Corporate financial hedging and firm value: a meta-analysis

Abstract

This study is a quantitative review of the empirical literature analyzing firm value effects of corporate financial hedging. Using meta-regression analysis to accumulate a hand-collected data set of 1016 estimates for the hedging premium reported in 71 previous studies, we find that reported firm value effects of hedging are systematically larger for foreign exchange hedgers as compared to interest rate and commodity price hedgers, for studies published in lower-ranked journals, and for models estimated without firm fixed effects and without controls for endogeneity. Our results also suggest that hedging premiums increase significantly when a study considers operational hedging strategies in addition to financial hedging. Moreover, we find evidence for a larger hedging premium in less developed financial markets and countries with higher tax rates. Aggregating the previous hedging literature and assuming a ‘best practice’ study design, we find an overall hedging premium of 1.8% for foreign currency hedgers and a firm value discount of −0.8% (−0.6%) for interest rate (commodity price) hedgers.

KEYWORDS:

• Financial derivatives
• shareholder value
• risk management
• meta-regression

JEL Classifications:

• C83
• G32

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Corporate hedging refers to all measures to reduce and control corporate exposure to risks. There are different strategies to hedge: Financial hedging makes use of derivative instruments or other financial hedging methods like foreign debt, foreign assets, or hedging substitutes such as cash management or dividend policy. Operational hedging refers to the use of various real options within a company, like opening a foreign subsidiary or flexibility in the adjustment of production volumes. The title of this study reflects that the majority of the articles included in our meta-analysis report estimates for the firm value impact of financial hedging activities, especially the use of financial derivatives. Nevertheless, the meta-analysis results also allow us to derive inferences about the impact of other hedging strategies.

2 The Online Appendix is available at https://myweb.rz.uni-augsburg.de/~geyerkje/research%20files/HedgeValue_Appendix.pdf.

3 The search term consists of keywords linking terms for hedging (hedging, hedger, risk management, derivatives, option, swap, forward, future) and firm value (firm value, premium, Tobin’s Q, market-to-book ratio).

4 Academic Search Premier, Business Source Premier, EconLit, Google Scholar, and the SSRN working paper database.

5 This is a common procedure in the hedging literature, which is reasoned by the fact that most financial firms are also market makers in derivatives markets (Allayannis and Weston 2001). Hence, their rationales for using derivatives may be different from non-financial firms.

6 If an estimate for the hedging premium refers to one specific country, we can clearly assign a value of the macroeconomic factors defined in Section 5.

7 Moreover, selecting just one estimate per study requires objective selection rules to decide which estimate to prefer and also leads to a loss of information about within-study variation (Stanley and Doucouliagos 2012).

8 Cross-country studies may also include country fixed effects.

9 As an extension of Equation (1), about 10% of estimates for $\hat{\beta }$ represent interactions of the hedging variable with other firm characteristics (e.g. capital expenditures). For these estimates, we follow Havranek, Horvath, and Zeynalov (2016) and evaluate the interaction term at the sample mean of the interacting variable to calculate the hedging premium. We then use the delta method to approximate the corresponding standard errors of the hedging premiums. For details, see Online Appendix B.

10 Publication bias arises when researchers or editors/reviewers discard undesired results from publication (Begg and Berlin 1988; Rothstein, Sutton, and Borenstein 2005; Stanley 2005). Undesirable outcomes might be statistically insignificant effects, outcomes without support of the ex-ante hypothesis, outcomes that are inconsistent with theoretical predictions, or outcomes that do not agree with what is found in the previous empirical literature. If uncontrolled, such an active selection of preferred statistical results might distort the summarized effects in a meta-analysis (Doucouliagos and Stanley 2013).

11 In addition, Online Appendix D outlines common issues occurring in a meta-analysis and how we approached them.

12 The journals in this category are The Journal of Finance, Journal of Corporate Finance, Review of Financial Studies, Review of Finance, Journal of International Economics, and Energy Economics.

13 We prefer a breakpoint dummy over the average sample year, as the average sample year is highly correlated with other variables exhibiting time trends (e.g. the macro variables explained in the next section).

14 This is typically measured by the actual quantity that is hedged (e.g. the volume of oil production hedged) divided by the actual risk exposure (e.g. the total oil production).

15 Tobin’s Q is defined by the ratio of the market value of financial claims and the replacement cost of the firm’s assets.

16 As ordinary least squares estimation of Equation (1) relies on the exogeneity assumption of the regressors, hedging premiums without accounting for endogeneity might be biased.

17 69% of the models with endogeneity correction use instrumental variables via two or three stage least squares regressions, 23% use generalized method of moment (GMM) estimation, the remaining 9% use other methods.

18 Following Allayannis, Ihrig, and Weston (2001), studies including a measure for geographical diversification are considered as controlling for operational hedging.

19 If a study’s sample period does not exactly correspond to the data available from the external sources, we follow Kysucky and Norden (2016) and use the closest available country-year observation.

20 This data is obtained from the Bank of International Settlements (BIS). As this information is only available on a triennial basis (starting in 1995), we estimate the missing annual values by linear interpolation.

21 The rule-of-law index measures the effectiveness of the legal system.

22 This is the sum of economic, financial, and political risk factors.

23 These indices are inverse measures of country risk, i.e. higher scores imply lower risk.

24 It should be noted that the number of observations including a broader hedging definition is rather small (5%). Thus, results for this variable can only be interpreted as a first indication of differences between derivatives usage and other hedging strategies.

25 It should be noted that the hedging premiums collected from primary models with a continuous hedging variable are evaluated at the sample mean hedging volume, i.e. they show the hedging premium the study implies for an average hedger. In contrast, the hedging premiums for the binary variable refer to the difference between hedgers and non-hedgers. Thus, the evaluation of the continuous variable at the sample mean might drive the systematically lower premiums as compared to models using a binary hedging variable. See also Online Appendix B.

26 In a robustness test (Online Appendix G), we examine the impact of major accounting changes and find that the hedging premium increased with each change. Moreover, we break down the continuous hedging measures in notional amounts, actual hedge ratios, fair values, and other hedging measures. The results suggest that fair values and actual hedge ratios yield systemically lower hedging premiums of 2.2% and 1.9% as compared to notional amounts.

27 See Online Appendix E for the correlation matrix.

28 A similar finding was recently exposed by Harvey, Liu, and Zhu (2016) for factor studies on the cross-section of expected returns.

29 The arithmetic mean across all hedging premiums is 9.7% for FX hedgers, 2.1% for interest rate hedgers, and 3.3% for commodity price hedgers.

30 Among others, we set the sample mean for the type of hedging variable (dummy vs. continuous) because both measures come with caveats and there is no clear preference which is the better proxy.