Fair-value Pension Accounting, Corporate Risk Management and Pension Investment Policy

Transcription

1 Fair-value Pension Accounting, Corporate Risk Management and Pension Investment Policy Yong Li Department of Management King s College London London SE1 9NH Phone: +44(0) yong.li@kcl.ac.uk Preliminary Draft This Version October, 2013

2 Fair-value Pension Accounting, Corporate Risk Management and Pension Investment Policy Abstract This paper investigates whether employer sponsors manage their pension risk exposure as an integral part of firm risk, or manage these two risk exposures separately over an extended period ( ) in the UK, when transition from a disclosure to recognition pension accounting regime is taking place. We test three competing explanations of corporate pension investment policy: coordinated risk management, contribution volatility, and separation hypotheses. We predict and find a negative association between the systematic pension risks and prior year s firm risk and operating asset risk during the period of recognition under fair value pension accounting (IAS 19). These findings hold after controlling for other economic determinants identified by prior literature as relevant for pension investment policy. Our empirical evidence supports the view that fair-value pension accounting has real economic consequences, and pension investment policy is a dynamic process that is ultimately determined by strategic corporate risk management considerations. Keywords: pension risk, fair-value, pension accounting, corporate risk management JEL Classifications: M41, J53

3 1. INTRODUCTION The growing size of pension plans assets and liabilities in relation to the market capitalisation of sponsoring companies raises the possibility that firms overall financial and risk positions may not only influence its strategy for funding their pension liabilities, but also its pension risk management policy. Theoretical research prior to the 1990s has developed competing hypotheses to explain pension asset allocation decision from a corporate financial perspective (Black, 1980; Tepper, 1981; Treynor, 1977; Sharpe, 1976; Harrison and Sharpe, 1983). The tax-based Black-Tepper hypothesis implies that firms with over-funded pension plans should invest in the most heavily taxed assets (such as bonds) to maximize their tax-savings because over-funded plans are less likely to default on their pension promises (Black, 1980; Tepper, 1981). The pension put hypothesis (Sharpe, 1976; Treynor, 1977) implies that firms with under-funded pension plans should invest more in riskier assets (such as equities) to maximize the value of the put option to default. However, the competing theoretical hypotheses lack consistent support from the small volume of empirical research on pension asset allocation (Friedman, 1983; Bodie et al., 1987; Amir and Benartzi, 1999; Frank, 2002; Rauh, 2009). Rauh (2009) shows that investment strategies of pension plans are not consistent with the risk-shifting behaviour as predicted by pension put hypothesis, but rather are chosen to minimize expected financial distress costs. The corporate asset allocation strategy over alternative asset investment categories (equities versus bonds) is not well understood based on findings from prior empirical literature. A recent US-based study (Jin, Merton, and Bodie, 2006) provides evidence suggesting the systematic risk of US firms reflect the risk-taking in their sponsored pension plans during the period from 1993 to Jin, Merton and Bodie (2006) emphasize that values transparency (i.e. accounting disclosure of pension liability values) does not equate to or fully reflect risk transparency. Their research design has been motivated by the implicit assumption that management of sponsored pension plans is independent of the firm s financial strategy (Moore 1

4 and Pruitt, 1986; Stone, 1987). These traditional explanations applied to US settings when pension funds are over-funded and surplus reversions were common and arcane pension accounting kept a firm s pension assets and liabilities off-balance sheet. By contrast, in the UK, the mandatory compliance with new IFRS pension accounting standards (IAS 19, a similar version of which is now being considered by the FASB in the US), has not only transpired the risks of defined benefit pension sponsorship to investors, but also brought pension deficits onto the sponsoring firms balance sheet. To exploit the unique UK reporting environment, and extend prior research, this paper investigates two research questions: whether firms manage their pension risk exposures in the context of all the firm s assets and liabilities or they manage their pension risk independent of overall corporate risk management strategy? And whether the consolidation of pension assets and liabilities onto balance sheet has real economic impact on pension risk management policy? This paper seeks to document empirical regularities in firms management of pension risk exposures, and its interrelationship with corporate risk management over an extended period ( ), when transition from a disclosure to recognition pension accounting regime is taking place. Controlling for factors known as influencing pension investment policy postulated by prior research, the empirical evidence shows a significant negative relationship between the pension risks and prior year s firm systematic risk and operating asset risk under the pension recognition regime during the post-ias 19 period, suggesting that UK sponsors manage their pension risk exposure as an integral part of corporate risk exposures, rather than managing these two risk exposures independently. This paper extends prior research on corporate pension investment policy in a number of ways. First, our research design captures variations in risk exposures via pension investment policy using pension beta, rather than the conventional equity percentage used in prior pension research. Second, no prior research has examined what determines firms management of their pension risk exposures in a financial reporting environment where pension assets and liabilities are consolidated onto the balance sheet, and in settings where firms are underfunded, and where 2

