity book value and the present value of expected future abnormal earnings. Ohlson shows that balance sheetbased and earnings-based valuation models represent the two extreme cases resulting from limiting assumptions regarding the persistence of abnormal earnings (Barth et al., 2000). The notion of economic rents is captured in the persistence parameter on abnormal earnings. In our model, since I am allowing for the interaction of the Ohlson model components with the firm-level persistence, I am controlling for nonlinearities (Barth et al. 2000, Holthausen and Watts, 2000). Sensitivity analysis is an alternative way of controlling for nonlinearities that I will also perform. The Ohlson model states that: ∞ MVEit = π΅ππΈπ‘ + π½2 ππΏππ‘ + ∑ π ⌋ π −π πΈπ‘ ⌊ππ‘+π (3) π=1 Where R equals one plus the one-period interest rate, r, t denotes years and Et is the expectations operator with expectations based on available information at time t. The abnormal earnings are defined as: ππ‘π = πΈπ΄ππ πππ‘ − ππ΅ππΈπ‘−1 To restate the equation 3 in terms of the current information available at time t, it is usually assumed linear information dynamics to self specify a relation between current and future abnormal earnings: π ππ‘+1 = πππ‘π + ππ‘−π + ππ‘+1 ππ‘+1 = πΎππ‘ + ππ‘+1 Where ν represents “other information” about future abnormal earnings reflected in share price, but not yet captured by equity book value and earnings. Parameters 0≤ω≤1 and 0≤γ≤1 reflect the persistence of abnormal earnings and other information. The “other information” would be relevant if it useful to predict future (expected) abnormal earnings. Hence, to do a fair horserace with respect to historic cost estimates, I will do also include the earnings changes, as a proxy for growth. Also the Ohlson model assumes clean surplus relationship. In our sample, there are elements that goes directly to equity (hedge instruments, CTA, etc), but its magnitude is small, so I can neglect its importance 67 As in Badia and Duro (2011) 44 Proposal for Discussion Purposes Only Since I will run pooled regressions I will include fixed-year effects to control for value relevant information specific to each year. In a sensitivity analysis I will also control for the firm-year distinct volatility since the O&G price volatility is high and the firms disclose their results in a timeframe that starts up to 1st April for TSX firms and up to July 1st for TSX Venture firms. I will also perform year-by-year regressions. After studying the “measurement perspective” (Beaver, 1998) I will analyze whether the fair value is timely (“informational perspective”) with respect to historic cost estimates. I operationalize the model in the following way: 8 MVEit = β0 + ∑ πΌπ‘ π·π‘ + β1 π΅πππ‘ + π½2 πΈπ΄π ππππ‘ + π½3 π₯πΈπ΄π ππππ‘ + π½4 πΉπππ‘ π‘=1 2 + π½5 ππππππππ‘ ∗ πΉπππ‘ + π½6 ππππππππ‘ ∗ πΈπ΄π ππππ‘ + π½7 ππππππππ‘ ∗ π΅πππ‘ (2) + π½8 ππππππππ‘ ∗ π₯πΈπ΄π ππππ‘ + πππ‘ MVEit = β0 + β1 π΅πππ‘ + π½2 πΈπ΄π ππππ‘ + π½3 πΉπππ‘ + π½4 ππππππππ‘ ∗ πΉπππ‘ + πππ‘ MVEi,t≡ market value (number of shares outstanding times the stock price at the end of the following month after the Annual Information Form disclosure date) for firm i at year t Dt≡ dummy variable, 1 if it is year t, and 0 otherwise BV≡ Book Value of Equity for firm i at year t EARNS≡Net income before extraordinary items and discontinued operations for firm i and period t FV≡ Net Present Value of reserves is the discounted value of estimated future revenues from Proven Reserves less royalties, operating, development, and well abandonment costs. The figure used is based on forecast oil and gas prices, before tax, and are presented using various discount rates (0% and 10%), for 1P and 2P PScore≡ Precision score for firm i at year t (levels and changes). I use FVPrecision Score, KPIScore, and Pred alternatively as well as additional dummies with the interaction effect of both Scores and the remaining regressors. The level of precision gives an indication of the reliability of the fair value measure. Change in the precision score may indicate the effect of the build-up credibility on the actual values Firm characteristics≡ proxy for financial risk, either operating or leverage, such as size, leverage: NetDebt/2PBOE (Net debt is defined as: Total debt (defined as bank overdraft + current portion of long term debt + revolving credit facility (364-day loans) + short term convertible debentures + current portion of capital leases + long term debt + convertible debentures + long term capital leases) less cash and cash equivalents, restricted cash and marketable securities). All the variables are scaled for 2P BOE, using energy and revenue conversion (Alternatively using shares as deflator and undeflated). 