The Effect of Performance Measure Comparability and Discriminability on Measure Discontinuance

ABSTRACT

This study investigates the effect of performance measure characteristics―comparability and discriminability―on performance measure discontinuance. Relative performance evaluation using comparable information offers a means of filtering out common shocks and capturing the subordinates’ own performance relative to that of peers. In addition, a measure’s ability to discriminate good from poor performance is important in the incentive contract. We show that since subordinates have an incentive to put effort into the comparable and discriminable performance measures, the superior is more likely to retain performance measures with greater comparability and discriminability. Using a large sample of performance evaluation data from 15 state-owned enterprises in South Korea over the period 1985–2021, we find that comparable and discriminable performance measures are less likely to be discontinued. Our study contributes to the performance evaluation literature by identifying two measure characteristics that influence the evolution of performance measures.

Keywords:

• Comparability
• Discriminability
• Performance measure discontinuance
• Performance measure choice

Correction Statement

This article has been corrected with minor changes. These changes do not impact the academic content of the article.

Acknowledgements

We thank Josep Bisbe (editor) and the two anonymous reviewers for their valuable feedback. This paper is based on the dissertation of the first author at Seoul National University. We thank Iny Hwang, Karen Sedatole, Jae Yong Shin, Dae-Hee Yoon for the insightful discussions and comments. Jeong–Hoon Hyun gratefully acknowledges financial support from the Chung-Ang University Research Grants in 2023.

Disclosure statement

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

Data Availability

Data are available from the public sources identified in the text.

Notes

1 Common measures are typically applied universally to all SOEs. However, they may also be selectively applied to a subset of SOEs, indicating their usage for either all SOEs or only for two or more out of the entire set. As a result, we can differentiate common measures based on their RPE, categorizing them as having more RPE (i.e., used for all SOEs) or less RPE (i.e., applied to a subset of SOEs). Further details are provided in Section IV.

2 Some prior studies have examined the peer selection process using US listed firms. For instance, Albuquerque (2009) adopts implicit matching approach to find RPE peers using both industry and size. However, Faulkender and Yang (2010) and Gong et al. (2011) hand–collect proxy disclosures and find that about 25% of the S&P 1500 firms explicitly use RPE; these firms select highly paid peers to justify their CEO compensation in setting of their own executive compensation. In addition, De Franco et al. (2015) investigate how analysts select peers and determine that peer companies with high valuations are selected according to analysts’ incentives and ability. Ding et al. (2019) provide a machine learning–based peer selection method using clustering techniques.

3 This study utilizes the annual performance evaluation results of 15 South Korean SOEs from 1985 to 2021. The SOEs are as follows: Korea Electric Power Corporation, Korea Minting & Security Printing Corporation, Korea Coal Corporation, Korea Resources Corporation, Korea National Oil Corporation, Korea Trade Investment Promotion Agency, Korea Highway Corporation, Korea National Housing Corporation, Korea Water Resources Corporation, Korea Land Corporation, Korea Rural Community & Agriculture Corporation, Agricultural and Fishery Marketing Corporation, Korea National Tourism Organization, Korea Railroad Corporation, and Korea Land and Housing Corporation. The Korea Land Corporation and the Korea National Housing Corporation were merged into the Korea Land and Housing Corporation in 2009.

4 Considerable attention has been paid to the performance evaluation system of Korean SOEs, thanks to its uniqueness and potential as a valuable research setting (e.g., Ahn et al., 2010; Ahn et al., 2018; Hyun et al., 2022). Following these prior studies, we define the Government and the evaluation committee as ‘superior’ and the SOEs and their CEOs as ‘subordinates.’

5 In addition, SOEs can select common measures based upon the approval by the committee. For example, the committee suggests several possible financial measures such as asset turnover ( = sales/asset), interest coverage ratio ( = operating income/interest expense), leverage ( = liability/equity), operating margin ratio ( = operating income/sales), and ROE ( = EBT/equity). Then, 15 SOEs select two or three measures and discuss with the committee about whether they are well-matched with operation of the SOE and then, the measure selection is finalized before the beginning of the evaluation period. This approach reflects the uniqueness of SOE’s business and increases the SOE’s acceptance and procedural fairness.

6 The RPE literature has long discussed the importance of peer quality to motivate employees (Bizjak et al., 2008; Albuquerque et al., 2013), in addition to bonus determination. There might be concerns that RPE is less applicable to the SOE setting because there is heterogeneity in the business operations of the 15 SOEs. The different business models of each SOE can affect common and financial measures but SOEs commonly pursue the public social welfare. The SOE evaluation system adopts both public (i.e., nonfinancial) and financial measures in a balanced way. The common and nonfinancial measures are more likely to be related to RPE because the measures are commonly applied, while common and financial measures capture the differences in financial status and profit generating activities of SOEs. We try to reflect the differences between financial and nonfinancial measures in common measure category in terms of the degree of RPE application in Table 5.

7 In the untabulated result, the average (maximum) number of performance measures per SOE and year is 31 (46), comprising both common and unique measures.

8 For instance, A+, A0, B+, … , and E0 ratings are 100%, 90%, 80%, … , and 20% point of Attain ratio, respectively.

9 Ahn et al. (2010) define discriminability as the average over the previous three years of the cross-sectional standard deviation of the score of a performance measure. As the appropriateness of performance measures is reviewed every year, their non-trivial numbers are dropped within three years. If Ahn et al.’s (2010) definition is applied to our empirical model, the size of the sample is significantly reduced. Therefore, we redefine discriminability as the cross-sectional standard deviation of the score on a performance measure for the previous year.

10 Our untabulated results show that an objective performance measure has a higher standard deviation (is more discriminable) than the subjective one at a 1% significance level. Moreover, when we limit the sample to common measures, we find that the cross-sectional standard deviation of the performance score rate of objective performance measure is also much higher than that of subjective one at a 1% significance level. These results are consistent with our construct that the proxy OBJECTIVE represents the discriminability of the performance measures.

11 To reflect this feature, we replace the year fixed-effect with the presidential cycle dummy variable. The results are consistent with our main empirical results (untabulated).

12 Common measures whose discriminability increases from the first-quartile value (0.058) to the third-quartile value (0.135) are 2.055 percent less likely to be dropped.

13 Untabulated results confirm that, when we additionally consider macroeconomic factors, such as GDP amounts and GDP growth, the results are consistent with our main findings.

14 When the performance measures are discontinued, WEIGHTDEC has a value of 1, and ΔWEIGHT is minus one times the previous WEIGHT value, given that the discontinued measure can be translated into the weight adjusted to zero.

15 Since the dependent variables DROP_t+2 and DROP_t+3 require future two- and three-year periods, respectively, the number of observations varies according to the dependent variables.