5 managers may face incentives to increase their discretion over critical pension assumption parameters that are available under new fair value reporting rules. Third, this study directly addresses an important empirical question, that is, whether pension risks are corporate risks. Prior research (e.g. Jin, Merton, and Bodie, 2006) has not explicitly tested the risk management implications of integral versus separation perspectives underlying a firm s defined benefit pension sponsorship. The remaining sections of this paper are organized as follows. Section 2 provides a brief overview of prior research on this topic and explains in detail the institutional setting in which UK firms have migrated to IFRS fair value pension reporting regime. Section 3 develops hypotheses that differentiate between pension-firm risk relationships implied by competing perspectives of optimal corporate risk management policy. Section 4 describes the sample, data and research design. Section 5 presents the empirical results and finally section 6 provides a summary and conclusion. 2. BACKGROUND AND PRIOR RESEARCH This section introduces the institutional and theoretical background required to understand how the lack of transparent pension information may prevent investors from fully processing pension risks. Section 2.1 discusses the institutional issues and section 2.2 reviews the prior research Corporate pension schemes and evolution of pension accounting Defined benefit pension schemes provide employees with retirement benefits linked to the final salary upon retirement or to the date the employee terminated its employment with the company. To understand the extent to which US and UK legislative features governing private pensions share broad similarities, a comparison of regulatory regimes of both countries is useful. The similarities include the fiduciary duty of employer sponsor to make up the funding shortfall in the plan assets. The minimum funding standards set out by the ERISA 1974 are slightly more 3

6 stringent than the equivalent UK statutory funding requirement effective under Pension Act And the pension fund asset allocation strategies in both countries are broadly similar with a majority of investment in equities, and secondarily in bonds. In terms of benefit guarantee, ERISA 1974 established the Pension Benefit Guaranty Corporation (PBGC), which guaranteed the payment of defined benefit pension benefits in case of insolvency of employer sponsors. 1 In March 2005, UK government established a new pension regulator, Pension Protection Fund (PPF), similar in nature with PBGC in the US, which guarantees defined benefit pension obligations. 2 Prior to the 1980s, pension accounting in the UK was based on the cash contributions made by employer sponsors during the accounting period. Statement of Standard Accounting Practice 24 (hereafter SSAP 24) Accounting for Pension Costs was issued in 1988 by the UK Accounting Standards Committee (ASC), the predecessor body to the UK Accounting Standards Board (ASB). SSAP24 emphasized the primary accounting objective of maintaining a smoothing stream of regular pension cost each year, and any variations from the regular costs were recognised gradually over the remaining service life of employees. However, SSAP 24 did not address the issue of the appropriate treatment of the pension asset and liability on an employer sponsor s balance sheet. On 30 November 2000, the Accounting Standards Board (ASB) issued Financial Reporting Standard 17 ( Retirement Benefits, FRS 17) which represented a significant pension accounting regime change in the UK. FRS 17 adopts the fair value approach in measuring pension assets and liabilities and to achieve transparency and comparability in accounting for pension costs (FRS 17, para 1). One of the most significant and controversial changes required the recognition of the pension surplus or deficit as a net pension liability/asset on balance sheet (FRS17, para.37). The surplus or deficit is required to be valued annually, using market prices for pension 1 See ERISA 1974 section 4002(a). 2 The PPF protects 100 per cent of the pension for members above scheme pension age, and 90 per cent of the promised pension for members below scheme pension age (up to a maximum of 25,000 at age a65) using a mixture of scheme individual rates and standardised rules. The PPF is largely funded by charging a levy on participating pension schemes. 4

7 investments and the AA corporate bond rate to discount pension obligations (FRS17, para14). The recognition of this pension balance in a firm s financial statement is consistent with the corporate finance perspective which would treat the pension fund as part of the employer firm s net worth, requiring full consolidation on its balance sheet. In 2002 the ASB delayed the mandatory implementation of FRS 17 in order to allow UK and International Accounting Standards Boards (IASB) an opportunity to agree how to converge their different approaches to the recognition of actuarial gains and losses on post-retirement benefits for pensions. All listed companies across Europe are required to adopt IAS 19 Employee Benefits in their consolidated accounts from 1 Jan 2005,, as part of EU-endorsed International Financial Reporting Standards ( IFRS ). Prior to IAS 19, UK companies have the choice under the transitional regime of FRS 17 to recognise the net pension liabilities (assets) on balance sheet; or continue to report under SSAP 24, and provide comprehensive mandatory disclosures in pension footnotes the effects on the main financial statements as if FRS 17 had been fully implemented. From 1 Jan 2006, all UK firms adopting IAS 19 had to recognise the pension assets and liabilities on balance sheets, and most of UK firms have opted to recognise actuarial gains and losses in full in the other comprehensive income. The US pension accounting rules are consistent with IAS 19 approach in mandating balance sheet recognition of the net pension liability/asset. In September 2006, the US FASB adopted SFAS 158 (Employers' Accounting for Defined Benefit Pension and Other Postretirement Plans, an Amendment of FASB Statements Nos. 87, 88, 106, and 132(R)), which requires balance-sheet recognition of the funding status of defined benefit pension and OPEB plans Prior research Very limited amount of empirical research focused on explaining corporate asset allocation decisions. Bodie et al. (1987) find that the proportion of assets allocated to equities is negatively related to the level of funding and positively related to the size of the company. In other words, under-funded plans tend to hold more equities and less fixed income securities. The negative 5