45 Proposal for Discussion Purposes Only Following prior research, to ensure accounting information is in the public domain, I measure P six months after fiscal year-end (Lang, Raedy, and Yetman, 2003; Lang, Raedy, and Wilson, 2006). The value relevance metric is the incremental adjusted R2 from the regression. I perform this regression using pool cross-sectional time-series and alternatively FamaMacbeth regressions to control for cross-sectional correlations. It is not expected to find a high serial correlation given that the historic cost estimates and fair value estimates are very sensitive to unpredictable shocks in prices (also for the fair value the additional presence of unpredictable new discoveries). Anyway, there could be an errors-invariables issue due to measurement errors. As explained in Peterson (2009) the presence of a time-effect, i.e. observations in different firms within the same year are correlated, can be corrected through Fama-Macbeth approach. Clustering by time allows observations to be cross-sectionally correlated, but assumes independence over time. This approach give better estimates than Roger standard errors (Huber-White standard errors), since the number of years is small. This is the most likely presence of correlation, due to the presence of industry simultaneous shocks to prices (market-wide shocks). But since the variables could have some level of predictability or persistence (firm effects, ie correlation of firms across time), I also run the regressions controlling for both dependences (clustering by firm and year). As in Thompson (2009), the formula for both clustering without persistent common shocks (firms in different periods are uncorrelated): But Thompson (2009) advices that the double clustering (firm and year effects) is not relevant when we have much more firms than years (our case). So, following Gow, Ormazabal and Taylor (2010), they suggest that researchers using asymptotic methods in small samples may wish to consider using two-way cluster-robust standard errors using the cluster bootstrapping technique discussed in Cameron et al. (2008). Finally, it is important to notice that the dependent variable, if we assume prices follow a random walk, is a unit root process, i.e. it is not stationary, having a time-variant mean and variance and also causing inference issues. To solve that, I will add the lagged price as an independent variable. Additionally, I will first difference, performing returns regressions. (is it possible they are cointegrated?) Hence, after performing the levels regressions, I perform changes regressions. The question to be asked is different though. A return specification is testing whether the accounting information is disclosing new information incremental to that from other public sources in a timely relevant fashion (“information perspective”). As Lambert (1996) explained, that the information is new is not the only attribute we should ask to the accounting information. But for testing the timely feature of the accounting and fair value information, I will also perform price changes (returns) specifications. The returns specifications have its own issues (Barth, 2000), as for instance the matching between the economic event and the accounting recognition. For our study, that matching is almost perfect, since the fair value can only be partially known through the returns period, but it is not possible to advance it sooner. 46 Proposal for Discussion Purposes Only Penman (2011) warns us that “we should first attend to specification (of regression equations) so that the interpretation of observed correlations can be made within a (regression) framework that incorporates all known structural relationships”. As it is stated by accounting: ΔBit = OIit – FCFit β NFEit + FCFit – dit = Earningsit – dit Where OI is operating earnings from the business less net financing expenses, FCF is the free cash of the business, NFE is Net Financing Expenses and d are dividends. It is worthwhile to notice that FCF is drop when calculating the incremental Book Value of Equity (following of course clean surplus assumptions). As Penman states, under a perfect measurement of the true value through the Book Value we would expect the following relation to hold: Rt ο½ Pt ο Pt ο1 ο« d t Earnings t , ο½ Pt ο1 P t ο1 If the Book Value measures price with errors, we would expect: Pt – Pt-1 + dt = Earningst + (Pt – Bt) – (Pt-1 – Bt-1) This recognizes that stock returns are equal to earnings plus a change in the price premium over book value. Hence, there are two possibilities for earnings to be sufficient: no change in premium and a perfect identification of Book Value with price. So for information to be relevant beyond earnings, it has to provide a link to the change in premium, through information about the differences between the book value amounts and the future discount rates news, cash flows news and the expected realization (Campbell, 1991). Rt≡Et-1(Rt)+Ncf-NR As per the Ohlson Model and assuming constant required rate and growth, the premium at each point τ is: Pο΄ ο Bο΄ ο½ E ( Earningsο΄ ο«1 ο rBο΄ ) rοg Then, substituting equation X in the equation Y we obtain: Pt ο Pt ο1 ο« d t ο½ Earnings t ο« 47 E οEarnings t ο«1 ο