8 correlation between funding and the proportion of assets allocated to equities provides some support to the pension put hypothesis. Their findings on asset allocations taken together suggest that firms do not manage their sponsored pension funds as if they are entirely separate entities from the sponsor (Feldstein and Morck, 1983; Bodie et al., 1987). Nevertheless, US-based empirical research during the 1990s provides no evidence supporting a tax arbitrage based pension investment strategy. Motivated by the inconsistency among theoretical and empirical studies, Frank (2002) re-examines the extent to which taxes affect a firm s decision to allocate its defined benefit plan s assets between equity and bonds within a simultaneous system of equations, which attempts to capture the joint corporate capital structure and pension asset allocation decision in such arbitrage strategy. Contrary to prior empirical research, Frank (2002) finds evidence consistent with firms trading off tax benefits and non-tax factors as described by Black (1980). SFAS 87 allows US firms to keep most of pension-related information off-balance sheet and only additional minimum pension liabilities are recognised into balance sheet. Thus it represents a compromise between the integration and separate legal entity perspectives of employer sponsored pension plans. Such lack of transparency in US firms pension exposure raises the question of whether investors can accurately process the complexity inherent in pension accounting and fully understand and apply it in valuation of share prices (Glaum, 2011). A body of empirical research has emerged to determine capital market participants assessment of the value-relevance of pension assets and liabilities to the sponsoring firm s shareholders. Findings from these studies suggested that pension and postretirement benefits information is not fully impounded in share prices, consistent with some degree of capital market inefficiency. Coronado and Sharpe (2003, 2008) find supportive evidence that US capital market participants failed to interpret the pension expense correctly throughout the late 1990s leading to the over-valuation of firms. Franzoni and Marin (2006) find that US firms sponsoring underfunded defined benefit pension plans appear to be undervalued relative to those sponsoring overfunded pension plans. 6

9 Picconi (2006) investigates whether analysts and investors fully incorporate the information contained in pension footnote disclosures. His results indicate that analysts do not explicitly incorporate the information from changes in pension plan parameters into their initial forecasts so that these changes predict future earnings surprises. In addition, Picconi finds that the offbalance-sheet portion of the pension plan s funded status and the projected benefit obligations (PBO) are predictive of future returns while the on-balance-sheet portion of the funded status is not. This implies that investors do not accurately assess the long-run cash flow and earnings implications of these off-balance-sheet pension disclosures. Jin, Merton and Bodie (2006) argue that value transparency does not equate to risk transparency, as the former is a static measure but the latter dynamic. Their study differed from the prior studies in that Jin, Merton and Bodie (2006) focused examining whether the systematic risk of the firm s stock returns (instead of the firm s market value) reflects the systematic risk of the pension plan. For a sample of large Compustat firms with pension asset allocation data (Form 5500) during period 1993 to 1998, they found that the beta of the firm s stock reflect the beta of the pension plan. Most interestingly, their findings suggest this contemporaneous cross-sectional positive relationship between pension risk and firm risk holds for sponsors during time periods when their pension plans are mostly overfunded and in a reporting environment when pension liabilities are off-balance sheet liabilities. This study extends their line of inquiry by examining directly the important empirical question of whether pension risks are corporate risks, and the interrelationship between firm risk, operating asset risk and pension risk against a rich pension reporting environment when pension assets and liability transitioned from off- to on- balance sheet items, and when pension plans are mostly underfunded. 3. HYPOTHESES This section develops alternative hypotheses on pension and firm risk relationship from alternative perspectives of corporate pension sponsorship. 7

10 3.1. Coordinated Risk Management Hypothesis Financial economists model a firm s exposure to defined benefit pension plan obligation as if it is integrated or consolidated on the firm s own balance sheet (Black, 1980; Tepper, 1981). This integrated perspective concerns managing the firm s extended balance sheet, including both its conventional assets and liabilities, and its pension assets and liabilities, in the best interests of shareholders. Under this view, the net pension liability is simply a form of nominal debt and thus should reduce a firm s market value dollar for dollar. The adoption of fair-value pension accounting reinforces and reflects the integration (corporate finance) views of the pension contract. In contrast, prior to IAS 19, firms may believe that markets do not properly impound the extent of its pension plans systematic risks, thus choose pension investment policies that have substantially mismatched risks with aims to maximize asset returns but at the same time adding risks to pension plans. From the risk management perspective, the mandatory adoption of IAS 19 may alter the nature or perception of risks of employers pension exposure. UK sponsoring firms are required under IAS 19 to recognize their past funding practices on a fair value basis onto the balance sheets. Consolidation of pension risk onto balance sheet may facilitate a shift towards managing their pension risk exposure in the context of entire risk of the corporate asset base (ie all the firm s assets and all its liabilities). One possible pension risk management strategy is to take an integrated perspective in managing firms pension and operating asset risk (Jin, Merton, and Bodie, 2006; Merton, 2007). If a sponsoring firm s non-pension risk is high, lower pension risk exposure can mitigate net operating asset risk exposure, which in turn would reduce firm s total asset risk. Such coordinated risk management can be attained by choosing a different mix of equities and bonds, thus maintaining an optimal asset exposure which is negatively correlated with the operating assets of the firm. Froot et al. (1993) risk management framework on financial institutions bear similar implications on our coordinated risk management hypothesis in that Froot et al. (1993) suggest that if a firm allows internal funds to run down, it will have to choose between cutting highly rewarding investments and incurring the high costs of external finance. 8

11 Thus, with capital-market imperfections, and exposure asymmetry, a firm can dynamically adjust its pension investment to the firm s own investing opportunity. 3 Prudent value-maximizing managers may face strong incentives to engage in coordinating the management of pension risks, which is attributable largely from mismatched pension assets (invested in equities) and liabilities (bond-like), with management of operating asset risk exposures under a fair-value recognition pension accounting regime, and when pension plans are mostly under-funded. Following Jin, Merton and Bodie (2006), a systematic risk measure of pension plans (PENBETA) is constructed to capture variation in firms exposure to asset and liability mismatch risk. PENBETA is calculated as pension asset beta minus pension liability beta, adjusted by the value of pension assets and liabilities as a percentage of a sponsor total market value. We use the volatility of the sponsor s operating cash flow (STDCF) to proxy for its operating risk. STDCF is calculated as the standard deviation of operating cash flows over the preceding 8 years deflated by the book value of equity. Hypothesis 1 (H1a): Ceteris paribus, the systematic risk of pension plan is negative related to the sponsoring firm s operating risk during the post-ias 19 period. Hypothesis 1 (H1b): Ceteris paribus, the systematic risk of pension plan is negative related to the sponsoring firm s systematic risk during the post-ias 19 period Contribution Volatility Risk Risky assets, such as equities, are characterized by their volatile returns. Financial theory suggests that higher risk of equity investment is awarded by the higher return it generates (Markowitz, 1952; Sharpe, 1964). However, if the volatile return on the pension assets translates into volatility in the required cash contributions, then risky assets will translate into more riskiness of contribution flows to the pension plan. The main concern of cash flow risk managers is to minimize the volatility of changes in cash flows (Culp, 2001). Friedman (1983) 3 The Actuary, March 2008, notes that the Froot et al. (1993) framework can potentially explain the risk management relationship between a firm and its sponsored defined benefit pension fund. 9

12 finds some evidence that firms have incentives to time their pension contributions so as to smooth the reported earnings. The lack of transparency in pension financial reporting systems meant many stakeholders did not understand the risks in DB pension plans. Shareholders and investors may judge the firm s performance by its reported earnings rather than by more comprehensive cash flow measures. Coronado and Sharpe (2003, 2008) show that investors failed to distinguish between pension and operating earnings and capitalize them similarly. Enhanced transparency under IAS 19 may facilitate the investors awareness of the impact of pension mismatch risk. Consequently, UK firms also face incentives to hedge the riskiness of their future cash contributions via immunization, ie duration-matching pension assets with liabilities. Duration-matching allows sponsoring firms effectively reduce the volatility of their pension contributions, thus dynamically hedge their cash contribution risks. If the incentive is strong for sponsors to minimize the volatility of pension contributions, then it would be observed that firms with both extremely over-funded and under-funded plans invest in bonds because such extreme over-funding and under-funding afford less flexibility to adjust the timing of pension contributions than do firms with moderate funding levels. Pension plans with greater deficits are required to make deficit-reduction contributions and those with greater surpluses have to conform to the tax regulations. 4 By contrast, pension contributions are fairly predictable for moderate funding levels, but less predictable when funding levels become more extreme. The contribution risk hypothesis thus predicts a nonlinearity relationship between pension funding level and pension risk, and provides an alternative contribution risk explanation for the conflicting results from prior studies on the effect of pension funding on asset allocation. Amir and Benartzi (1999) find some empirical evidence on such a nonlinear relationship between pension funding and asset allocation. Hypothesis 2 (H2): Ceteris paribus, the systematic risk of pension plan increases as the 4 UK corporate sponsors of defined benefit pension plans with a funded status in excess of 105 percent were subject to taxation at a rate of 35 percent. 10