rBt ο ( Earnings t ο rBt ο1 )ο rοg Proposal for Discussion Purposes Only g ( Earnings t ο rBt ο1 ) rοg As Earningst+1 β rBt β (Earningst β rBt-1) = Earningst+1 + rdt β (1+r)Earningst, one can refer to the growth as abnormal (cum-dividend) earnings growth that explains the P/E ratio, as in Ohlson and Juettner-Nauroth (2005). It implies that the change in premium is an earning characteristic. So, any other information is in fact giving information about future earnings growth (or discount rates news). If we do not include that growth variable the equation would be: ο½ Earnings t ο« Pt ο Pt ο1 ο« d t Earnings t B ο½ Rt ο½ a ο« b1 ο« b2 t ο1 ο« ο₯ t Pt ο1 Pt ο1 Pt ο1 Changes in earnings (growth in earnings) have an informational role if they forecast subsequent growth that induces a change in premium. The specification also recognizes a role for book value (other than for the case where Pt = Bt): Book-to-price loads with a non-zero coefficient if it indicates growth. So the main point from Penman (2011) is that we have got a structural explanation for the returns model specification and the role of the changes in earnings and book value are useful as a predictor of growth in earnings. The missing point is the information that can be inferred from earnings and other information concerning discount rate news (see Sadka and Sadka, 2009). In any case, regardless of the explanation, the inclusion of the book value is relevant versus prior studies, starting from Ball and Brown (1968). Hence, following Penman’s advice I include the prior year Book Value, so my model specification is as follows: R it = α0 + β1 πΈπ΄π ππππ‘ + π½2 π₯πΈπ΄π ππππ‘ + β3 π΅πππ‘−1 +β4 π₯πΉπππ‘ + π½5 ππππππππ‘ ∗ π₯πΉπππ‘ + π½6 ππππππππ‘ ∗ π₯πΈπ΄π ππππ‘ +π½7ππππππππ‘ ∗ πΈπ΄π ππππ‘ + π½8 ππππππππ‘ ∗ π΅πππ‘−1 + πππ‘ (1) R it = β0 + β2 π΅πππ‘−1 + π½3 π₯πΈπ΄π ππππ‘ +β4 π₯πΉπππ‘ + π½5 ππππππππ‘ ∗ π₯πΉπππ‘ + πππ‘ Rit≡ Twelve month cum-dividend cumulative return for year and firm i (ending when the annual reports of year t are disclosed) EARNSit≡ Earnings for firm i and period t ΔEARNSit≡ change in earnings for firm i and period t BVit-1≡ Book Value for firm i and period t-1 PScoreit≡ Precision score for firm i at year t (levels and changes). The precision level shed light on how reliable are the changes in the other measures. The precision changes indicate the precision build-up in the prior years and their impact on current changes in the other estimates (I should also interact precision changes with levels). All variables are scaled by last year stock price (and alternatively the deflators used for the levels regression). That scalar, as noticed by Penman (2011) initializes for information in price at the beginning of the period, so variables are relative to expectations at that point. 48 Proposal for Discussion Purposes Only Second Consequence: Future returns and mispricing Some could argue that TSX Venture companies are arguably less efficient than TSX firms. But, as per Damodaran (book) “if stocks that are 'neglected' by institutional investors are more likely to be undervalued and earn excess returns, the odds of finding undervalued firms should increase in this sub-sample.” So I will also control for TSX and TSX-V firms. The variable pred represents the expected values of future technical revisions, so under the market efficiency hypothesis it would be reasonable to expect that investors have already impounded that information in prices. If that is not the case, an alternative explanation, under the same paradigm, would be the presence of costs associated to predicting the future technical revisions, or some constraints that prevent investors from arbitraging away this effect (liquidity, transaction costs, etc). I investigate whether the FV Precision Score (alone or combined with the KPI Score) or the pred variable predict one-year-ahead stock returns directly. I sort the firms into 4 portfolios, based on the mentioned variables (one at a time). I calculate the (equally weighted) mean buy-and-hold raw and size-adjusted returns for the following twelve months (analyze survivorship bias). I compute the size adjusted return on a size-matched, value weighted portfolio formed from size-quartiles in the TSX-V and the CFMRC. The difference between the mean twelve-month raw return for portfolio 1 and that for Portfolio 4 is XX%, with a t-statistic estimated from the time-series of 8 returns of X.X. The corresponding return difference for size-adjusted returns is XX%, with a t-statistic of X.X. When I compute these returns for two-years-ahead the effect is reduced, consistent with a mispricing explanation (a risk explanation would cause a non short-lived return effect) Those observed returns could reflect risk factors. Since they are based on the same industry and are size-adjusted they mitigate partially this explanation. Anyway, to rule out that effect I give the results of estimating returns regressions including Fama-French factors. The coefficient on pred (and FV Score) is still significant π΅π,π‘−1 +α4 πΏπΈπππ‘ + πΌ5 ππππππππ‘ ππ,π‘ (1) R i,t+1 = α0 + πΌ1 π½ππ‘ + πΌ2 πΏππππ‘ + α3 ππ + πΌ6 ππππππ‘ +πππ‘ Third Consequence: betas and liquidity (asymmetry of information and information risk) To be written (Verecchia and Easton approaches). VII. Results 49 Proposal for Discussion Purposes Only To be written VIII. Sensitivity Analysis The above analysis investigates the effects of the fair value precision score on the value relevance, future returns (mispricing) and information risk. The following step is to analyze under which context or firm characteristics investors attach less importance to the fair value precision score and/or pred variables. Specifically, I examine the effect of size, full cost versus successful effort methods, ratio of probable to proved reserves, quality of estimates, firm’s legal form, stock exchange, December fiscal year-end firms, credit adjusted returns vs 10%, price volatility at disclosure date and price (and/or quantity) changes. I also do some robustness tests using FV before taxes, and disaggregating 2P between Proved and Probable NPV reserves. Size Full Cost vs. Successful Effort Method These methods differ in the treatment of specific operating expenses relating to exploration costs (as opposed to acquisition or development costs, which are capitalized in both methods). Exploration costs are costs relating to carrying and retaining undeveloped properties, costs of the collection and analysis of geophysical and seismic data, and costs incurred with drilling an exploratory well. The successful efforts method capitalizes only those exploration costs associated with successfully locating new reserves. For unsuccessful (or dry hole) results, the associated exploration costs are immediately expensed. The full cost method capitalizes all exploration spending regardless of the outcome. The use of full cost accounting tends to smooth out charges against earnings from year-to-year and quarter-to-quarter. Hence, small oil and gas producers, whose earnings can be significantly impacted by a costly dry hole, tend to favor full cost accounting. Ratio of probable to proved reserves Expected Prices and Quality of estimates Evaluators Reserves Committee Forecast prices vs. constant prices (Marc’s subsample) Firms’ legal form: As explained in Badia and Duro (2011), the two major legal forms in the Canadian O&G industry are trusts and corporations. An open-ended investment trust is a legal structure that holds income-producing assets and pays income to unit-holders through 50 Proposal for Discussion Purposes Only distributions. Units trade like stocks. Distributions are tax deductible for the trust, eliminating corporate taxes, so it establishes a single level for distributions to the unitholders. This is the key difference between trusts and corporations. As of October 31st 2006 the Canadian federal government revoked this preferential tax treatment for income trusts, imposing a distribution tax on the income distributed to the unitholders. As a result, the trust would end up paying the same amount of a corporation. The new tax treatment applies for new trusts as of 2007, and its application is deferred until 2011 for trusts that were publicly traded at the date of announcement of the revoke (PriceWaterhouseCoopers, 2008 April). Almost all the trusts were converted into corporations in 2011 (exceptions two foreign trusts that using some loopholes still have a preferential tax treatment). Canadian royalty trusts are different from U.S. royalty trusts (Badia and Duro, 2011). The U.S. trusts pay out the cash flow generated by their O&G properties, but they do not acquire new properties. Consequently, their cash flow declines over time as their assets are depleted. Canadian trusts, by contrast, try to replenish depleted properties with new acquisitions. Since royalty trusts distribute most of their income to unitholders, they must raise cash to fund acquisitions either by borrowing or by selling more units. There are not trusts in the TSX Venture Stock exchange, since firms trading in that exchange have less stringer requirements, and most of the times they are exploration companies with a small volume of resources, so they hardly have cash flows to distribute. Stock Exchange There are two main stock exchanges in Canada: the TSX (the main market) and the TSX Venture68. The TSX Venture provides growth capital for early stage companies. On December 31, 2010, there were 1516 issuers listed on the TSX (from them 491 graduated from TSX-V) and 2154 issuers listed on the TSX-V. Using the definitions of small and microcap firms given by the Advisory Committee on smaller companies (SEC, 2006), Nicholls (2006) estimates that about 98% of TSXV companies could be considered microcap companies. The median (average) market capitalization for TSX firms was $152.8Mn vs. $8.2Mn ($32.7Mn vs. $1.45 Bn) for TSX Venture firms The TSX-Venture (Tier 1) has very low initial listing requirements: new companies can list with either (i) $500,000 in proved developed producing reserves or (ii) $750,000 in proved plus probable reserves. On the contrary TSX firms need to have at least $3Mn in proved reserves Hence, the main differences between firms traded in the two exchanges are the size (in our sample as of December 31 2010 XX% of the TSX-V firms have lower than $50 Mn, vs XX% at TSX), the analyst coverage (for less than $50Mn, the average number of analysts is 2, vs from 3 to 12 for firms with higher than $150 market cap, as per TMX), and the liquidity (note that TSX-V does not have market makers). 68 Additionally TSX-V can be tier 1 and 2. The tier 1 have stringer requirements 51 Proposal for Discussion Purposes Only December fiscal year-end firms One of the main differences between December fiscal year and the other firms is the possibility of different O&G future price expectations impounded in the calculation of the fair value. Credit adjusted returns (discount rate variability): The fair value is calculated with a standard 10% discount for all the firms. But firmspecific discount rates may differ from 10 percent, causing value relevance differences due to this omitted variable (Boone, 2002). I proxy for the degree of distortion introduced by the uniform 10 percent discount rate, calculated as βr - 0.10β, where r is the credit adjusted rate free rate the firm uses in accounting for asset retirement obligation. If that data does not exist I use the debt rate + the median spread for the remaining firms between the median industry debt rate and the median credit adjusted rate. D = 1 if the value of “rate” is above its cross-sectional sample median value, and 0 otherwise. Price volatility at disclosure date: I measure the Price volatility69 as the standard deviation of the annualized standard deviation in the natural logarithm of the oil price relative, calculated as πΜπ √250, where πΜπ is the estimated standard deviation of the log-transformed ratio of day t to day t-1 spot price of oil and gas (Boone, 2002). It is calculated across days 0 to -59 where day 0 corresponds to the Annual Information Form date disclosure. The subscript “i” corresponds to firms-years. Since firms have a different percentage of Oil and Gas reserves I calculate the weighted average: πΜπ = ππππ,π,π‘(2π) ∗ πΜ πππ + πππΊ,π,π‘ (2π) ∗ πΜ ππΊ ωOil,i,t≡ percentage of oil 2P reserves over the total 2P reserves for firm i at year t. I use WTI70 as the benchmark. ωNG,i,t≡ percentage of natural gas 2P reserves over the total 2P reserves for firm i at year t (gas leverage). I use Henry Hub as the benchmark. Then, I use D≡ 1 if the value of OIL-VOLATILITY is above its cross-sectional sample median value, and 0 otherwise. 69 It can be used the O&G VIX, but the data is only available from 2007 I calculate the correlation from 1996 to 2011 and from our sample period (2003-2011) between WTI and the other benchmarks (Brent and Edmonton) and the correlation is 1 and 0.88 for the long period, and 1 and 0.74 for the sample period. Analogously, the correlation between Henry Hub (natural gas) and other benchmark (AECO) is 0.99 for both periods. This is consistent with conversations with the evaluator firm Sproule, in which they states that “all the other Crude prices (different qualities and International prices too) are derived by applying historical average offsets against WTI. Our only exception is in the case of Edmonton Par, where we apply the actual transportation charges and exchange rate adjustment to WTI.” 70 52 Proposal for Discussion Purposes Only Price and/or quantity changes: Prior evidence suggests that managers engage in strategic earnings management during periods in which their environment is volatile or changing (Cornier and Magna, 2002). This conclusion is consistent with investors paying less attention to the precision of the fair value when there are other effects that mask the precision effect. Basically, the E&P firms’ performance is based on the compounded effect of the price and the quantity of O&G barrels. So it is possible that when there are big price increases or the quantity of barrels have been increased via acquisitions, discoveries, etc. investors neglect the importance of the technical revisions, not only directly, as a decrease in quantities, but indirectly, as a quality or precision signal of the remaining quantities. Other Amount of resources71 After taxes 1P for a subset (2009, 2010 and some firms that Marc has got). IX. Conclusions, caveats and future research Conclusions (to be written) π Μ 2 Caveats (to be written) Future research could examine the other two main criticisms to fair value estimates in the presence of historic cost estimates and with extant fair value probabilistic disclosures. Those criticisms are the unintended effects on real activities and the allegedly induced stock market volatility. The first criticism is pertinent to this industry, since O&G firms take long term decisions of