13 funding level increases up to a specific point, then decreases as the funding level increases beyond this point (i.e. a non-linear relationship) Separation Hypothesis The alternative, separate entity, view implies that the pension fund is a separate legal entity from the employer sponsor. Therefore the net surplus or deficit, defined as the fair market value of plan assets minus the projected benefit obligation (PBO), would not be shown on the employer s balance sheet. In other words, plan assets are separate pools of capital that collateralize the future pension liabilities. The separate legal entity view assumes that pension funds should be managed without regard to either corporate financial policy or the interests of its shareholders. The risk management implications of the separation view are that firms should manage their pension risk exposures on the basis of the expected future stream of employee pension liabilities, and in the best interests of plan beneficiaries. Sponsoring firms are willing to honour the ongoing needs to fully fund future pension benefit accruals together with the existing benefits of the plan. Blake (2003) asserts that matching pension liabilities with fixed income securities of similar duration (i.e. immunization of the pension liabilities) in pension investment portfolio, and consequently reducing pension risk exposures, should provide maximum assurance of benefit securities with plan participant. Such pension risk management strategy is independent of firms assets and liabilities, and thus has zero correlation with exposures of a firm s net operating asset risk. As a result, our null hypothesis, separation hypothesis, predicts that firms pension risk is not associated with either systematic beta of the firm, or pension funding ratio. Hypothesis 3 (H3): Ceteris paribus, there is no relationship between systematic risks of pension and firm under either pre-ias 19 or post-ias 19 period. 11

14 4. DATA AND METHOD 4.1. Sample and Data Our initial sample includes the 350 UK companies listed on the London Stock Exchange, which were constituents of the FTSE 350 index during the period from 2003 to We exclude firms from the initial sample that did not sponsor defined benefit pension schemes, and firms that are financial firms and investment trusts, as well as firms following US GAAP and firm-year adopting FRS 17 early during our sample period. We collect pension fund related variables, firm price and return, other firm characteristics data from Thomson Reuter s Data Stream. This sampling procedure yields a primary sample of 1431 firm-year observations. In our regression analysis, we use firm-year observations with available data and delete the firm-year observations with negative book value of equity. The resulting number of sample firms varies over time, with a minimum of 169 in 2003, and a maximum of 186 in Estimation of Firm Risk To operationalize the coordinated risk management hypothesis, we estimate firm risk using Capital Asset Pricing Model (CAPM) framework, following methodology in Jin, Merton and Bodie (2006). From the balance sheet identity we get the following relationship: OA + PA = E + D + PL where OA is the value of firms operating assets, E is the value of equity, D is the value of debt, and PA is the value of pension assets, and PL is the value of pension liabilities. We first calculate firms unlevered equity beta using the market model, as a variant of CAPM: E (r i ) = α i + β i E (r m ) where E (ri) is the expected stock return for firm i in period t, and E (r m ) is the return on the market portfolio m in period t. Monthly returns on specific stocks in our sample are regressed on a market risk factor proxied by a value-weighted index return (FTSE 350) to obtain the beta coefficient for each firm for the sample period. We exclude the stocks which are thinly traded 12

15 (see Dimson, 1979) to avoid the potential biases caused by nonsychronous trading. Consistent with Jin, Merton and Bodie (2006), we define the firm risk as the systematic risk borne by the equity and debt holders of the firm. Rearranging the balance sheet identity ( OA + PA = E + D + PL ), we have E + D = OA + PA PL where E + D is firms invested capital which equals to the value of operating assets plus PA PL, the net pension surplus (deficit). Therefore, the systematic risk of firm assets, FIRMBETA ( β E+ D ) is calculated as the market value weighted average beta of debt and equity, and is given by the following equation: E E + D E + D = β E + β D D E + D β (1) 4.3. Estimation of Pension Risk After estimation of firm risk in equation (1), we follow the balance sheet identity ( OA + PA = E + D + PL ) and it gives: β E + D = β PA PA E + D β PL PL E + D β OA OA + E + D Thus, net pension plan risk is measured by the following equation: Pension β PA PA β PL PL = E + D E + D β (2) where PENBETA (β Pension ) it is estimated as pension asset beta minus pension liability beta, adjusted by the value of pension assets and liabilities as a percentage of sponsor total market value. β PA is the pension asset beta, or the weighted average beta of all asset classes in a pension fund. The weight of each asset class is collected from firms annual reports. The betas of individual asset classes are from Jin, Merton, Bodie (2006, Table 4). β PL is estimated using 30- year UK treasury bond rate as the benchmark for pension liabilities. The estimated pension liability beta is 0.38 if we do a 60-month rolling regression estimation. We also use pension 13

16 liability beta of 0.18 used in Jin, Merton and Bodie (2006) as a robustness check in our reported results, as they note beta of pension liability varies, depending on the estimation method. As a result, compared with pension asset beta, pension liability beta estimates are noisier. The implicit assumption in our research design is that there are no effects of taxes and Pension Protection Funds (PPF) on firms that are not distressed for reasons consistent with Jin, Merton, Bodie (2006). 5 We compute three empirical measures to distinguish the financially distressed firms: (1) return on investment as a measure of operating business distress; (2) leverage as a measure of financial distress; (3) book to market value as a combination measure of both types of distress. We rank all firms in the previous year by each measure of the distress, and take the decile of firms with the most severe measure as distressed, and the rest as nondistressed Empirical models To investigate the relationship between firm risks and its management of pension risk exposures as reflected in pension beta, we estimate the following panel regression model: PENBETA it = α 0 + α1firmbetai, t 1 + α 2STDCFi, t 1 + α3fundi, t 1 + α 4t FUNDSQi, t 1 Φ Χ 1 i, t + ε t + (3) where the dependent variable is pension beta (PENBETA), which is calculated as the market value weighted average beta of pension asset and pension liability: β PA PA/(E+D) - β PL PL/(E+D). FIRMBETA t-1 is one-year lagged systematic risk of sponsoring firm, which is 5 Jin, Merton and Bodie discussed three primary reasons for the restriction to non-distressed firms: (1) the effect of the benefit guarantee provided by PBGC is potentially large, and must be taken into account. This is applicable in the UK after PPF became operational; (2) betas of the debt and equity of firms in distress are non-linear functions of the value of the firm; (3) book value of debt as approximation of market value works reasonably well when firms are not in distress. 6 Return on investment is calculated as net income divided by total assets; financial leverage is defined as debt divided by total assets; book-to market ratio is the book value of common equity plus balance sheet deferred taxes for fiscal year t-1, over market equity for December of year t-1. 14

17 calculated as the market value weighted average beta of debt and equity: β E E/(E+D) - β D D/(E+D). STDCF t-1 is one-year lagged standard deviation of operating cash flows (earnings before extraordinary items plus depreciation expenses) over the preceding 8 years deflated by the book value of equity. FUND t-1 is the reported funding status; FUNDSQ t-1 is the squared value of FUND; X it is a vector of control variables postulated by prior research as influencing pension risk-taking strategy. The lower pension beta, the better hedged is pension liability risks to its pension asset investment portfolios. We predict a negative coefficient on the prior year s firm systematic risk (FIRMBETA) and operating asset risk (STDCF) to the extent that our sample of firms coordinate the management of their pension and firm risks under the recognition regime (post- IAS 19). We use Fama-MacBeth (1973) methodology to estimate equation (3). Fama-Macbeth (1973) estimation procedure consists of two steps: in the first step, for each single time period a cross-sectional regression is performed. Then, in the second step, the final coefficient estimates are obtained as the average of the first step coefficient estimates. The standard error estimates incorporate the Newey-West (1987) heteroscedasticity and autocorrelation correction Control Variables Observed systematic relationships between systematic risk of pension plan and its sponsoring firm might be attributable to the effect of omitted correlated variables pension risk. We address this issue by including a comprehensive list of control variables posited by prior literature which could explain variations in pension beta risk. The pension put hypothesis (Sharpe, 1976; Treynor, 1977) implies that firms for which the pension put option is more valuable (that is, more in the money) will hold more of the most risky assets, presumably equities, and vice versa. The put option is likely to be more valuable for severely under-funded plans, for unprofitable companies, or for firms with more debt. Consistent with prior literature (Friedman, 1983; Bodie et al., 1987), two proxies used to measure the value of pension put option to sponsoring firms include leverage (LEV) and profitability (PROF). 15

18 The consolidated leverage (LEV) is calculated as the long-term debt divided by total assets. For firms adopting SSAP 24 during pre-ifrs period, we treat the pension liability (PBO) as a long-term liability. Higher leverage implies less debt covenant slack and higher probability of financial distress. Thus the put option is more valuable to highly levered firms. PROF is calculated as the mean return on share-holders equity over the preceding eight years, which is a proxy for long-term profitability. Less profitable firms are less likely to fulfil the fixed payments of retirees benefits and thus more likely to invest more in equities to maximize the value of the put option to default (Sharpe, 1976). It is important we also control for various economic determinants of pension asset allocation posited by prior research. Prior research find the plan demographics influences asset allocations (e.g. Amir and Benartzi, 1999; Friedman, 1983). The duration of the pension liabilities may be an important determinant affecting the asset allocation decision by UK sponsors in the current regulatory economic environment when most of funds gradually mature, demand non-trivial fixed benefit payments. Firms with mature pensions may wish to invest more assets in bonds so as to achieve better asset/liability matching, thus reduce the likelihood of the assets falling short of obligations. We construct a proxy as the duration of the pension liability (Duration) using the ratio of current service cost over the sum of current service and interest cost. Service cost is equal to the present value of the pension benefits earned by employees during the year; in essence, it is the cost of deferred compensation. Interest cost is calculated as the beginning-of-year value of pension obligations multiplied by the plan s assumed discount rate; this represents the cost of financing the outstanding pension obligation, that is, the increase in the benefit obligation resulting from the passage of time. It is also argued that UK pension fund management is partially driven by the herding behaviour (Klumpes and Whittington, 2003). Rauh (2009) also show that the lagged investment return is strongly positively associated with allocation to riskier assets such as equities. This suggests that asset investment policy may be driven by benchmarking their investment performance against prior period performances. The lagged investment return on pension asset 16