investing and financing. Therefore, fair value disclosures and big short-term swings in O&G prices could have an unintended effect of shortening the horizon of their exploration and production planning72. In Canada, since the reserves are calculated using forecasted O&G prices for a long horizon, we may have the opposite effect, that is, an attenuation of the effect of the O&G price swings in the real investment decisions. So it would be interesting to analyze whether reserves calculated using forecasted cash flows and prices have a real effect in the investment activity of those companies or, alternatively, companies follow spot O&G prices for E&P decisions, regardless of the fair value disclosures. The latter would invalidate the role of the disclosures as a catalyst of investment decisions, and would imply that the cause of the lack of investments is more 71 The widespread implementation of horizontal drilling, combined with the use of hydraulic fracturing, helped unlock enormous quantities of tight gas trapped in shale rock formations thousands of feet below the surface. It soon became apparent that a horizontal well could provide 5x the production of a vertical well at 3x the cost. At that point the industry had entered a new era (Analyst Report, Sterne Agee, Dec 15, 2010) 72 Even more when using discounted cash flows for valuation, since the potential time premium on options embedded in the undeveloped or out-of-the-money reserves is not recognized. This effect could have played a role soothing the effects of the O&G price swings. 53 Proposal for Discussion Purposes Only related to myopic decisions, or as per Robert Hertz, rational decisions when investors see their wealth decrease. Concerning the second criticism, since O&G is a price-taker industry, the volatility of the O&G price and its effects in earnings, cost of debt73 and reinvestments are material. It would be interesting to test whether the disclosures using forecasted O&G prices make any impact on reducing the so-called induced spurious stock market volatility. Critics have argued that disclosing fair value using current spot O&G prices would cause artificial volatility in earnings and reserves estimates, inducing in turn undesired stock returns volatility. O&G industry players are price takers, so it seems reasonable to assume that the market (implied) O&G price expected volatility or the fair value disclosures do not affect the spot O&G price. On the contrary, the spot price will likely affect the expected volatility and the fair value disclosures. Analogously, stock market volatility has no effect on fair value disclosures (although we could find correlation, the causation goes in the opposite direction). What it is a priori unclear is the direction of the causation between fair value disclosures and expected O&G price volatility74, and between expected O&G price volatility and stock market volatility. Hence, the analysis of the impact of the fair value disclosures on stock market volatility either directly (if stock investors directly look at fair value disclosures to make stock decisions) or indirectly through the options and futures market, whose effects on the stock market are captured by the implied expected volatility (if stock investors look at the expected volatility extracted from the Options and Futures market, which is influenced by the fair value disclosures) is an open empirical question and a potential interesting research avenue. 73 Higher volatility makes companies to be perceived as riskier, increasing their cost of debt. We have to notice that although the forecasted prices used to value the reserves take into consideration firm’s physical O&G hedging contracted, it also includes firm’s own price estimates and adjustments (price deck). The adjustments are performed to reflect the different firm’s price realizations: the evaluator needs to determine a price adjustment to apply to the benchmark crude, adjustments that reflect the unique location of the crude oil production, the quality of the oil produced, price controls as well as inflation assumptions. Hence, the prices inputted in the model may vary with respect to forecasted prices implied from options and futures instruments. As a conclusion, the forecasted prices used can be only partially understood as the possibility of an investor of buying the security at these forecasted prices and also hedging in the market against deviations from these forecasted prices. 74 54 Proposal for Discussion Purposes Only Variable Definitions: Board Lot A standard trading unit as defined in UMIR (Universal Market Integrity Rules). The board lot size of a security on Toronto Stock Exchange or TSX Venture Exchange depends on the trading price of the security, as follows: ο· ο· ο· Trading price per unit is less than $0.10 - board lot size is 1,000 units Trading price per unit is $0.10 to $0.99 - board lot size is 500 units Trading price per unit is $1.00 or more - board lot size is 100 units 55