19 portfolio (LAGRTN) is included to control for this argument. LAGRTN is calculated as the actual return over the prior year s market value of pension assets. We also include plan size, (LNSIZE) measured as the natural logarithm of the fair value of plan assets as a control. It is possible that pension risk mismatch is attributable to managers accounting discretion over the long-term expected rate of return on pension investments (hereinafter ERR ). Prior USbased research has demonstrated that firm management has attempted to engage in smoothing and spreading of pension costs over time (Picconi, 2006; Bergstressser et al., 2006; Li and Klumpes, 2013). Therefore, an aggressive risk-taking pension investment policy could help firms to justify the choice of high ERR. We therefore included the ERR assumptions into our empirical model as a control variable Descriptive statistics 5. EMPIRICAL RESULTS Table 1 provides summary statistics for equity beta, firm beta and pension beta for estimating model (1). It shows a large variation in the pension, equity and firm betas. Further untabulated analysis shows variations in pension betas are largely cross-sectional. For any specific firm, the pension beta changes more slowly over time. [Insert Table 1 about here] Panel A, Table 2 presents the descriptive statistics on the distribution of pension asset composition in the sample. At the end of 2003, sponsoring sample UK firms allocate on average around 66 percent of their assets to equities and 27 percent to fixed income securities. By the end of 2010, the allocation to equities has decreased to 48 percent, and the portion of fixed income securities has increased to 37 percent. On average, the sample UK firms still invest significantly more in equities than in bonds. This evidence is consistent with the pension asset allocation of relevant comparative population in the US. US firms invest on average about percent of their total pension funds in equities (Rauh, 2009), and 30 percent in fixed income securities. Panel B of Table 2 presents the cross-sectional quintile distribution of %EQUITY by year. 17

20 %EQUITY varies significantly across the firms with a standard deviation of for the pooled sample, and the cross-sectional variations between firms increases year on year during the sample period. Overall the sample pension asset allocation exhibits a gradual and slow trend towards a greater percentage of bonds among the overall pension asset composition for the sample period ( ). [Insert Table 2 about here] To examine the frequency of asset allocation revisions, changes in %EQUITY are calculated over one, two and three years. Untabulated results suggest that most firms maintain a stable allocation to equities. Over a one-year period, more than 80 percent of the firms remained within 5 percentage points of their beginning allocation to equities. Over a three-year period, 80 percent of the firms decreased their allocation to equities by less than 12 percent, or increased it by less than 7 percent. This suggests that pension beta is likely to be correlated over time, given the relative stability of equity allocation over the sample period ( ) Univariate analysis Table 3 examines the relation between each of the key explanatory variables in equation (3) and pension risk (PENBETA), using a nonparametric portfolio analysis. Each independent variable is divided into five equal-size portfolios based on pension risk, where portfolio 1(5) contains firms with the lowest (highest) values. Consistent with hypothesis H1, the measure of firm systematic risk decrease monotonically as pension risk increases. Portfolio with highest exposure to pension risk (fifth quintile) has average firm beta of 0.364, whereas portfolio with the smallest exposure (first quintile) has average firm beta of The relation between the funding ratio of the pension plan (FUND) and pension risk is not entirely supportive with the non-linear relationship as predicted in hypothesis H2. Funding ratio increases from 78 percent for the first quintile to 89 percent for the fourth quintile, and then it only decreases slightly to 86 percent for the fifth quintile. There is a positive association between the proxy for the duration of pension plan and pension risks. Portfolio with highest exposure to 18

21 pension risk (fifth quintile) has significantly longer duration of pension liabilities than portfolio with lowest pension risk (1 st quintile). This evidence is consistent with prior findings in Amir and Benartzi (1999) and Peterson (1996) that firms with shorter pension liability duration (more mature aging distribution of plan participants) allocate more in bonds than equities. The effect of long-term profitability (PROF) and pension plan size (LNSIZE) are statistically significant in explaining pension risks. Portfolio with highest exposure to pension risk (fifth quintile) is significantly less profitable than portfolio with lowest pension risk (1 st quintile). This evidence is more consistent with pension put hypothesis that pension put option is more valuable for less profitable firms. [Insert Table 3 about here] 5.3. Multivariate regression analysis Table 4 presents regression results on determinants of pension beta using Fama and MacBeth (1973) methodology. The estimation is run for each year, controlling for fixed effect at industry level in the first step, then the final coefficient estimates are obtained as the average of the first step coefficient estimates. The regression coefficients and their t-statistics are reported in Table 4. Panel A, Table 4 reports the regression results for the sub-sample period prior to IAS 19 adoption. The coefficient on FIRMBETA is of predicted negative sign, but not statistically significant. Similarly, no statistically significant relationships are found with firms prior year operating cash flow volatility (STDCF), and pension funding ratio (FUND, FUNDSQ). Nonetheless, three explanatory variables are significant and of predicted sign under the pension disclosure regime: the prior year s actual rate of return on pension assets, and consolidated book leverage ratio adjusted for pension liability (PBO) as long-term liability, and duration of pension liabilities. The empirical evidence suggest that firms with higher prior year s pension investment return, higher expected financial distress risk, and longer pension liability duration assume higher pension risks. Furthermore, results show that pension liability duration is statistically significant in explaining pension systematic risk only under a pension disclosure regime. Other 19

22 things being equal, longer pension liability duration is associated with higher pension risks. To summarize the results, it appears corporate pension risk management policy is determined primarily by pension fund characteristics prior to firms adopting IAS 19, and pension risks and firm risks seem to be managed separately, consistent with our separation hypothesis. Panel B, Table 4 reports the regression results for the sub-sample adopting IAS 19/FRS 17. The negative and statistically significant coefficient on the prior year s firm systematic risk is supportive of H1b, which predicts firms with higher systematic risks takes less pension risk, but only under the recognition regime of pension accounting. One unit increase of the prior year s firm risk is associated with the decrease in pension risk by units. Furthermore, a firm with higher prior year s operating cash flow risk takes less pension risk. The significant negative relationship between pension beta and operating risk is supportive of H1a prediction that firms manage their pension risk exposure to correlate with their net operating asset risk. Taken together, the results are consistent with coordinated risk management hypothesis. This evidence suggests that pension liability recognition has real risk management consequences for UK sponsors, and pension risk exposures are managed as an integral part of firms total assets and liabilities risk management strategy, rather than being considered as two risk exposures separately. Another interesting finding is the significant negative relationship between firms long-term profitability (PROF) and pension risk during post-ias 19 period. This finding suggests that sponsors with higher long-run profitability take less pension mismatch risks, consistent with the univariate analysis. This evidence is in contrast with the risk-offsetting story advocated by Friedman (1983) that the less profitable firms face higher risk to default on fixed payments, thus prefer bonds to equities. But it is more consistent with the risk-shifting story which predicts that less profitable firms facing higher probability of default face greater incentive to shift risk to their pension funds to maximize the value of the put option. Other two explanatory variables are significant and of predicted signs under the pension recognition regime: the prior year s actual rate of return on pension assets, and consolidated book leverage ratio adjusted for pension 20

23 liability (PBO) as long-term liability. Overall evidence suggests that firm-level determinants play a more significant role in explaining pension investment policy during the post-ias 19 period. We do not find evidence suggesting the effect of funding level on pension risk follows a nonlinear relationship. H2 predicts that immunization (minimizing pension mismatch risk) can minimize the cash flow risk caused by volatile pension contributions, as substantially underfunded plans have stronger incentive to avoid the accelerated funding requirements. Conversely, substantially over-funded plans have stronger incentive to avoid exceeding the full-funding limits. Regression results suggest that the coefficient on FUND is positive and it is negative on FUNDSQ, but not significant. This evidence is not supportive of H2 under either recognition or disclosure pension accounting regime, and for the full sample. [Insert Table 4 about here] 5.4. Robustness check Endogeneity One potential issue with respect to the test above is that pension risk, firms long-term pension funding, the consolidated financial leverage and firm discretion over the long-term expected rate of return on pension investments ( ERR assumptions) could be endogenously determined. Corporate sponsors can exercise the discretion in the level of ERR assumptions to change the pattern and magnitude of pension liabilities, thus mitigating their contribution volatility risk. They could also increase the allocation of equities in their pension fund investment portfolio to justify the higher level of the ERR assumptions. Picconi (2006) shows that US firms engage in smoothing and spreading of pension costs over time. Klumpes, Li and Whittington (2009) provide evidence that UK firms pension termination decision is indeed not independent of their discretionary ERR choices. Consequently, we adopt the General Method of Moments (GMM) method of Arellano and Bond (1991) to control for the endogeneity in testing the coordinated risk management hypothesis. 21

24 Our GMM estimation consists of two steps. First, the regression equation is rewritten as a dynamic model that includes lagged firm risk (with one lag) as an explanatory variable. Second, the model is estimated by GMM and use lagged values of the firm risk, operating risk proxy, pension funding, and other firm and pension characteristics as instruments. Using lagged variables as instruments for the present values of these variables controls for potential simultaneity and reverse causality. In addition, this estimation procedure allows some of the explanatory variables (i.e., the firm risk proxy, pension funding, and ERR assumption) to be treated as endogenous. Intuitively, such a dynamic panel GMM setup corresponds to estimating a dynamic simultaneous equations system. Table 5 presents results from GMM estimation of equation (3) for the pre- and post-ifrs period, and for the pooled sample over the entire period. As compared with the Panel B of Table 4, the coefficient on firm risk remains significant at 5% level, but its economic magnitude is smaller (-0.02, t-statistics=-2.01) when the dynamic endogeneity, and simultaneity are accounted for based on the dynamic GMM estimator. Similarly, the coefficient estimate of operating asset risk proxy (STDCF) remains significant at 5% level with a smaller magnitude (-0.009, t- statistics=-2.26). Pension funding status remains non-significant in explaining pension risk management policy. This is probably attributable to the fact that most of our sample firms sponsor underfunded pension plans, and the funding effect may have been captured by the leverage ratio. [Insert Table 5 about here] Change analysis The previous analysis focuses on the cross-sectional relationship between the levels of pension and firm risk while controlling for a variety of observable pension and firm characteristics that might account for the variations in pension risk. To control for potentially omitted firm and/or pension characteristics, and further corroborate the core empirical results on the association between systematic pension and firm risk, we conduct a changes analysis to 22